The following changes since commit aa3a285b5bc56a4208b3b57d4a55291e9c260107: Merge tag 'mem-2024-12-21' of https://github.com/davidhildenbrand/qemu into staging (2024-12-22 14:33:27 -0500) are available in the Git repository at: https://gitlab.com/rth7680/qemu.git tags/pull-tcg-20241224 for you to fetch changes up to e4a8e093dc74be049f4829831dce76e5edab0003: accel/tcg: Move gen_intermediate_code to TCGCPUOps.translate_core (2024-12-24 08:32:15 -0800) ---------------------------------------------------------------- tcg/optimize: Remove in-flight mask data from OptContext fpu: Add float*_muladd_scalbn fpu: Remove float_muladd_halve_result fpu: Add float_round_nearest_even_max fpu: Add float_muladd_suppress_add_product_zero target/hexagon: Use float32_muladd accel/tcg: Move gen_intermediate_code to TCGCPUOps.translate_core ---------------------------------------------------------------- Ilya Leoshkevich (1): tests/tcg: Do not use inttypes.h in multiarch/system/memory.c Pierrick Bouvier (1): plugins: optimize cpu_index code generation Richard Henderson (70): tcg/optimize: Split out finish_bb, finish_ebb tcg/optimize: Split out fold_affected_mask tcg/optimize: Copy mask writeback to fold_masks tcg/optimize: Split out fold_masks_zs tcg/optimize: Augment s_mask from z_mask in fold_masks_zs tcg/optimize: Change representation of s_mask tcg/optimize: Use finish_folding in fold_add, fold_add_vec, fold_addsub2 tcg/optimize: Introduce const value accessors for TempOptInfo tcg/optimize: Use fold_masks_zs in fold_and tcg/optimize: Use fold_masks_zs in fold_andc tcg/optimize: Use fold_masks_zs in fold_bswap tcg/optimize: Use fold_masks_zs in fold_count_zeros tcg/optimize: Use fold_masks_z in fold_ctpop tcg/optimize: Use fold_and and fold_masks_z in fold_deposit tcg/optimize: Compute sign mask in fold_deposit tcg/optimize: Use finish_folding in fold_divide tcg/optimize: Use finish_folding in fold_dup, fold_dup2 tcg/optimize: Use fold_masks_s in fold_eqv tcg/optimize: Use fold_masks_z in fold_extract tcg/optimize: Use finish_folding in fold_extract2 tcg/optimize: Use fold_masks_zs in fold_exts tcg/optimize: Use fold_masks_z in fold_extu tcg/optimize: Use fold_masks_zs in fold_movcond tcg/optimize: Use finish_folding in fold_mul* tcg/optimize: Use fold_masks_s in fold_nand tcg/optimize: Use fold_masks_z in fold_neg_no_const tcg/optimize: Use fold_masks_s in fold_nor tcg/optimize: Use fold_masks_s in fold_not tcg/optimize: Use fold_masks_zs in fold_or tcg/optimize: Use fold_masks_zs in fold_orc tcg/optimize: Use fold_masks_zs in fold_qemu_ld tcg/optimize: Return true from fold_qemu_st, fold_tcg_st tcg/optimize: Use finish_folding in fold_remainder tcg/optimize: Distinguish simplification in fold_setcond_zmask tcg/optimize: Use fold_masks_z in fold_setcond tcg/optimize: Use fold_masks_s in fold_negsetcond tcg/optimize: Use fold_masks_z in fold_setcond2 tcg/optimize: Use finish_folding in fold_cmp_vec tcg/optimize: Use finish_folding in fold_cmpsel_vec tcg/optimize: Use fold_masks_zs in fold_sextract tcg/optimize: Use fold_masks_zs, fold_masks_s in fold_shift tcg/optimize: Simplify sign bit test in fold_shift tcg/optimize: Use finish_folding in fold_sub, fold_sub_vec tcg/optimize: Use fold_masks_zs in fold_tcg_ld tcg/optimize: Use finish_folding in fold_tcg_ld_memcopy tcg/optimize: Use fold_masks_zs in fold_xor tcg/optimize: Use finish_folding in fold_bitsel_vec tcg/optimize: Use finish_folding as default in tcg_optimize tcg/optimize: Remove z_mask, s_mask from OptContext tcg/optimize: Re-enable sign-mask optimizations tcg/optimize: Move fold_bitsel_vec into alphabetic sort tcg/optimize: Move fold_cmp_vec, fold_cmpsel_vec into alphabetic sort softfloat: Add float{16,32,64}_muladd_scalbn target/arm: Use float*_muladd_scalbn target/sparc: Use float*_muladd_scalbn softfloat: Remove float_muladd_halve_result softfloat: Add float_round_nearest_even_max softfloat: Add float_muladd_suppress_add_product_zero target/hexagon: Use float32_mul in helper_sfmpy target/hexagon: Use float32_muladd for helper_sffma target/hexagon: Use float32_muladd for helper_sffms target/hexagon: Use float32_muladd_scalbn for helper_sffma_sc target/hexagon: Use float32_muladd for helper_sffm[as]_lib target/hexagon: Remove internal_fmafx target/hexagon: Expand GEN_XF_ROUND target/hexagon: Remove Float target/hexagon: Remove Double target/hexagon: Use mulu64 for int128_mul_6464 target/hexagon: Simplify internal_mpyhh setup accel/tcg: Move gen_intermediate_code to TCGCPUOps.translate_core include/exec/translator.h | 14 - include/fpu/softfloat-types.h | 2 + include/fpu/softfloat.h | 14 +- include/hw/core/tcg-cpu-ops.h | 13 + target/alpha/cpu.h | 2 + target/arm/internals.h | 2 + target/avr/cpu.h | 2 + target/hexagon/cpu.h | 2 + target/hexagon/fma_emu.h | 3 - target/hppa/cpu.h | 2 + target/i386/tcg/helper-tcg.h | 2 + target/loongarch/internals.h | 2 + target/m68k/cpu.h | 2 + target/microblaze/cpu.h | 2 + target/mips/tcg/tcg-internal.h | 2 + target/openrisc/cpu.h | 2 + target/ppc/cpu.h | 2 + target/riscv/cpu.h | 3 + target/rx/cpu.h | 2 + target/s390x/s390x-internal.h | 2 + target/sh4/cpu.h | 2 + target/sparc/cpu.h | 2 + target/sparc/helper.h | 4 +- target/tricore/cpu.h | 2 + target/xtensa/cpu.h | 2 + accel/tcg/cpu-exec.c | 8 +- accel/tcg/plugin-gen.c | 9 + accel/tcg/translate-all.c | 8 +- fpu/softfloat.c | 63 +-- target/alpha/cpu.c | 1 + target/alpha/translate.c | 4 +- target/arm/cpu.c | 1 + target/arm/tcg/cpu-v7m.c | 1 + target/arm/tcg/helper-a64.c | 6 +- target/arm/tcg/translate.c | 5 +- target/avr/cpu.c | 1 + target/avr/translate.c | 6 +- target/hexagon/cpu.c | 1 + target/hexagon/fma_emu.c | 496 ++++++--------------- target/hexagon/op_helper.c | 125 ++---- target/hexagon/translate.c | 4 +- target/hppa/cpu.c | 1 + target/hppa/translate.c | 4 +- target/i386/tcg/tcg-cpu.c | 1 + target/i386/tcg/translate.c | 5 +- target/loongarch/cpu.c | 1 + target/loongarch/tcg/translate.c | 4 +- target/m68k/cpu.c | 1 + target/m68k/translate.c | 4 +- target/microblaze/cpu.c | 1 + target/microblaze/translate.c | 4 +- target/mips/cpu.c | 1 + target/mips/tcg/translate.c | 4 +- target/openrisc/cpu.c | 1 + target/openrisc/translate.c | 4 +- target/ppc/cpu_init.c | 1 + target/ppc/translate.c | 4 +- target/riscv/tcg/tcg-cpu.c | 1 + target/riscv/translate.c | 4 +- target/rx/cpu.c | 1 + target/rx/translate.c | 4 +- target/s390x/cpu.c | 1 + target/s390x/tcg/translate.c | 4 +- target/sh4/cpu.c | 1 + target/sh4/translate.c | 4 +- target/sparc/cpu.c | 1 + target/sparc/fop_helper.c | 8 +- target/sparc/translate.c | 84 ++-- target/tricore/cpu.c | 1 + target/tricore/translate.c | 5 +- target/xtensa/cpu.c | 1 + target/xtensa/translate.c | 4 +- tcg/optimize.c | 857 +++++++++++++++++++----------------- tests/tcg/multiarch/system/memory.c | 9 +- fpu/softfloat-parts.c.inc | 16 +- 75 files changed, 866 insertions(+), 1009 deletions(-) From: Ilya Leoshkevich <iii@linux.ibm.com> make check-tcg fails on Fedora with the following error message: alpha-linux-gnu-gcc [...] qemu/tests/tcg/multiarch/system/memory.c -o memory [...] qemu/tests/tcg/multiarch/system/memory.c:17:10: fatal error: inttypes.h: No such file or directory 17 | #include <inttypes.h> | ^~~~~~~~~~~~ compilation terminated. The reason is that Fedora has cross-compilers, but no cross-glibc headers. Fix by hardcoding the format specifiers and dropping the include. An alternative fix would be to introduce a configure check for inttypes.h. But this would make it impossible to use Fedora cross-compilers for softmmu tests, which used to work so far. Fixes: ecbcc9ead2f8 ("tests/tcg: add a system test to check memory instrumentation") Signed-off-by: Ilya Leoshkevich <iii@linux.ibm.com> Reviewed-by: Paolo Bonzini <pbonzini@redhat.com> Message-ID: <20241010085906.226249-1-iii@linux.ibm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tests/tcg/multiarch/system/memory.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/tests/tcg/multiarch/system/memory.c b/tests/tcg/multiarch/system/memory.c index XXXXXXX..XXXXXXX 100644 --- a/tests/tcg/multiarch/system/memory.c +++ b/tests/tcg/multiarch/system/memory.c @@ -XXX,XX +XXX,XX @@ #include <stdint.h> #include <stdbool.h> -#include <inttypes.h> #include <minilib.h> #ifndef CHECK_UNALIGNED @@ -XXX,XX +XXX,XX @@ int main(void) int i; bool ok = true; - ml_printf("Test data start: 0x%"PRIxPTR"\n", &test_data[0]); - ml_printf("Test data end: 0x%"PRIxPTR"\n", &test_data[TEST_SIZE]); + ml_printf("Test data start: 0x%lx\n", (unsigned long)&test_data[0]); + ml_printf("Test data end: 0x%lx\n", (unsigned long)&test_data[TEST_SIZE]); /* Run through the unsigned tests first */ for (i = 0; i < ARRAY_SIZE(init_ufns) && ok; i++) { @@ -XXX,XX +XXX,XX @@ int main(void) ok = do_signed_reads(true); } - ml_printf("Test data read: %"PRId32"\n", test_read_count); - ml_printf("Test data write: %"PRId32"\n", test_write_count); + ml_printf("Test data read: %lu\n", (unsigned long)test_read_count); + ml_printf("Test data write: %lu\n", (unsigned long)test_write_count); ml_printf("Test complete: %s\n", ok ? "PASSED" : "FAILED"); return ok ? 0 : -1; } -- 2.43.0 From: Pierrick Bouvier <pierrick.bouvier@linaro.org> When running with a single vcpu, we can return a constant instead of a load when accessing cpu_index. A side effect is that all tcg operations using it are optimized, most notably scoreboard access. When running a simple loop in user-mode, the speedup is around 20%. Signed-off-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Reviewed-by: Richard Henderson <richard.henderson@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Message-ID: <20241128213843.1023080-1-pierrick.bouvier@linaro.org> --- accel/tcg/plugin-gen.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/accel/tcg/plugin-gen.c b/accel/tcg/plugin-gen.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/plugin-gen.c +++ b/accel/tcg/plugin-gen.c @@ -XXX,XX +XXX,XX @@ static void gen_disable_mem_helper(void) static TCGv_i32 gen_cpu_index(void) { + /* + * Optimize when we run with a single vcpu. All values using cpu_index, + * including scoreboard index, will be optimized out. + * User-mode calls tb_flush when setting this flag. In system-mode, all + * vcpus are created before generating code. + */ + if (!tcg_cflags_has(current_cpu, CF_PARALLEL)) { + return tcg_constant_i32(current_cpu->cpu_index); + } TCGv_i32 cpu_index = tcg_temp_ebb_new_i32(); tcg_gen_ld_i32(cpu_index, tcg_env, -offsetof(ArchCPU, env) + offsetof(CPUState, cpu_index)); -- 2.43.0 Call them directly from the opcode switch statement in tcg_optimize, rather than in finish_folding based on opcode flags. Adjust folding of conditional branches to match. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 47 +++++++++++++++++++++++++++++++---------------- 1 file changed, 31 insertions(+), 16 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static void copy_propagate(OptContext *ctx, TCGOp *op, } } +static void finish_bb(OptContext *ctx) +{ + /* We only optimize memory barriers across basic blocks. */ + ctx->prev_mb = NULL; +} + +static void finish_ebb(OptContext *ctx) +{ + finish_bb(ctx); + /* We only optimize across extended basic blocks. */ + memset(&ctx->temps_used, 0, sizeof(ctx->temps_used)); + remove_mem_copy_all(ctx); +} + static void finish_folding(OptContext *ctx, TCGOp *op) { const TCGOpDef *def = &tcg_op_defs[op->opc]; int i, nb_oargs; - /* - * We only optimize extended basic blocks. If the opcode ends a BB - * and is not a conditional branch, reset all temp data. - */ - if (def->flags & TCG_OPF_BB_END) { - ctx->prev_mb = NULL; - if (!(def->flags & TCG_OPF_COND_BRANCH)) { - memset(&ctx->temps_used, 0, sizeof(ctx->temps_used)); - remove_mem_copy_all(ctx); - } - return; - } - nb_oargs = def->nb_oargs; for (i = 0; i < nb_oargs; i++) { TCGTemp *ts = arg_temp(op->args[i]); @@ -XXX,XX +XXX,XX @@ static bool fold_brcond(OptContext *ctx, TCGOp *op) if (i > 0) { op->opc = INDEX_op_br; op->args[0] = op->args[3]; + finish_ebb(ctx); + } else { + finish_bb(ctx); } - return false; + return true; } static bool fold_brcond2(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ static bool fold_brcond2(OptContext *ctx, TCGOp *op) } op->opc = INDEX_op_br; op->args[0] = label; - break; + finish_ebb(ctx); + return true; } - return false; + + finish_bb(ctx); + return true; } static bool fold_bswap(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ void tcg_optimize(TCGContext *s) CASE_OP_32_64_VEC(xor): done = fold_xor(&ctx, op); break; + case INDEX_op_set_label: + case INDEX_op_br: + case INDEX_op_exit_tb: + case INDEX_op_goto_tb: + case INDEX_op_goto_ptr: + finish_ebb(&ctx); + done = true; + break; default: break; } -- 2.43.0 There are only a few logical operations which can compute an "affected" mask. Split out handling of this optimization to a separate function, only to be called when applicable. Remove the a_mask field from OptContext, as the mask is no longer stored anywhere. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 42 +++++++++++++++++++++++++++--------------- 1 file changed, 27 insertions(+), 15 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ typedef struct OptContext { QSIMPLEQ_HEAD(, MemCopyInfo) mem_free; /* In flight values from optimization. */ - uint64_t a_mask; /* mask bit is 0 iff value identical to first input */ uint64_t z_mask; /* mask bit is 0 iff value bit is 0 */ uint64_t s_mask; /* mask of clrsb(value) bits */ TCGType type; @@ -XXX,XX +XXX,XX @@ static bool fold_const2_commutative(OptContext *ctx, TCGOp *op) static bool fold_masks(OptContext *ctx, TCGOp *op) { - uint64_t a_mask = ctx->a_mask; uint64_t z_mask = ctx->z_mask; uint64_t s_mask = ctx->s_mask; @@ -XXX,XX +XXX,XX @@ static bool fold_masks(OptContext *ctx, TCGOp *op) * type changing opcodes. */ if (ctx->type == TCG_TYPE_I32) { - a_mask = (int32_t)a_mask; z_mask = (int32_t)z_mask; s_mask |= MAKE_64BIT_MASK(32, 32); ctx->z_mask = z_mask; @@ -XXX,XX +XXX,XX @@ static bool fold_masks(OptContext *ctx, TCGOp *op) if (z_mask == 0) { return tcg_opt_gen_movi(ctx, op, op->args[0], 0); } + return false; +} + +/* + * An "affected" mask bit is 0 if and only if the result is identical + * to the first input. Thus if the entire mask is 0, the operation + * is equivalent to a copy. + */ +static bool fold_affected_mask(OptContext *ctx, TCGOp *op, uint64_t a_mask) +{ + if (ctx->type == TCG_TYPE_I32) { + a_mask = (uint32_t)a_mask; + } if (a_mask == 0) { return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); } @@ -XXX,XX +XXX,XX @@ static bool fold_and(OptContext *ctx, TCGOp *op) * Known-zeros does not imply known-ones. Therefore unless * arg2 is constant, we can't infer affected bits from it. */ - if (arg_is_const(op->args[2])) { - ctx->a_mask = z1 & ~z2; + if (arg_is_const(op->args[2]) && + fold_affected_mask(ctx, op, z1 & ~z2)) { + return true; } return fold_masks(ctx, op); @@ -XXX,XX +XXX,XX @@ static bool fold_andc(OptContext *ctx, TCGOp *op) */ if (arg_is_const(op->args[2])) { uint64_t z2 = ~arg_info(op->args[2])->z_mask; - ctx->a_mask = z1 & ~z2; + if (fold_affected_mask(ctx, op, z1 & ~z2)) { + return true; + } z1 &= z2; } ctx->z_mask = z1; @@ -XXX,XX +XXX,XX @@ static bool fold_extract(OptContext *ctx, TCGOp *op) z_mask_old = arg_info(op->args[1])->z_mask; z_mask = extract64(z_mask_old, pos, len); - if (pos == 0) { - ctx->a_mask = z_mask_old ^ z_mask; + if (pos == 0 && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) { + return true; } ctx->z_mask = z_mask; ctx->s_mask = smask_from_zmask(z_mask); @@ -XXX,XX +XXX,XX @@ static bool fold_exts(OptContext *ctx, TCGOp *op) ctx->z_mask = z_mask; ctx->s_mask = s_mask; - if (!type_change) { - ctx->a_mask = s_mask & ~s_mask_old; + if (!type_change && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { + return true; } return fold_masks(ctx, op); @@ -XXX,XX +XXX,XX @@ static bool fold_extu(OptContext *ctx, TCGOp *op) ctx->z_mask = z_mask; ctx->s_mask = smask_from_zmask(z_mask); - if (!type_change) { - ctx->a_mask = z_mask_old ^ z_mask; + if (!type_change && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) { + return true; } return fold_masks(ctx, op); } @@ -XXX,XX +XXX,XX @@ static bool fold_sextract(OptContext *ctx, TCGOp *op) s_mask |= MAKE_64BIT_MASK(len, 64 - len); ctx->s_mask = s_mask; - if (pos == 0) { - ctx->a_mask = s_mask & ~s_mask_old; + if (pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { + return true; } return fold_masks(ctx, op); @@ -XXX,XX +XXX,XX @@ void tcg_optimize(TCGContext *s) } /* Assume all bits affected, no bits known zero, no sign reps. */ - ctx.a_mask = -1; ctx.z_mask = -1; ctx.s_mask = 0; -- 2.43.0 Use of fold_masks should be restricted to those opcodes that can reliably make use of it -- those with a single output, and from higher-level folders that set up the masks. Prepare for conversion of each folder in turn. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 17 ++++++++++++++--- 1 file changed, 14 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_masks(OptContext *ctx, TCGOp *op) { uint64_t z_mask = ctx->z_mask; uint64_t s_mask = ctx->s_mask; + const TCGOpDef *def = &tcg_op_defs[op->opc]; + TCGTemp *ts; + TempOptInfo *ti; + + /* Only single-output opcodes are supported here. */ + tcg_debug_assert(def->nb_oargs == 1); /* * 32-bit ops generate 32-bit results, which for the purpose of @@ -XXX,XX +XXX,XX @@ static bool fold_masks(OptContext *ctx, TCGOp *op) if (ctx->type == TCG_TYPE_I32) { z_mask = (int32_t)z_mask; s_mask |= MAKE_64BIT_MASK(32, 32); - ctx->z_mask = z_mask; - ctx->s_mask = s_mask; } if (z_mask == 0) { return tcg_opt_gen_movi(ctx, op, op->args[0], 0); } - return false; + + ts = arg_temp(op->args[0]); + reset_ts(ctx, ts); + + ti = ts_info(ts); + ti->z_mask = z_mask; + ti->s_mask = s_mask; + return true; } /* -- 2.43.0 Add a routine to which masks can be passed directly, rather than storing them into OptContext. To be used in upcoming patches. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_const2_commutative(OptContext *ctx, TCGOp *op) return fold_const2(ctx, op); } -static bool fold_masks(OptContext *ctx, TCGOp *op) +/* + * Record "zero" and "sign" masks for the single output of @op. + * See TempOptInfo definition of z_mask and s_mask. + * If z_mask allows, fold the output to constant zero. + */ +static bool fold_masks_zs(OptContext *ctx, TCGOp *op, + uint64_t z_mask, uint64_t s_mask) { - uint64_t z_mask = ctx->z_mask; - uint64_t s_mask = ctx->s_mask; const TCGOpDef *def = &tcg_op_defs[op->opc]; TCGTemp *ts; TempOptInfo *ti; @@ -XXX,XX +XXX,XX @@ static bool fold_masks(OptContext *ctx, TCGOp *op) return true; } +static bool fold_masks(OptContext *ctx, TCGOp *op) +{ + return fold_masks_zs(ctx, op, ctx->z_mask, ctx->s_mask); +} + /* * An "affected" mask bit is 0 if and only if the result is identical * to the first input. Thus if the entire mask is 0, the operation -- 2.43.0 Consider the passed s_mask to be a minimum deduced from either existing s_mask or from a sign-extension operation. We may be able to deduce more from the set of known zeros. Remove identical logic from several opcode folders. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 21 ++++++--------------- 1 file changed, 6 insertions(+), 15 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_const2_commutative(OptContext *ctx, TCGOp *op) * Record "zero" and "sign" masks for the single output of @op. * See TempOptInfo definition of z_mask and s_mask. * If z_mask allows, fold the output to constant zero. + * The passed s_mask may be augmented by z_mask. */ static bool fold_masks_zs(OptContext *ctx, TCGOp *op, uint64_t z_mask, uint64_t s_mask) @@ -XXX,XX +XXX,XX @@ static bool fold_masks_zs(OptContext *ctx, TCGOp *op, ti = ts_info(ts); ti->z_mask = z_mask; - ti->s_mask = s_mask; + ti->s_mask = s_mask | smask_from_zmask(z_mask); return true; } @@ -XXX,XX +XXX,XX @@ static bool fold_bswap(OptContext *ctx, TCGOp *op) default: g_assert_not_reached(); } - s_mask = smask_from_zmask(z_mask); + s_mask = 0; switch (op->args[2] & (TCG_BSWAP_OZ | TCG_BSWAP_OS)) { case TCG_BSWAP_OZ: break; @@ -XXX,XX +XXX,XX @@ static bool fold_bswap(OptContext *ctx, TCGOp *op) default: /* The high bits are undefined: force all bits above the sign to 1. */ z_mask |= sign << 1; - s_mask = 0; break; } ctx->z_mask = z_mask; @@ -XXX,XX +XXX,XX @@ static bool fold_count_zeros(OptContext *ctx, TCGOp *op) g_assert_not_reached(); } ctx->z_mask = arg_info(op->args[2])->z_mask | z_mask; - ctx->s_mask = smask_from_zmask(ctx->z_mask); return false; } @@ -XXX,XX +XXX,XX @@ static bool fold_ctpop(OptContext *ctx, TCGOp *op) default: g_assert_not_reached(); } - ctx->s_mask = smask_from_zmask(ctx->z_mask); return false; } @@ -XXX,XX +XXX,XX @@ static bool fold_extract(OptContext *ctx, TCGOp *op) return true; } ctx->z_mask = z_mask; - ctx->s_mask = smask_from_zmask(z_mask); return fold_masks(ctx, op); } @@ -XXX,XX +XXX,XX @@ static bool fold_extu(OptContext *ctx, TCGOp *op) } ctx->z_mask = z_mask; - ctx->s_mask = smask_from_zmask(z_mask); if (!type_change && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool fold_qemu_ld(OptContext *ctx, TCGOp *op) int width = 8 * memop_size(mop); if (width < 64) { - ctx->s_mask = MAKE_64BIT_MASK(width, 64 - width); - if (!(mop & MO_SIGN)) { + if (mop & MO_SIGN) { + ctx->s_mask = MAKE_64BIT_MASK(width, 64 - width); + } else { ctx->z_mask = MAKE_64BIT_MASK(0, width); - ctx->s_mask <<= 1; } } @@ -XXX,XX +XXX,XX @@ static bool fold_setcond(OptContext *ctx, TCGOp *op) fold_setcond_tst_pow2(ctx, op, false); ctx->z_mask = 1; - ctx->s_mask = smask_from_zmask(1); return false; } @@ -XXX,XX +XXX,XX @@ static bool fold_setcond2(OptContext *ctx, TCGOp *op) } ctx->z_mask = 1; - ctx->s_mask = smask_from_zmask(1); return false; do_setcond_const: @@ -XXX,XX +XXX,XX @@ static bool fold_tcg_ld(OptContext *ctx, TCGOp *op) break; CASE_OP_32_64(ld8u): ctx->z_mask = MAKE_64BIT_MASK(0, 8); - ctx->s_mask = MAKE_64BIT_MASK(9, 55); break; CASE_OP_32_64(ld16s): ctx->s_mask = MAKE_64BIT_MASK(16, 48); break; CASE_OP_32_64(ld16u): ctx->z_mask = MAKE_64BIT_MASK(0, 16); - ctx->s_mask = MAKE_64BIT_MASK(17, 47); break; case INDEX_op_ld32s_i64: ctx->s_mask = MAKE_64BIT_MASK(32, 32); break; case INDEX_op_ld32u_i64: ctx->z_mask = MAKE_64BIT_MASK(0, 32); - ctx->s_mask = MAKE_64BIT_MASK(33, 31); break; default: g_assert_not_reached(); -- 2.43.0 Change the representation from sign bit repetitions to all bits equal to the sign bit, including the sign bit itself. The previous format has a problem in that it is difficult to recreate a valid sign mask after a shift operation: the "repetitions" part of the previous format meant that applying the same shift as for the value lead to an off-by-one value. The new format, including the sign bit itself, means that the sign mask can be manipulated in exactly the same way as the value, canonicalization is easier. Canonicalize the s_mask in fold_masks_zs, rather than requiring callers to do so. Treat 0 as a non-canonical but typeless input for no sign information, which will be reset as appropriate for the data type. We can easily fold in the data from z_mask while canonicalizing. Temporarily disable optimizations using s_mask while each operation is converted to use fold_masks_zs and to the new form. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 64 ++++++++++++-------------------------------------- 1 file changed, 15 insertions(+), 49 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ typedef struct TempOptInfo { QSIMPLEQ_HEAD(, MemCopyInfo) mem_copy; uint64_t val; uint64_t z_mask; /* mask bit is 0 if and only if value bit is 0 */ - uint64_t s_mask; /* a left-aligned mask of clrsb(value) bits. */ + uint64_t s_mask; /* mask bit is 1 if value bit matches msb */ } TempOptInfo; typedef struct OptContext { @@ -XXX,XX +XXX,XX @@ typedef struct OptContext { /* In flight values from optimization. */ uint64_t z_mask; /* mask bit is 0 iff value bit is 0 */ - uint64_t s_mask; /* mask of clrsb(value) bits */ + uint64_t s_mask; /* mask bit is 1 if value bit matches msb */ TCGType type; } OptContext; -/* Calculate the smask for a specific value. */ -static uint64_t smask_from_value(uint64_t value) -{ - int rep = clrsb64(value); - return ~(~0ull >> rep); -} - -/* - * Calculate the smask for a given set of known-zeros. - * If there are lots of zeros on the left, we can consider the remainder - * an unsigned field, and thus the corresponding signed field is one bit - * larger. - */ -static uint64_t smask_from_zmask(uint64_t zmask) -{ - /* - * Only the 0 bits are significant for zmask, thus the msb itself - * must be zero, else we have no sign information. - */ - int rep = clz64(zmask); - if (rep == 0) { - return 0; - } - rep -= 1; - return ~(~0ull >> rep); -} - -/* - * Recreate a properly left-aligned smask after manipulation. - * Some bit-shuffling, particularly shifts and rotates, may - * retain sign bits on the left, but may scatter disconnected - * sign bits on the right. Retain only what remains to the left. - */ -static uint64_t smask_from_smask(int64_t smask) -{ - /* Only the 1 bits are significant for smask */ - return smask_from_zmask(~smask); -} - static inline TempOptInfo *ts_info(TCGTemp *ts) { return ts->state_ptr; @@ -XXX,XX +XXX,XX @@ static void init_ts_info(OptContext *ctx, TCGTemp *ts) ti->is_const = true; ti->val = ts->val; ti->z_mask = ts->val; - ti->s_mask = smask_from_value(ts->val); + ti->s_mask = INT64_MIN >> clrsb64(ts->val); } else { ti->is_const = false; ti->z_mask = -1; @@ -XXX,XX +XXX,XX @@ static void finish_folding(OptContext *ctx, TCGOp *op) */ if (i == 0) { ts_info(ts)->z_mask = ctx->z_mask; - ts_info(ts)->s_mask = ctx->s_mask; } } } @@ -XXX,XX +XXX,XX @@ static bool fold_const2_commutative(OptContext *ctx, TCGOp *op) * The passed s_mask may be augmented by z_mask. */ static bool fold_masks_zs(OptContext *ctx, TCGOp *op, - uint64_t z_mask, uint64_t s_mask) + uint64_t z_mask, int64_t s_mask) { const TCGOpDef *def = &tcg_op_defs[op->opc]; TCGTemp *ts; TempOptInfo *ti; + int rep; /* Only single-output opcodes are supported here. */ tcg_debug_assert(def->nb_oargs == 1); @@ -XXX,XX +XXX,XX @@ static bool fold_masks_zs(OptContext *ctx, TCGOp *op, */ if (ctx->type == TCG_TYPE_I32) { z_mask = (int32_t)z_mask; - s_mask |= MAKE_64BIT_MASK(32, 32); + s_mask |= INT32_MIN; } if (z_mask == 0) { @@ -XXX,XX +XXX,XX @@ static bool fold_masks_zs(OptContext *ctx, TCGOp *op, ti = ts_info(ts); ti->z_mask = z_mask; - ti->s_mask = s_mask | smask_from_zmask(z_mask); + + /* Canonicalize s_mask and incorporate data from z_mask. */ + rep = clz64(~s_mask); + rep = MAX(rep, clz64(z_mask)); + rep = MAX(rep - 1, 0); + ti->s_mask = INT64_MIN >> rep; + return true; } @@ -XXX,XX +XXX,XX @@ static bool fold_exts(OptContext *ctx, TCGOp *op) ctx->z_mask = z_mask; ctx->s_mask = s_mask; - if (!type_change && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { + if (0 && !type_change && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool fold_sextract(OptContext *ctx, TCGOp *op) s_mask |= MAKE_64BIT_MASK(len, 64 - len); ctx->s_mask = s_mask; - if (pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { + if (0 && pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool fold_shift(OptContext *ctx, TCGOp *op) ctx->z_mask = do_constant_folding(op->opc, ctx->type, z_mask, sh); s_mask = do_constant_folding(op->opc, ctx->type, s_mask, sh); - ctx->s_mask = smask_from_smask(s_mask); return fold_masks(ctx, op); } -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 9 +++++---- 1 file changed, 5 insertions(+), 4 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static void finish_ebb(OptContext *ctx) remove_mem_copy_all(ctx); } -static void finish_folding(OptContext *ctx, TCGOp *op) +static bool finish_folding(OptContext *ctx, TCGOp *op) { const TCGOpDef *def = &tcg_op_defs[op->opc]; int i, nb_oargs; @@ -XXX,XX +XXX,XX @@ static void finish_folding(OptContext *ctx, TCGOp *op) ts_info(ts)->z_mask = ctx->z_mask; } } + return true; } /* @@ -XXX,XX +XXX,XX @@ static bool fold_add(OptContext *ctx, TCGOp *op) fold_xi_to_x(ctx, op, 0)) { return true; } - return false; + return finish_folding(ctx, op); } /* We cannot as yet do_constant_folding with vectors. */ @@ -XXX,XX +XXX,XX @@ static bool fold_add_vec(OptContext *ctx, TCGOp *op) fold_xi_to_x(ctx, op, 0)) { return true; } - return false; + return finish_folding(ctx, op); } static bool fold_addsub2(OptContext *ctx, TCGOp *op, bool add) @@ -XXX,XX +XXX,XX @@ static bool fold_addsub2(OptContext *ctx, TCGOp *op, bool add) op->args[4] = arg_new_constant(ctx, bl); op->args[5] = arg_new_constant(ctx, bh); } - return false; + return finish_folding(ctx, op); } static bool fold_add2(OptContext *ctx, TCGOp *op) -- 2.43.0 Introduce ti_is_const, ti_const_val, ti_is_const_val. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 20 +++++++++++++++++--- 1 file changed, 17 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static inline TempOptInfo *arg_info(TCGArg arg) return ts_info(arg_temp(arg)); } +static inline bool ti_is_const(TempOptInfo *ti) +{ + return ti->is_const; +} + +static inline uint64_t ti_const_val(TempOptInfo *ti) +{ + return ti->val; +} + +static inline bool ti_is_const_val(TempOptInfo *ti, uint64_t val) +{ + return ti_is_const(ti) && ti_const_val(ti) == val; +} + static inline bool ts_is_const(TCGTemp *ts) { - return ts_info(ts)->is_const; + return ti_is_const(ts_info(ts)); } static inline bool ts_is_const_val(TCGTemp *ts, uint64_t val) { - TempOptInfo *ti = ts_info(ts); - return ti->is_const && ti->val == val; + return ti_is_const_val(ts_info(ts), val); } static inline bool arg_is_const(TCGArg arg) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Sink mask computation below fold_affected_mask early exit. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 30 ++++++++++++++++-------------- 1 file changed, 16 insertions(+), 14 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_add2(OptContext *ctx, TCGOp *op) static bool fold_and(OptContext *ctx, TCGOp *op) { - uint64_t z1, z2; + uint64_t z1, z2, z_mask, s_mask; + TempOptInfo *t1, *t2; if (fold_const2_commutative(ctx, op) || fold_xi_to_i(ctx, op, 0) || @@ -XXX,XX +XXX,XX @@ static bool fold_and(OptContext *ctx, TCGOp *op) return true; } - z1 = arg_info(op->args[1])->z_mask; - z2 = arg_info(op->args[2])->z_mask; - ctx->z_mask = z1 & z2; - - /* - * Sign repetitions are perforce all identical, whether they are 1 or 0. - * Bitwise operations preserve the relative quantity of the repetitions. - */ - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; + t1 = arg_info(op->args[1]); + t2 = arg_info(op->args[2]); + z1 = t1->z_mask; + z2 = t2->z_mask; /* * Known-zeros does not imply known-ones. Therefore unless * arg2 is constant, we can't infer affected bits from it. */ - if (arg_is_const(op->args[2]) && - fold_affected_mask(ctx, op, z1 & ~z2)) { + if (ti_is_const(t2) && fold_affected_mask(ctx, op, z1 & ~z2)) { return true; } - return fold_masks(ctx, op); + z_mask = z1 & z2; + + /* + * Sign repetitions are perforce all identical, whether they are 1 or 0. + * Bitwise operations preserve the relative quantity of the repetitions. + */ + s_mask = t1->s_mask & t2->s_mask; + + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_andc(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Avoid double inversion of the value of second const operand. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 21 +++++++++++---------- 1 file changed, 11 insertions(+), 10 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_and(OptContext *ctx, TCGOp *op) static bool fold_andc(OptContext *ctx, TCGOp *op) { - uint64_t z1; + uint64_t z_mask, s_mask; + TempOptInfo *t1, *t2; if (fold_const2(ctx, op) || fold_xx_to_i(ctx, op, 0) || @@ -XXX,XX +XXX,XX @@ static bool fold_andc(OptContext *ctx, TCGOp *op) return true; } - z1 = arg_info(op->args[1])->z_mask; + t1 = arg_info(op->args[1]); + t2 = arg_info(op->args[2]); + z_mask = t1->z_mask; /* * Known-zeros does not imply known-ones. Therefore unless * arg2 is constant, we can't infer anything from it. */ - if (arg_is_const(op->args[2])) { - uint64_t z2 = ~arg_info(op->args[2])->z_mask; - if (fold_affected_mask(ctx, op, z1 & ~z2)) { + if (ti_is_const(t2)) { + uint64_t v2 = ti_const_val(t2); + if (fold_affected_mask(ctx, op, z_mask & v2)) { return true; } - z1 &= z2; + z_mask &= ~v2; } - ctx->z_mask = z1; - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; - return fold_masks(ctx, op); + s_mask = t1->s_mask & t2->s_mask; + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_brcond(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Always set s_mask along the BSWAP_OS path, since the result is being explicitly sign-extended. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 21 ++++++++++----------- 1 file changed, 10 insertions(+), 11 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_brcond2(OptContext *ctx, TCGOp *op) static bool fold_bswap(OptContext *ctx, TCGOp *op) { uint64_t z_mask, s_mask, sign; + TempOptInfo *t1 = arg_info(op->args[1]); - if (arg_is_const(op->args[1])) { - uint64_t t = arg_info(op->args[1])->val; - - t = do_constant_folding(op->opc, ctx->type, t, op->args[2]); - return tcg_opt_gen_movi(ctx, op, op->args[0], t); + if (ti_is_const(t1)) { + return tcg_opt_gen_movi(ctx, op, op->args[0], + do_constant_folding(op->opc, ctx->type, + ti_const_val(t1), + op->args[2])); } - z_mask = arg_info(op->args[1])->z_mask; - + z_mask = t1->z_mask; switch (op->opc) { case INDEX_op_bswap16_i32: case INDEX_op_bswap16_i64: @@ -XXX,XX +XXX,XX @@ static bool fold_bswap(OptContext *ctx, TCGOp *op) /* If the sign bit may be 1, force all the bits above to 1. */ if (z_mask & sign) { z_mask |= sign; - s_mask = sign << 1; } + /* The value and therefore s_mask is explicitly sign-extended. */ + s_mask = sign; break; default: /* The high bits are undefined: force all bits above the sign to 1. */ z_mask |= sign << 1; break; } - ctx->z_mask = z_mask; - ctx->s_mask = s_mask; - return fold_masks(ctx, op); + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_call(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Compute s_mask from the union of the maximum count and the op2 fallback for op1 being zero. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 15 ++++++++++----- 1 file changed, 10 insertions(+), 5 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_call(OptContext *ctx, TCGOp *op) static bool fold_count_zeros(OptContext *ctx, TCGOp *op) { - uint64_t z_mask; + uint64_t z_mask, s_mask; + TempOptInfo *t1 = arg_info(op->args[1]); + TempOptInfo *t2 = arg_info(op->args[2]); - if (arg_is_const(op->args[1])) { - uint64_t t = arg_info(op->args[1])->val; + if (ti_is_const(t1)) { + uint64_t t = ti_const_val(t1); if (t != 0) { t = do_constant_folding(op->opc, ctx->type, t, 0); @@ -XXX,XX +XXX,XX @@ static bool fold_count_zeros(OptContext *ctx, TCGOp *op) default: g_assert_not_reached(); } - ctx->z_mask = arg_info(op->args[2])->z_mask | z_mask; - return false; + s_mask = ~z_mask; + z_mask |= t2->z_mask; + s_mask &= t2->s_mask; + + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_ctpop(OptContext *ctx, TCGOp *op) -- 2.43.0 Add fold_masks_z as a trivial wrapper around fold_masks_zs. Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_masks_zs(OptContext *ctx, TCGOp *op, return true; } +static bool fold_masks_z(OptContext *ctx, TCGOp *op, uint64_t z_mask) +{ + return fold_masks_zs(ctx, op, z_mask, 0); +} + static bool fold_masks(OptContext *ctx, TCGOp *op) { return fold_masks_zs(ctx, op, ctx->z_mask, ctx->s_mask); @@ -XXX,XX +XXX,XX @@ static bool fold_count_zeros(OptContext *ctx, TCGOp *op) static bool fold_ctpop(OptContext *ctx, TCGOp *op) { + uint64_t z_mask; + if (fold_const1(ctx, op)) { return true; } switch (ctx->type) { case TCG_TYPE_I32: - ctx->z_mask = 32 | 31; + z_mask = 32 | 31; break; case TCG_TYPE_I64: - ctx->z_mask = 64 | 63; + z_mask = 64 | 63; break; default: g_assert_not_reached(); } - return false; + return fold_masks_z(ctx, op, z_mask); } static bool fold_deposit(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. When we fold to and, use fold_and. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 35 +++++++++++++++++------------------ 1 file changed, 17 insertions(+), 18 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_ctpop(OptContext *ctx, TCGOp *op) static bool fold_deposit(OptContext *ctx, TCGOp *op) { + TempOptInfo *t1 = arg_info(op->args[1]); + TempOptInfo *t2 = arg_info(op->args[2]); + int ofs = op->args[3]; + int len = op->args[4]; TCGOpcode and_opc; + uint64_t z_mask; - if (arg_is_const(op->args[1]) && arg_is_const(op->args[2])) { - uint64_t t1 = arg_info(op->args[1])->val; - uint64_t t2 = arg_info(op->args[2])->val; - - t1 = deposit64(t1, op->args[3], op->args[4], t2); - return tcg_opt_gen_movi(ctx, op, op->args[0], t1); + if (ti_is_const(t1) && ti_is_const(t2)) { + return tcg_opt_gen_movi(ctx, op, op->args[0], + deposit64(ti_const_val(t1), ofs, len, + ti_const_val(t2))); } switch (ctx->type) { @@ -XXX,XX +XXX,XX @@ static bool fold_deposit(OptContext *ctx, TCGOp *op) } /* Inserting a value into zero at offset 0. */ - if (arg_is_const_val(op->args[1], 0) && op->args[3] == 0) { - uint64_t mask = MAKE_64BIT_MASK(0, op->args[4]); + if (ti_is_const_val(t1, 0) && ofs == 0) { + uint64_t mask = MAKE_64BIT_MASK(0, len); op->opc = and_opc; op->args[1] = op->args[2]; op->args[2] = arg_new_constant(ctx, mask); - ctx->z_mask = mask & arg_info(op->args[1])->z_mask; - return false; + return fold_and(ctx, op); } /* Inserting zero into a value. */ - if (arg_is_const_val(op->args[2], 0)) { - uint64_t mask = deposit64(-1, op->args[3], op->args[4], 0); + if (ti_is_const_val(t2, 0)) { + uint64_t mask = deposit64(-1, ofs, len, 0); op->opc = and_opc; op->args[2] = arg_new_constant(ctx, mask); - ctx->z_mask = mask & arg_info(op->args[1])->z_mask; - return false; + return fold_and(ctx, op); } - ctx->z_mask = deposit64(arg_info(op->args[1])->z_mask, - op->args[3], op->args[4], - arg_info(op->args[2])->z_mask); - return false; + z_mask = deposit64(t1->z_mask, ofs, len, t2->z_mask); + return fold_masks_z(ctx, op, z_mask); } static bool fold_divide(OptContext *ctx, TCGOp *op) -- 2.43.0 The input which overlaps the sign bit of the output can have its input s_mask propagated to the output s_mask. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_deposit(OptContext *ctx, TCGOp *op) TempOptInfo *t2 = arg_info(op->args[2]); int ofs = op->args[3]; int len = op->args[4]; + int width; TCGOpcode and_opc; - uint64_t z_mask; + uint64_t z_mask, s_mask; if (ti_is_const(t1) && ti_is_const(t2)) { return tcg_opt_gen_movi(ctx, op, op->args[0], @@ -XXX,XX +XXX,XX @@ static bool fold_deposit(OptContext *ctx, TCGOp *op) switch (ctx->type) { case TCG_TYPE_I32: and_opc = INDEX_op_and_i32; + width = 32; break; case TCG_TYPE_I64: and_opc = INDEX_op_and_i64; + width = 64; break; default: g_assert_not_reached(); @@ -XXX,XX +XXX,XX @@ static bool fold_deposit(OptContext *ctx, TCGOp *op) return fold_and(ctx, op); } + /* The s_mask from the top portion of the deposit is still valid. */ + if (ofs + len == width) { + s_mask = t2->s_mask << ofs; + } else { + s_mask = t1->s_mask & ~MAKE_64BIT_MASK(0, ofs + len); + } + z_mask = deposit64(t1->z_mask, ofs, len, t2->z_mask); - return fold_masks_z(ctx, op, z_mask); + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_divide(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_divide(OptContext *ctx, TCGOp *op) fold_xi_to_x(ctx, op, 1)) { return true; } - return false; + return finish_folding(ctx, op); } static bool fold_dup(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_dup(OptContext *ctx, TCGOp *op) t = dup_const(TCGOP_VECE(op), t); return tcg_opt_gen_movi(ctx, op, op->args[0], t); } - return false; + return finish_folding(ctx, op); } static bool fold_dup2(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ static bool fold_dup2(OptContext *ctx, TCGOp *op) op->opc = INDEX_op_dup_vec; TCGOP_VECE(op) = MO_32; } - return false; + return finish_folding(ctx, op); } static bool fold_eqv(OptContext *ctx, TCGOp *op) -- 2.43.0 Add fold_masks_s as a trivial wrapper around fold_masks_zs. Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 13 ++++++++++--- 1 file changed, 10 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_masks_z(OptContext *ctx, TCGOp *op, uint64_t z_mask) return fold_masks_zs(ctx, op, z_mask, 0); } +static bool fold_masks_s(OptContext *ctx, TCGOp *op, uint64_t s_mask) +{ + return fold_masks_zs(ctx, op, -1, s_mask); +} + static bool fold_masks(OptContext *ctx, TCGOp *op) { return fold_masks_zs(ctx, op, ctx->z_mask, ctx->s_mask); @@ -XXX,XX +XXX,XX @@ static bool fold_dup2(OptContext *ctx, TCGOp *op) static bool fold_eqv(OptContext *ctx, TCGOp *op) { + uint64_t s_mask; + if (fold_const2_commutative(ctx, op) || fold_xi_to_x(ctx, op, -1) || fold_xi_to_not(ctx, op, 0)) { return true; } - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; - return false; + s_mask = arg_info(op->args[1])->s_mask + & arg_info(op->args[2])->s_mask; + return fold_masks_s(ctx, op, s_mask); } static bool fold_extract(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 15 ++++++--------- 1 file changed, 6 insertions(+), 9 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_eqv(OptContext *ctx, TCGOp *op) static bool fold_extract(OptContext *ctx, TCGOp *op) { uint64_t z_mask_old, z_mask; + TempOptInfo *t1 = arg_info(op->args[1]); int pos = op->args[2]; int len = op->args[3]; - if (arg_is_const(op->args[1])) { - uint64_t t; - - t = arg_info(op->args[1])->val; - t = extract64(t, pos, len); - return tcg_opt_gen_movi(ctx, op, op->args[0], t); + if (ti_is_const(t1)) { + return tcg_opt_gen_movi(ctx, op, op->args[0], + extract64(ti_const_val(t1), pos, len)); } - z_mask_old = arg_info(op->args[1])->z_mask; + z_mask_old = t1->z_mask; z_mask = extract64(z_mask_old, pos, len); if (pos == 0 && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) { return true; } - ctx->z_mask = z_mask; - return fold_masks(ctx, op); + return fold_masks_z(ctx, op, z_mask); } static bool fold_extract2(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_extract2(OptContext *ctx, TCGOp *op) } return tcg_opt_gen_movi(ctx, op, op->args[0], v1 | v2); } - return false; + return finish_folding(ctx, op); } static bool fold_exts(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Explicitly sign-extend z_mask instead of doing that manually. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 29 ++++++++++++----------------- 1 file changed, 12 insertions(+), 17 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_extract2(OptContext *ctx, TCGOp *op) static bool fold_exts(OptContext *ctx, TCGOp *op) { - uint64_t s_mask_old, s_mask, z_mask, sign; + uint64_t s_mask_old, s_mask, z_mask; bool type_change = false; + TempOptInfo *t1; if (fold_const1(ctx, op)) { return true; } - z_mask = arg_info(op->args[1])->z_mask; - s_mask = arg_info(op->args[1])->s_mask; + t1 = arg_info(op->args[1]); + z_mask = t1->z_mask; + s_mask = t1->s_mask; s_mask_old = s_mask; switch (op->opc) { CASE_OP_32_64(ext8s): - sign = INT8_MIN; - z_mask = (uint8_t)z_mask; + s_mask |= INT8_MIN; + z_mask = (int8_t)z_mask; break; CASE_OP_32_64(ext16s): - sign = INT16_MIN; - z_mask = (uint16_t)z_mask; + s_mask |= INT16_MIN; + z_mask = (int16_t)z_mask; break; case INDEX_op_ext_i32_i64: type_change = true; QEMU_FALLTHROUGH; case INDEX_op_ext32s_i64: - sign = INT32_MIN; - z_mask = (uint32_t)z_mask; + s_mask |= INT32_MIN; + z_mask = (int32_t)z_mask; break; default: g_assert_not_reached(); } - if (z_mask & sign) { - z_mask |= sign; - } - s_mask |= sign << 1; - - ctx->z_mask = z_mask; - ctx->s_mask = s_mask; if (0 && !type_change && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { return true; } - return fold_masks(ctx, op); + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_extu(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_extu(OptContext *ctx, TCGOp *op) g_assert_not_reached(); } - ctx->z_mask = z_mask; if (!type_change && fold_affected_mask(ctx, op, z_mask_old ^ z_mask)) { return true; } - return fold_masks(ctx, op); + + return fold_masks_z(ctx, op, z_mask); } static bool fold_mb(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 19 +++++++++++-------- 1 file changed, 11 insertions(+), 8 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_mov(OptContext *ctx, TCGOp *op) static bool fold_movcond(OptContext *ctx, TCGOp *op) { + uint64_t z_mask, s_mask; + TempOptInfo *tt, *ft; int i; /* If true and false values are the same, eliminate the cmp. */ @@ -XXX,XX +XXX,XX @@ static bool fold_movcond(OptContext *ctx, TCGOp *op) return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[4 - i]); } - ctx->z_mask = arg_info(op->args[3])->z_mask - | arg_info(op->args[4])->z_mask; - ctx->s_mask = arg_info(op->args[3])->s_mask - & arg_info(op->args[4])->s_mask; + tt = arg_info(op->args[3]); + ft = arg_info(op->args[4]); + z_mask = tt->z_mask | ft->z_mask; + s_mask = tt->s_mask & ft->s_mask; - if (arg_is_const(op->args[3]) && arg_is_const(op->args[4])) { - uint64_t tv = arg_info(op->args[3])->val; - uint64_t fv = arg_info(op->args[4])->val; + if (ti_is_const(tt) && ti_is_const(ft)) { + uint64_t tv = ti_const_val(tt); + uint64_t fv = ti_const_val(ft); TCGOpcode opc, negopc = 0; TCGCond cond = op->args[5]; @@ -XXX,XX +XXX,XX @@ static bool fold_movcond(OptContext *ctx, TCGOp *op) } } } - return false; + + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_mul(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_mul(OptContext *ctx, TCGOp *op) fold_xi_to_x(ctx, op, 1)) { return true; } - return false; + return finish_folding(ctx, op); } static bool fold_mul_highpart(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ static bool fold_mul_highpart(OptContext *ctx, TCGOp *op) fold_xi_to_i(ctx, op, 0)) { return true; } - return false; + return finish_folding(ctx, op); } static bool fold_multiply2(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ static bool fold_multiply2(OptContext *ctx, TCGOp *op) tcg_opt_gen_movi(ctx, op2, rh, h); return true; } - return false; + return finish_folding(ctx, op); } static bool fold_nand(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_multiply2(OptContext *ctx, TCGOp *op) static bool fold_nand(OptContext *ctx, TCGOp *op) { + uint64_t s_mask; + if (fold_const2_commutative(ctx, op) || fold_xi_to_not(ctx, op, -1)) { return true; } - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; - return false; + s_mask = arg_info(op->args[1])->s_mask + & arg_info(op->args[2])->s_mask; + return fold_masks_s(ctx, op, s_mask); } static bool fold_neg_no_const(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 9 ++------- 1 file changed, 2 insertions(+), 7 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_neg_no_const(OptContext *ctx, TCGOp *op) { /* Set to 1 all bits to the left of the rightmost. */ uint64_t z_mask = arg_info(op->args[1])->z_mask; - ctx->z_mask = -(z_mask & -z_mask); + z_mask = -(z_mask & -z_mask); - /* - * Because of fold_sub_to_neg, we want to always return true, - * via finish_folding. - */ - finish_folding(ctx, op); - return true; + return fold_masks_z(ctx, op, z_mask); } static bool fold_neg(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_neg(OptContext *ctx, TCGOp *op) static bool fold_nor(OptContext *ctx, TCGOp *op) { + uint64_t s_mask; + if (fold_const2_commutative(ctx, op) || fold_xi_to_not(ctx, op, 0)) { return true; } - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; - return false; + s_mask = arg_info(op->args[1])->s_mask + & arg_info(op->args[2])->s_mask; + return fold_masks_s(ctx, op, s_mask); } static bool fold_not(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 7 +------ 1 file changed, 1 insertion(+), 6 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_not(OptContext *ctx, TCGOp *op) if (fold_const1(ctx, op)) { return true; } - - ctx->s_mask = arg_info(op->args[1])->s_mask; - - /* Because of fold_to_not, we want to always return true, via finish. */ - finish_folding(ctx, op); - return true; + return fold_masks_s(ctx, op, arg_info(op->args[1])->s_mask); } static bool fold_or(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 13 ++++++++----- 1 file changed, 8 insertions(+), 5 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_not(OptContext *ctx, TCGOp *op) static bool fold_or(OptContext *ctx, TCGOp *op) { + uint64_t z_mask, s_mask; + TempOptInfo *t1, *t2; + if (fold_const2_commutative(ctx, op) || fold_xi_to_x(ctx, op, 0) || fold_xx_to_x(ctx, op)) { return true; } - ctx->z_mask = arg_info(op->args[1])->z_mask - | arg_info(op->args[2])->z_mask; - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; - return fold_masks(ctx, op); + t1 = arg_info(op->args[1]); + t2 = arg_info(op->args[2]); + z_mask = t1->z_mask | t2->z_mask; + s_mask = t1->s_mask & t2->s_mask; + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_orc(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_or(OptContext *ctx, TCGOp *op) static bool fold_orc(OptContext *ctx, TCGOp *op) { + uint64_t s_mask; + if (fold_const2(ctx, op) || fold_xx_to_i(ctx, op, -1) || fold_xi_to_x(ctx, op, -1) || @@ -XXX,XX +XXX,XX @@ static bool fold_orc(OptContext *ctx, TCGOp *op) return true; } - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; - return false; + s_mask = arg_info(op->args[1])->s_mask + & arg_info(op->args[2])->s_mask; + return fold_masks_s(ctx, op, s_mask); } static bool fold_qemu_ld(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Be careful not to call fold_masks_zs when the memory operation is wide enough to require multiple outputs, so split into two functions: fold_qemu_ld_1reg and fold_qemu_ld_2reg. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 26 +++++++++++++++++++++----- 1 file changed, 21 insertions(+), 5 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_orc(OptContext *ctx, TCGOp *op) return fold_masks_s(ctx, op, s_mask); } -static bool fold_qemu_ld(OptContext *ctx, TCGOp *op) +static bool fold_qemu_ld_1reg(OptContext *ctx, TCGOp *op) { const TCGOpDef *def = &tcg_op_defs[op->opc]; MemOpIdx oi = op->args[def->nb_oargs + def->nb_iargs]; MemOp mop = get_memop(oi); int width = 8 * memop_size(mop); + uint64_t z_mask = -1, s_mask = 0; if (width < 64) { if (mop & MO_SIGN) { - ctx->s_mask = MAKE_64BIT_MASK(width, 64 - width); + s_mask = MAKE_64BIT_MASK(width - 1, 64 - (width - 1)); } else { - ctx->z_mask = MAKE_64BIT_MASK(0, width); + z_mask = MAKE_64BIT_MASK(0, width); } } /* Opcodes that touch guest memory stop the mb optimization. */ ctx->prev_mb = NULL; - return false; + + return fold_masks_zs(ctx, op, z_mask, s_mask); +} + +static bool fold_qemu_ld_2reg(OptContext *ctx, TCGOp *op) +{ + /* Opcodes that touch guest memory stop the mb optimization. */ + ctx->prev_mb = NULL; + return finish_folding(ctx, op); } static bool fold_qemu_st(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ void tcg_optimize(TCGContext *s) break; case INDEX_op_qemu_ld_a32_i32: case INDEX_op_qemu_ld_a64_i32: + done = fold_qemu_ld_1reg(&ctx, op); + break; case INDEX_op_qemu_ld_a32_i64: case INDEX_op_qemu_ld_a64_i64: + if (TCG_TARGET_REG_BITS == 64) { + done = fold_qemu_ld_1reg(&ctx, op); + break; + } + QEMU_FALLTHROUGH; case INDEX_op_qemu_ld_a32_i128: case INDEX_op_qemu_ld_a64_i128: - done = fold_qemu_ld(&ctx, op); + done = fold_qemu_ld_2reg(&ctx, op); break; case INDEX_op_qemu_st8_a32_i32: case INDEX_op_qemu_st8_a64_i32: -- 2.43.0 Stores have no output operands, and so need no further work. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 11 +++++------ 1 file changed, 5 insertions(+), 6 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_qemu_st(OptContext *ctx, TCGOp *op) { /* Opcodes that touch guest memory stop the mb optimization. */ ctx->prev_mb = NULL; - return false; + return true; } static bool fold_remainder(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ static bool fold_tcg_st(OptContext *ctx, TCGOp *op) if (op->args[1] != tcgv_ptr_arg(tcg_env)) { remove_mem_copy_all(ctx); - return false; + return true; } switch (op->opc) { @@ -XXX,XX +XXX,XX @@ static bool fold_tcg_st(OptContext *ctx, TCGOp *op) g_assert_not_reached(); } remove_mem_copy_in(ctx, ofs, ofs + lm1); - return false; + return true; } static bool fold_tcg_st_memcopy(OptContext *ctx, TCGOp *op) @@ -XXX,XX +XXX,XX @@ static bool fold_tcg_st_memcopy(OptContext *ctx, TCGOp *op) TCGType type; if (op->args[1] != tcgv_ptr_arg(tcg_env)) { - fold_tcg_st(ctx, op); - return false; + return fold_tcg_st(ctx, op); } src = arg_temp(op->args[0]); @@ -XXX,XX +XXX,XX @@ static bool fold_tcg_st_memcopy(OptContext *ctx, TCGOp *op) last = ofs + tcg_type_size(type) - 1; remove_mem_copy_in(ctx, ofs, last); record_mem_copy(ctx, type, src, ofs, last); - return false; + return true; } static bool fold_xor(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_remainder(OptContext *ctx, TCGOp *op) fold_xx_to_i(ctx, op, 0)) { return true; } - return false; + return finish_folding(ctx, op); } static bool fold_setcond_zmask(OptContext *ctx, TCGOp *op, bool neg) -- 2.43.0 Change return from bool to int; distinguish between complete folding, simplification, and no change. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 22 ++++++++++++++-------- 1 file changed, 14 insertions(+), 8 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_remainder(OptContext *ctx, TCGOp *op) return finish_folding(ctx, op); } -static bool fold_setcond_zmask(OptContext *ctx, TCGOp *op, bool neg) +/* Return 1 if finished, -1 if simplified, 0 if unchanged. */ +static int fold_setcond_zmask(OptContext *ctx, TCGOp *op, bool neg) { uint64_t a_zmask, b_val; TCGCond cond; @@ -XXX,XX +XXX,XX @@ static bool fold_setcond_zmask(OptContext *ctx, TCGOp *op, bool neg) op->opc = xor_opc; op->args[2] = arg_new_constant(ctx, 1); } - return false; + return -1; } } - - return false; + return 0; } static void fold_setcond_tst_pow2(OptContext *ctx, TCGOp *op, bool neg) @@ -XXX,XX +XXX,XX @@ static bool fold_setcond(OptContext *ctx, TCGOp *op) return tcg_opt_gen_movi(ctx, op, op->args[0], i); } - if (fold_setcond_zmask(ctx, op, false)) { + i = fold_setcond_zmask(ctx, op, false); + if (i > 0) { return true; } - fold_setcond_tst_pow2(ctx, op, false); + if (i == 0) { + fold_setcond_tst_pow2(ctx, op, false); + } ctx->z_mask = 1; return false; @@ -XXX,XX +XXX,XX @@ static bool fold_negsetcond(OptContext *ctx, TCGOp *op) return tcg_opt_gen_movi(ctx, op, op->args[0], -i); } - if (fold_setcond_zmask(ctx, op, true)) { + i = fold_setcond_zmask(ctx, op, true); + if (i > 0) { return true; } - fold_setcond_tst_pow2(ctx, op, true); + if (i == 0) { + fold_setcond_tst_pow2(ctx, op, true); + } /* Value is {0,-1} so all bits are repetitions of the sign. */ ctx->s_mask = -1; -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_setcond(OptContext *ctx, TCGOp *op) fold_setcond_tst_pow2(ctx, op, false); } - ctx->z_mask = 1; - return false; + return fold_masks_z(ctx, op, 1); } static bool fold_negsetcond(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_negsetcond(OptContext *ctx, TCGOp *op) } /* Value is {0,-1} so all bits are repetitions of the sign. */ - ctx->s_mask = -1; - return false; + return fold_masks_s(ctx, op, -1); } static bool fold_setcond2(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_setcond2(OptContext *ctx, TCGOp *op) return fold_setcond(ctx, op); } - ctx->z_mask = 1; - return false; + return fold_masks_z(ctx, op, 1); do_setcond_const: return tcg_opt_gen_movi(ctx, op, op->args[0], i); -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_cmp_vec(OptContext *ctx, TCGOp *op) if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) { op->args[3] = tcg_swap_cond(op->args[3]); } - return false; + return finish_folding(ctx, op); } static bool fold_cmpsel_vec(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_cmpsel_vec(OptContext *ctx, TCGOp *op) if (swap_commutative(op->args[0], &op->args[4], &op->args[3])) { op->args[5] = tcg_invert_cond(op->args[5]); } - return false; + return finish_folding(ctx, op); } static bool fold_sextract(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 24 +++++++++--------------- 1 file changed, 9 insertions(+), 15 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_cmpsel_vec(OptContext *ctx, TCGOp *op) static bool fold_sextract(OptContext *ctx, TCGOp *op) { uint64_t z_mask, s_mask, s_mask_old; + TempOptInfo *t1 = arg_info(op->args[1]); int pos = op->args[2]; int len = op->args[3]; - if (arg_is_const(op->args[1])) { - uint64_t t; - - t = arg_info(op->args[1])->val; - t = sextract64(t, pos, len); - return tcg_opt_gen_movi(ctx, op, op->args[0], t); + if (ti_is_const(t1)) { + return tcg_opt_gen_movi(ctx, op, op->args[0], + sextract64(ti_const_val(t1), pos, len)); } - z_mask = arg_info(op->args[1])->z_mask; - z_mask = sextract64(z_mask, pos, len); - ctx->z_mask = z_mask; - - s_mask_old = arg_info(op->args[1])->s_mask; - s_mask = sextract64(s_mask_old, pos, len); - s_mask |= MAKE_64BIT_MASK(len, 64 - len); - ctx->s_mask = s_mask; + s_mask_old = t1->s_mask; + s_mask = s_mask_old >> pos; + s_mask |= -1ull << (len - 1); if (0 && pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { return true; } - return fold_masks(ctx, op); + z_mask = sextract64(t1->z_mask, pos, len); + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_shift(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 27 ++++++++++++++------------- 1 file changed, 14 insertions(+), 13 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_sextract(OptContext *ctx, TCGOp *op) static bool fold_shift(OptContext *ctx, TCGOp *op) { uint64_t s_mask, z_mask, sign; + TempOptInfo *t1, *t2; if (fold_const2(ctx, op) || fold_ix_to_i(ctx, op, 0) || @@ -XXX,XX +XXX,XX @@ static bool fold_shift(OptContext *ctx, TCGOp *op) return true; } - s_mask = arg_info(op->args[1])->s_mask; - z_mask = arg_info(op->args[1])->z_mask; + t1 = arg_info(op->args[1]); + t2 = arg_info(op->args[2]); + s_mask = t1->s_mask; + z_mask = t1->z_mask; - if (arg_is_const(op->args[2])) { - int sh = arg_info(op->args[2])->val; - - ctx->z_mask = do_constant_folding(op->opc, ctx->type, z_mask, sh); + if (ti_is_const(t2)) { + int sh = ti_const_val(t2); + z_mask = do_constant_folding(op->opc, ctx->type, z_mask, sh); s_mask = do_constant_folding(op->opc, ctx->type, s_mask, sh); - return fold_masks(ctx, op); + return fold_masks_zs(ctx, op, z_mask, s_mask); } switch (op->opc) { @@ -XXX,XX +XXX,XX @@ static bool fold_shift(OptContext *ctx, TCGOp *op) * Arithmetic right shift will not reduce the number of * input sign repetitions. */ - ctx->s_mask = s_mask; - break; + return fold_masks_s(ctx, op, s_mask); CASE_OP_32_64(shr): /* * If the sign bit is known zero, then logical right shift - * will not reduced the number of input sign repetitions. + * will not reduce the number of input sign repetitions. */ - sign = (s_mask & -s_mask) >> 1; + sign = -s_mask; if (sign && !(z_mask & sign)) { - ctx->s_mask = s_mask; + return fold_masks_s(ctx, op, s_mask); } break; default: break; } - return false; + return finish_folding(ctx, op); } static bool fold_sub_to_neg(OptContext *ctx, TCGOp *op) -- 2.43.0 Merge the two conditions, sign != 0 && !(z_mask & sign), by testing ~z_mask & sign. If sign == 0, the logical and will produce false. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_sextract(OptContext *ctx, TCGOp *op) static bool fold_shift(OptContext *ctx, TCGOp *op) { - uint64_t s_mask, z_mask, sign; + uint64_t s_mask, z_mask; TempOptInfo *t1, *t2; if (fold_const2(ctx, op) || @@ -XXX,XX +XXX,XX @@ static bool fold_shift(OptContext *ctx, TCGOp *op) * If the sign bit is known zero, then logical right shift * will not reduce the number of input sign repetitions. */ - sign = -s_mask; - if (sign && !(z_mask & sign)) { + if (~z_mask & -s_mask) { return fold_masks_s(ctx, op, s_mask); } break; -- 2.43.0 Duplicate fold_sub_vec into fold_sub instead of calling it, now that fold_sub_vec always returns true. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_sub_vec(OptContext *ctx, TCGOp *op) fold_sub_to_neg(ctx, op)) { return true; } - return false; + return finish_folding(ctx, op); } static bool fold_sub(OptContext *ctx, TCGOp *op) { - if (fold_const2(ctx, op) || fold_sub_vec(ctx, op)) { + if (fold_const2(ctx, op) || + fold_xx_to_i(ctx, op, 0) || + fold_xi_to_x(ctx, op, 0) || + fold_sub_to_neg(ctx, op)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool fold_sub(OptContext *ctx, TCGOp *op) ? INDEX_op_add_i32 : INDEX_op_add_i64); op->args[2] = arg_new_constant(ctx, -val); } - return false; + return finish_folding(ctx, op); } static bool fold_sub2(OptContext *ctx, TCGOp *op) -- 2.43.0 Avoid the use of the OptContext slots. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_sub2(OptContext *ctx, TCGOp *op) static bool fold_tcg_ld(OptContext *ctx, TCGOp *op) { + uint64_t z_mask = -1, s_mask = 0; + /* We can't do any folding with a load, but we can record bits. */ switch (op->opc) { CASE_OP_32_64(ld8s): - ctx->s_mask = MAKE_64BIT_MASK(8, 56); + s_mask = INT8_MIN; break; CASE_OP_32_64(ld8u): - ctx->z_mask = MAKE_64BIT_MASK(0, 8); + z_mask = MAKE_64BIT_MASK(0, 8); break; CASE_OP_32_64(ld16s): - ctx->s_mask = MAKE_64BIT_MASK(16, 48); + s_mask = INT16_MIN; break; CASE_OP_32_64(ld16u): - ctx->z_mask = MAKE_64BIT_MASK(0, 16); + z_mask = MAKE_64BIT_MASK(0, 16); break; case INDEX_op_ld32s_i64: - ctx->s_mask = MAKE_64BIT_MASK(32, 32); + s_mask = INT32_MIN; break; case INDEX_op_ld32u_i64: - ctx->z_mask = MAKE_64BIT_MASK(0, 32); + z_mask = MAKE_64BIT_MASK(0, 32); break; default: g_assert_not_reached(); } - return false; + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_tcg_ld_memcopy(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_tcg_ld_memcopy(OptContext *ctx, TCGOp *op) TCGType type; if (op->args[1] != tcgv_ptr_arg(tcg_env)) { - return false; + return finish_folding(ctx, op); } type = ctx->type; -- 2.43.0 Avoid the use of the OptContext slots. Find TempOptInfo once. Remove fold_masks as the function becomes unused. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 18 ++++++++---------- 1 file changed, 8 insertions(+), 10 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_masks_s(OptContext *ctx, TCGOp *op, uint64_t s_mask) return fold_masks_zs(ctx, op, -1, s_mask); } -static bool fold_masks(OptContext *ctx, TCGOp *op) -{ - return fold_masks_zs(ctx, op, ctx->z_mask, ctx->s_mask); -} - /* * An "affected" mask bit is 0 if and only if the result is identical * to the first input. Thus if the entire mask is 0, the operation @@ -XXX,XX +XXX,XX @@ static bool fold_tcg_st_memcopy(OptContext *ctx, TCGOp *op) static bool fold_xor(OptContext *ctx, TCGOp *op) { + uint64_t z_mask, s_mask; + TempOptInfo *t1, *t2; + if (fold_const2_commutative(ctx, op) || fold_xx_to_i(ctx, op, 0) || fold_xi_to_x(ctx, op, 0) || @@ -XXX,XX +XXX,XX @@ static bool fold_xor(OptContext *ctx, TCGOp *op) return true; } - ctx->z_mask = arg_info(op->args[1])->z_mask - | arg_info(op->args[2])->z_mask; - ctx->s_mask = arg_info(op->args[1])->s_mask - & arg_info(op->args[2])->s_mask; - return fold_masks(ctx, op); + t1 = arg_info(op->args[1]); + t2 = arg_info(op->args[2]); + z_mask = t1->z_mask | t2->z_mask; + s_mask = t1->s_mask & t2->s_mask; + return fold_masks_zs(ctx, op, z_mask, s_mask); } static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op) -- 2.43.0 Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op) return fold_orc(ctx, op); } } - return false; + return finish_folding(ctx, op); } /* Propagate constants and copies, fold constant expressions. */ -- 2.43.0 All non-default cases now finish folding within each function. Do the same with the default case and assert it is done after. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 6 ++---- 1 file changed, 2 insertions(+), 4 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ void tcg_optimize(TCGContext *s) done = true; break; default: + done = finish_folding(&ctx, op); break; } - - if (!done) { - finish_folding(&ctx, op); - } + tcg_debug_assert(done); } } -- 2.43.0 All mask setting is now done with parameters via fold_masks_*. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 13 ------------- 1 file changed, 13 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ typedef struct OptContext { QSIMPLEQ_HEAD(, MemCopyInfo) mem_free; /* In flight values from optimization. */ - uint64_t z_mask; /* mask bit is 0 iff value bit is 0 */ - uint64_t s_mask; /* mask bit is 1 if value bit matches msb */ TCGType type; } OptContext; @@ -XXX,XX +XXX,XX @@ static bool finish_folding(OptContext *ctx, TCGOp *op) for (i = 0; i < nb_oargs; i++) { TCGTemp *ts = arg_temp(op->args[i]); reset_ts(ctx, ts); - /* - * Save the corresponding known-zero/sign bits mask for the - * first output argument (only one supported so far). - */ - if (i == 0) { - ts_info(ts)->z_mask = ctx->z_mask; - } } return true; } @@ -XXX,XX +XXX,XX @@ void tcg_optimize(TCGContext *s) ctx.type = TCG_TYPE_I32; } - /* Assume all bits affected, no bits known zero, no sign reps. */ - ctx.z_mask = -1; - ctx.s_mask = 0; - /* * Process each opcode. * Sorted alphabetically by opcode as much as possible. -- 2.43.0 All instances of s_mask have been converted to the new representation. We can now re-enable usage. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_exts(OptContext *ctx, TCGOp *op) g_assert_not_reached(); } - if (0 && !type_change && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { + if (!type_change && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { return true; } @@ -XXX,XX +XXX,XX @@ static bool fold_sextract(OptContext *ctx, TCGOp *op) s_mask = s_mask_old >> pos; s_mask |= -1ull << (len - 1); - if (0 && pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { + if (pos == 0 && fold_affected_mask(ctx, op, s_mask & ~s_mask_old)) { return true; } -- 2.43.0 The big comment just above says functions should be sorted. Add forward declarations as needed. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 114 +++++++++++++++++++++++++------------------------ 1 file changed, 59 insertions(+), 55 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_xx_to_x(OptContext *ctx, TCGOp *op) * 3) those that produce information about the result value. */ +static bool fold_or(OptContext *ctx, TCGOp *op); +static bool fold_orc(OptContext *ctx, TCGOp *op); +static bool fold_xor(OptContext *ctx, TCGOp *op); + static bool fold_add(OptContext *ctx, TCGOp *op) { if (fold_const2_commutative(ctx, op) || @@ -XXX,XX +XXX,XX @@ static bool fold_andc(OptContext *ctx, TCGOp *op) return fold_masks_zs(ctx, op, z_mask, s_mask); } +static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op) +{ + /* If true and false values are the same, eliminate the cmp. */ + if (args_are_copies(op->args[2], op->args[3])) { + return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[2]); + } + + if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])) { + uint64_t tv = arg_info(op->args[2])->val; + uint64_t fv = arg_info(op->args[3])->val; + + if (tv == -1 && fv == 0) { + return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); + } + if (tv == 0 && fv == -1) { + if (TCG_TARGET_HAS_not_vec) { + op->opc = INDEX_op_not_vec; + return fold_not(ctx, op); + } else { + op->opc = INDEX_op_xor_vec; + op->args[2] = arg_new_constant(ctx, -1); + return fold_xor(ctx, op); + } + } + } + if (arg_is_const(op->args[2])) { + uint64_t tv = arg_info(op->args[2])->val; + if (tv == -1) { + op->opc = INDEX_op_or_vec; + op->args[2] = op->args[3]; + return fold_or(ctx, op); + } + if (tv == 0 && TCG_TARGET_HAS_andc_vec) { + op->opc = INDEX_op_andc_vec; + op->args[2] = op->args[1]; + op->args[1] = op->args[3]; + return fold_andc(ctx, op); + } + } + if (arg_is_const(op->args[3])) { + uint64_t fv = arg_info(op->args[3])->val; + if (fv == 0) { + op->opc = INDEX_op_and_vec; + return fold_and(ctx, op); + } + if (fv == -1 && TCG_TARGET_HAS_orc_vec) { + op->opc = INDEX_op_orc_vec; + op->args[2] = op->args[1]; + op->args[1] = op->args[3]; + return fold_orc(ctx, op); + } + } + return finish_folding(ctx, op); +} + static bool fold_brcond(OptContext *ctx, TCGOp *op) { int i = do_constant_folding_cond1(ctx, op, NO_DEST, &op->args[0], @@ -XXX,XX +XXX,XX @@ static bool fold_xor(OptContext *ctx, TCGOp *op) return fold_masks_zs(ctx, op, z_mask, s_mask); } -static bool fold_bitsel_vec(OptContext *ctx, TCGOp *op) -{ - /* If true and false values are the same, eliminate the cmp. */ - if (args_are_copies(op->args[2], op->args[3])) { - return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[2]); - } - - if (arg_is_const(op->args[2]) && arg_is_const(op->args[3])) { - uint64_t tv = arg_info(op->args[2])->val; - uint64_t fv = arg_info(op->args[3])->val; - - if (tv == -1 && fv == 0) { - return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[1]); - } - if (tv == 0 && fv == -1) { - if (TCG_TARGET_HAS_not_vec) { - op->opc = INDEX_op_not_vec; - return fold_not(ctx, op); - } else { - op->opc = INDEX_op_xor_vec; - op->args[2] = arg_new_constant(ctx, -1); - return fold_xor(ctx, op); - } - } - } - if (arg_is_const(op->args[2])) { - uint64_t tv = arg_info(op->args[2])->val; - if (tv == -1) { - op->opc = INDEX_op_or_vec; - op->args[2] = op->args[3]; - return fold_or(ctx, op); - } - if (tv == 0 && TCG_TARGET_HAS_andc_vec) { - op->opc = INDEX_op_andc_vec; - op->args[2] = op->args[1]; - op->args[1] = op->args[3]; - return fold_andc(ctx, op); - } - } - if (arg_is_const(op->args[3])) { - uint64_t fv = arg_info(op->args[3])->val; - if (fv == 0) { - op->opc = INDEX_op_and_vec; - return fold_and(ctx, op); - } - if (fv == -1 && TCG_TARGET_HAS_orc_vec) { - op->opc = INDEX_op_orc_vec; - op->args[2] = op->args[1]; - op->args[1] = op->args[3]; - return fold_orc(ctx, op); - } - } - return finish_folding(ctx, op); -} - /* Propagate constants and copies, fold constant expressions. */ void tcg_optimize(TCGContext *s) { -- 2.43.0 The big comment just above says functions should be sorted. Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- tcg/optimize.c | 60 +++++++++++++++++++++++++------------------------- 1 file changed, 30 insertions(+), 30 deletions(-) diff --git a/tcg/optimize.c b/tcg/optimize.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/optimize.c +++ b/tcg/optimize.c @@ -XXX,XX +XXX,XX @@ static bool fold_call(OptContext *ctx, TCGOp *op) return true; } +static bool fold_cmp_vec(OptContext *ctx, TCGOp *op) +{ + /* Canonicalize the comparison to put immediate second. */ + if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) { + op->args[3] = tcg_swap_cond(op->args[3]); + } + return finish_folding(ctx, op); +} + +static bool fold_cmpsel_vec(OptContext *ctx, TCGOp *op) +{ + /* If true and false values are the same, eliminate the cmp. */ + if (args_are_copies(op->args[3], op->args[4])) { + return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[3]); + } + + /* Canonicalize the comparison to put immediate second. */ + if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) { + op->args[5] = tcg_swap_cond(op->args[5]); + } + /* + * Canonicalize the "false" input reg to match the destination, + * so that the tcg backend can implement "move if true". + */ + if (swap_commutative(op->args[0], &op->args[4], &op->args[3])) { + op->args[5] = tcg_invert_cond(op->args[5]); + } + return finish_folding(ctx, op); +} + static bool fold_count_zeros(OptContext *ctx, TCGOp *op) { uint64_t z_mask, s_mask; @@ -XXX,XX +XXX,XX @@ static bool fold_setcond2(OptContext *ctx, TCGOp *op) return tcg_opt_gen_movi(ctx, op, op->args[0], i); } -static bool fold_cmp_vec(OptContext *ctx, TCGOp *op) -{ - /* Canonicalize the comparison to put immediate second. */ - if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) { - op->args[3] = tcg_swap_cond(op->args[3]); - } - return finish_folding(ctx, op); -} - -static bool fold_cmpsel_vec(OptContext *ctx, TCGOp *op) -{ - /* If true and false values are the same, eliminate the cmp. */ - if (args_are_copies(op->args[3], op->args[4])) { - return tcg_opt_gen_mov(ctx, op, op->args[0], op->args[3]); - } - - /* Canonicalize the comparison to put immediate second. */ - if (swap_commutative(NO_DEST, &op->args[1], &op->args[2])) { - op->args[5] = tcg_swap_cond(op->args[5]); - } - /* - * Canonicalize the "false" input reg to match the destination, - * so that the tcg backend can implement "move if true". - */ - if (swap_commutative(op->args[0], &op->args[4], &op->args[3])) { - op->args[5] = tcg_invert_cond(op->args[5]); - } - return finish_folding(ctx, op); -} - static bool fold_sextract(OptContext *ctx, TCGOp *op) { uint64_t z_mask, s_mask, s_mask_old; -- 2.43.0 We currently have a flag, float_muladd_halve_result, to scale the result by 2**-1. Extend this to handle arbitrary scaling. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/fpu/softfloat.h | 6 ++++ fpu/softfloat.c | 58 ++++++++++++++++++++++----------------- fpu/softfloat-parts.c.inc | 7 +++-- 3 files changed, 44 insertions(+), 27 deletions(-) diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat.h +++ b/include/fpu/softfloat.h @@ -XXX,XX +XXX,XX @@ float16 float16_add(float16, float16, float_status *status); float16 float16_sub(float16, float16, float_status *status); float16 float16_mul(float16, float16, float_status *status); float16 float16_muladd(float16, float16, float16, int, float_status *status); +float16 float16_muladd_scalbn(float16, float16, float16, + int, int, float_status *status); float16 float16_div(float16, float16, float_status *status); float16 float16_scalbn(float16, int, float_status *status); float16 float16_min(float16, float16, float_status *status); @@ -XXX,XX +XXX,XX @@ float32 float32_mul(float32, float32, float_status *status); float32 float32_div(float32, float32, float_status *status); float32 float32_rem(float32, float32, float_status *status); float32 float32_muladd(float32, float32, float32, int, float_status *status); +float32 float32_muladd_scalbn(float32, float32, float32, + int, int, float_status *status); float32 float32_sqrt(float32, float_status *status); float32 float32_exp2(float32, float_status *status); float32 float32_log2(float32, float_status *status); @@ -XXX,XX +XXX,XX @@ float64 float64_mul(float64, float64, float_status *status); float64 float64_div(float64, float64, float_status *status); float64 float64_rem(float64, float64, float_status *status); float64 float64_muladd(float64, float64, float64, int, float_status *status); +float64 float64_muladd_scalbn(float64, float64, float64, + int, int, float_status *status); float64 float64_sqrt(float64, float_status *status); float64 float64_log2(float64, float_status *status); FloatRelation float64_compare(float64, float64, float_status *status); diff --git a/fpu/softfloat.c b/fpu/softfloat.c index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -XXX,XX +XXX,XX @@ static FloatParts128 *parts128_mul(FloatParts128 *a, FloatParts128 *b, #define parts_mul(A, B, S) \ PARTS_GENERIC_64_128(mul, A)(A, B, S) -static FloatParts64 *parts64_muladd(FloatParts64 *a, FloatParts64 *b, - FloatParts64 *c, int flags, - float_status *s); -static FloatParts128 *parts128_muladd(FloatParts128 *a, FloatParts128 *b, - FloatParts128 *c, int flags, - float_status *s); +static FloatParts64 *parts64_muladd_scalbn(FloatParts64 *a, FloatParts64 *b, + FloatParts64 *c, int scale, + int flags, float_status *s); +static FloatParts128 *parts128_muladd_scalbn(FloatParts128 *a, FloatParts128 *b, + FloatParts128 *c, int scale, + int flags, float_status *s); -#define parts_muladd(A, B, C, Z, S) \ - PARTS_GENERIC_64_128(muladd, A)(A, B, C, Z, S) +#define parts_muladd_scalbn(A, B, C, Z, Y, S) \ + PARTS_GENERIC_64_128(muladd_scalbn, A)(A, B, C, Z, Y, S) static FloatParts64 *parts64_div(FloatParts64 *a, FloatParts64 *b, float_status *s); @@ -XXX,XX +XXX,XX @@ floatx80_mul(floatx80 a, floatx80 b, float_status *status) * Fused multiply-add */ -float16 QEMU_FLATTEN float16_muladd(float16 a, float16 b, float16 c, - int flags, float_status *status) +float16 QEMU_FLATTEN +float16_muladd_scalbn(float16 a, float16 b, float16 c, + int scale, int flags, float_status *status) { FloatParts64 pa, pb, pc, *pr; float16_unpack_canonical(&pa, a, status); float16_unpack_canonical(&pb, b, status); float16_unpack_canonical(&pc, c, status); - pr = parts_muladd(&pa, &pb, &pc, flags, status); + pr = parts_muladd_scalbn(&pa, &pb, &pc, scale, flags, status); return float16_round_pack_canonical(pr, status); } -static float32 QEMU_SOFTFLOAT_ATTR -soft_f32_muladd(float32 a, float32 b, float32 c, int flags, - float_status *status) +float16 float16_muladd(float16 a, float16 b, float16 c, + int flags, float_status *status) +{ + return float16_muladd_scalbn(a, b, c, 0, flags, status); +} + +float32 QEMU_SOFTFLOAT_ATTR +float32_muladd_scalbn(float32 a, float32 b, float32 c, + int scale, int flags, float_status *status) { FloatParts64 pa, pb, pc, *pr; float32_unpack_canonical(&pa, a, status); float32_unpack_canonical(&pb, b, status); float32_unpack_canonical(&pc, c, status); - pr = parts_muladd(&pa, &pb, &pc, flags, status); + pr = parts_muladd_scalbn(&pa, &pb, &pc, scale, flags, status); return float32_round_pack_canonical(pr, status); } -static float64 QEMU_SOFTFLOAT_ATTR -soft_f64_muladd(float64 a, float64 b, float64 c, int flags, - float_status *status) +float64 QEMU_SOFTFLOAT_ATTR +float64_muladd_scalbn(float64 a, float64 b, float64 c, + int scale, int flags, float_status *status) { FloatParts64 pa, pb, pc, *pr; float64_unpack_canonical(&pa, a, status); float64_unpack_canonical(&pb, b, status); float64_unpack_canonical(&pc, c, status); - pr = parts_muladd(&pa, &pb, &pc, flags, status); + pr = parts_muladd_scalbn(&pa, &pb, &pc, scale, flags, status); return float64_round_pack_canonical(pr, status); } @@ -XXX,XX +XXX,XX @@ float32_muladd(float32 xa, float32 xb, float32 xc, int flags, float_status *s) return ur.s; soft: - return soft_f32_muladd(ua.s, ub.s, uc.s, flags, s); + return float32_muladd_scalbn(ua.s, ub.s, uc.s, 0, flags, s); } float64 QEMU_FLATTEN @@ -XXX,XX +XXX,XX @@ float64_muladd(float64 xa, float64 xb, float64 xc, int flags, float_status *s) return ur.s; soft: - return soft_f64_muladd(ua.s, ub.s, uc.s, flags, s); + return float64_muladd_scalbn(ua.s, ub.s, uc.s, 0, flags, s); } float64 float64r32_muladd(float64 a, float64 b, float64 c, @@ -XXX,XX +XXX,XX @@ float64 float64r32_muladd(float64 a, float64 b, float64 c, float64_unpack_canonical(&pa, a, status); float64_unpack_canonical(&pb, b, status); float64_unpack_canonical(&pc, c, status); - pr = parts_muladd(&pa, &pb, &pc, flags, status); + pr = parts_muladd_scalbn(&pa, &pb, &pc, 0, flags, status); return float64r32_round_pack_canonical(pr, status); } @@ -XXX,XX +XXX,XX @@ bfloat16 QEMU_FLATTEN bfloat16_muladd(bfloat16 a, bfloat16 b, bfloat16 c, bfloat16_unpack_canonical(&pa, a, status); bfloat16_unpack_canonical(&pb, b, status); bfloat16_unpack_canonical(&pc, c, status); - pr = parts_muladd(&pa, &pb, &pc, flags, status); + pr = parts_muladd_scalbn(&pa, &pb, &pc, 0, flags, status); return bfloat16_round_pack_canonical(pr, status); } @@ -XXX,XX +XXX,XX @@ float128 QEMU_FLATTEN float128_muladd(float128 a, float128 b, float128 c, float128_unpack_canonical(&pa, a, status); float128_unpack_canonical(&pb, b, status); float128_unpack_canonical(&pc, c, status); - pr = parts_muladd(&pa, &pb, &pc, flags, status); + pr = parts_muladd_scalbn(&pa, &pb, &pc, 0, flags, status); return float128_round_pack_canonical(pr, status); } @@ -XXX,XX +XXX,XX @@ float32 float32_exp2(float32 a, float_status *status) float64_unpack_canonical(&rp, float64_one, status); for (i = 0 ; i < 15 ; i++) { + float64_unpack_canonical(&tp, float32_exp2_coefficients[i], status); - rp = *parts_muladd(&tp, &xnp, &rp, 0, status); + rp = *parts_muladd_scalbn(&tp, &xnp, &rp, 0, 0, status); xnp = *parts_mul(&xnp, &xp, status); } diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(mul)(FloatPartsN *a, FloatPartsN *b, * Requires A and C extracted into a double-sized structure to provide the * extra space for the widening multiply. */ -static FloatPartsN *partsN(muladd)(FloatPartsN *a, FloatPartsN *b, - FloatPartsN *c, int flags, float_status *s) +static FloatPartsN *partsN(muladd_scalbn)(FloatPartsN *a, FloatPartsN *b, + FloatPartsN *c, int scale, + int flags, float_status *s) { int ab_mask, abc_mask; FloatPartsW p_widen, c_widen; @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(muladd)(FloatPartsN *a, FloatPartsN *b, a->exp = p_widen.exp; return_normal: + /* TODO: Replace all use of float_muladd_halve_result with scale. */ if (flags & float_muladd_halve_result) { a->exp -= 1; } + a->exp += scale; finish_sign: if (flags & float_muladd_negate_result) { a->sign ^= 1; -- 2.43.0 Use the scalbn interface instead of float_muladd_halve_result. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/tcg/helper-a64.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) diff --git a/target/arm/tcg/helper-a64.c b/target/arm/tcg/helper-a64.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/helper-a64.c +++ b/target/arm/tcg/helper-a64.c @@ -XXX,XX +XXX,XX @@ uint32_t HELPER(rsqrtsf_f16)(uint32_t a, uint32_t b, float_status *fpst) (float16_is_infinity(b) && float16_is_zero(a))) { return float16_one_point_five; } - return float16_muladd(a, b, float16_three, float_muladd_halve_result, fpst); + return float16_muladd_scalbn(a, b, float16_three, -1, 0, fpst); } float32 HELPER(rsqrtsf_f32)(float32 a, float32 b, float_status *fpst) @@ -XXX,XX +XXX,XX @@ float32 HELPER(rsqrtsf_f32)(float32 a, float32 b, float_status *fpst) (float32_is_infinity(b) && float32_is_zero(a))) { return float32_one_point_five; } - return float32_muladd(a, b, float32_three, float_muladd_halve_result, fpst); + return float32_muladd_scalbn(a, b, float32_three, -1, 0, fpst); } float64 HELPER(rsqrtsf_f64)(float64 a, float64 b, float_status *fpst) @@ -XXX,XX +XXX,XX @@ float64 HELPER(rsqrtsf_f64)(float64 a, float64 b, float_status *fpst) (float64_is_infinity(b) && float64_is_zero(a))) { return float64_one_point_five; } - return float64_muladd(a, b, float64_three, float_muladd_halve_result, fpst); + return float64_muladd_scalbn(a, b, float64_three, -1, 0, fpst); } /* Floating-point reciprocal exponent - see FPRecpX in ARM ARM */ -- 2.43.0 Use the scalbn interface instead of float_muladd_halve_result. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/sparc/helper.h | 4 +- target/sparc/fop_helper.c | 8 ++-- target/sparc/translate.c | 80 +++++++++++++++++++++++---------------- 3 files changed, 54 insertions(+), 38 deletions(-) diff --git a/target/sparc/helper.h b/target/sparc/helper.h index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/helper.h +++ b/target/sparc/helper.h @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(faddd, TCG_CALL_NO_WG, f64, env, f64, f64) DEF_HELPER_FLAGS_3(fsubd, TCG_CALL_NO_WG, f64, env, f64, f64) DEF_HELPER_FLAGS_3(fmuld, TCG_CALL_NO_WG, f64, env, f64, f64) DEF_HELPER_FLAGS_3(fdivd, TCG_CALL_NO_WG, f64, env, f64, f64) -DEF_HELPER_FLAGS_5(fmaddd, TCG_CALL_NO_WG, f64, env, f64, f64, f64, i32) +DEF_HELPER_FLAGS_6(fmaddd, TCG_CALL_NO_WG, f64, env, f64, f64, f64, s32, i32) DEF_HELPER_FLAGS_3(fnaddd, TCG_CALL_NO_WG, f64, env, f64, f64) DEF_HELPER_FLAGS_3(fnmuld, TCG_CALL_NO_WG, f64, env, f64, f64) @@ -XXX,XX +XXX,XX @@ DEF_HELPER_FLAGS_3(fadds, TCG_CALL_NO_WG, f32, env, f32, f32) DEF_HELPER_FLAGS_3(fsubs, TCG_CALL_NO_WG, f32, env, f32, f32) DEF_HELPER_FLAGS_3(fmuls, TCG_CALL_NO_WG, f32, env, f32, f32) DEF_HELPER_FLAGS_3(fdivs, TCG_CALL_NO_WG, f32, env, f32, f32) -DEF_HELPER_FLAGS_5(fmadds, TCG_CALL_NO_WG, f32, env, f32, f32, f32, i32) +DEF_HELPER_FLAGS_6(fmadds, TCG_CALL_NO_WG, f32, env, f32, f32, f32, s32, i32) DEF_HELPER_FLAGS_3(fnadds, TCG_CALL_NO_WG, f32, env, f32, f32) DEF_HELPER_FLAGS_3(fnmuls, TCG_CALL_NO_WG, f32, env, f32, f32) diff --git a/target/sparc/fop_helper.c b/target/sparc/fop_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/fop_helper.c +++ b/target/sparc/fop_helper.c @@ -XXX,XX +XXX,XX @@ Int128 helper_fsqrtq(CPUSPARCState *env, Int128 src) } float32 helper_fmadds(CPUSPARCState *env, float32 s1, - float32 s2, float32 s3, uint32_t op) + float32 s2, float32 s3, int32_t sc, uint32_t op) { - float32 ret = float32_muladd(s1, s2, s3, op, &env->fp_status); + float32 ret = float32_muladd_scalbn(s1, s2, s3, sc, op, &env->fp_status); check_ieee_exceptions(env, GETPC()); return ret; } float64 helper_fmaddd(CPUSPARCState *env, float64 s1, - float64 s2, float64 s3, uint32_t op) + float64 s2, float64 s3, int32_t sc, uint32_t op) { - float64 ret = float64_muladd(s1, s2, s3, op, &env->fp_status); + float64 ret = float64_muladd_scalbn(s1, s2, s3, sc, op, &env->fp_status); check_ieee_exceptions(env, GETPC()); return ret; } diff --git a/target/sparc/translate.c b/target/sparc/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/translate.c +++ b/target/sparc/translate.c @@ -XXX,XX +XXX,XX @@ static void gen_op_fabsq(TCGv_i128 dst, TCGv_i128 src) static void gen_op_fmadds(TCGv_i32 d, TCGv_i32 s1, TCGv_i32 s2, TCGv_i32 s3) { - gen_helper_fmadds(d, tcg_env, s1, s2, s3, tcg_constant_i32(0)); + TCGv_i32 z = tcg_constant_i32(0); + gen_helper_fmadds(d, tcg_env, s1, s2, s3, z, z); } static void gen_op_fmaddd(TCGv_i64 d, TCGv_i64 s1, TCGv_i64 s2, TCGv_i64 s3) { - gen_helper_fmaddd(d, tcg_env, s1, s2, s3, tcg_constant_i32(0)); + TCGv_i32 z = tcg_constant_i32(0); + gen_helper_fmaddd(d, tcg_env, s1, s2, s3, z, z); } static void gen_op_fmsubs(TCGv_i32 d, TCGv_i32 s1, TCGv_i32 s2, TCGv_i32 s3) { - int op = float_muladd_negate_c; - gen_helper_fmadds(d, tcg_env, s1, s2, s3, tcg_constant_i32(op)); + TCGv_i32 z = tcg_constant_i32(0); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_c); + gen_helper_fmadds(d, tcg_env, s1, s2, s3, z, op); } static void gen_op_fmsubd(TCGv_i64 d, TCGv_i64 s1, TCGv_i64 s2, TCGv_i64 s3) { - int op = float_muladd_negate_c; - gen_helper_fmaddd(d, tcg_env, s1, s2, s3, tcg_constant_i32(op)); + TCGv_i32 z = tcg_constant_i32(0); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_c); + gen_helper_fmaddd(d, tcg_env, s1, s2, s3, z, op); } static void gen_op_fnmsubs(TCGv_i32 d, TCGv_i32 s1, TCGv_i32 s2, TCGv_i32 s3) { - int op = float_muladd_negate_c | float_muladd_negate_result; - gen_helper_fmadds(d, tcg_env, s1, s2, s3, tcg_constant_i32(op)); + TCGv_i32 z = tcg_constant_i32(0); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_c | + float_muladd_negate_result); + gen_helper_fmadds(d, tcg_env, s1, s2, s3, z, op); } static void gen_op_fnmsubd(TCGv_i64 d, TCGv_i64 s1, TCGv_i64 s2, TCGv_i64 s3) { - int op = float_muladd_negate_c | float_muladd_negate_result; - gen_helper_fmaddd(d, tcg_env, s1, s2, s3, tcg_constant_i32(op)); + TCGv_i32 z = tcg_constant_i32(0); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_c | + float_muladd_negate_result); + gen_helper_fmaddd(d, tcg_env, s1, s2, s3, z, op); } static void gen_op_fnmadds(TCGv_i32 d, TCGv_i32 s1, TCGv_i32 s2, TCGv_i32 s3) { - int op = float_muladd_negate_result; - gen_helper_fmadds(d, tcg_env, s1, s2, s3, tcg_constant_i32(op)); + TCGv_i32 z = tcg_constant_i32(0); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_result); + gen_helper_fmadds(d, tcg_env, s1, s2, s3, z, op); } static void gen_op_fnmaddd(TCGv_i64 d, TCGv_i64 s1, TCGv_i64 s2, TCGv_i64 s3) { - int op = float_muladd_negate_result; - gen_helper_fmaddd(d, tcg_env, s1, s2, s3, tcg_constant_i32(op)); + TCGv_i32 z = tcg_constant_i32(0); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_result); + gen_helper_fmaddd(d, tcg_env, s1, s2, s3, z, op); } /* Use muladd to compute (1 * src1) + src2 / 2 with one rounding. */ static void gen_op_fhadds(TCGv_i32 d, TCGv_i32 s1, TCGv_i32 s2) { - TCGv_i32 one = tcg_constant_i32(float32_one); - int op = float_muladd_halve_result; - gen_helper_fmadds(d, tcg_env, one, s1, s2, tcg_constant_i32(op)); + TCGv_i32 fone = tcg_constant_i32(float32_one); + TCGv_i32 mone = tcg_constant_i32(-1); + TCGv_i32 op = tcg_constant_i32(0); + gen_helper_fmadds(d, tcg_env, fone, s1, s2, mone, op); } static void gen_op_fhaddd(TCGv_i64 d, TCGv_i64 s1, TCGv_i64 s2) { - TCGv_i64 one = tcg_constant_i64(float64_one); - int op = float_muladd_halve_result; - gen_helper_fmaddd(d, tcg_env, one, s1, s2, tcg_constant_i32(op)); + TCGv_i64 fone = tcg_constant_i64(float64_one); + TCGv_i32 mone = tcg_constant_i32(-1); + TCGv_i32 op = tcg_constant_i32(0); + gen_helper_fmaddd(d, tcg_env, fone, s1, s2, mone, op); } /* Use muladd to compute (1 * src1) - src2 / 2 with one rounding. */ static void gen_op_fhsubs(TCGv_i32 d, TCGv_i32 s1, TCGv_i32 s2) { - TCGv_i32 one = tcg_constant_i32(float32_one); - int op = float_muladd_negate_c | float_muladd_halve_result; - gen_helper_fmadds(d, tcg_env, one, s1, s2, tcg_constant_i32(op)); + TCGv_i32 fone = tcg_constant_i32(float32_one); + TCGv_i32 mone = tcg_constant_i32(-1); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_c); + gen_helper_fmadds(d, tcg_env, fone, s1, s2, mone, op); } static void gen_op_fhsubd(TCGv_i64 d, TCGv_i64 s1, TCGv_i64 s2) { - TCGv_i64 one = tcg_constant_i64(float64_one); - int op = float_muladd_negate_c | float_muladd_halve_result; - gen_helper_fmaddd(d, tcg_env, one, s1, s2, tcg_constant_i32(op)); + TCGv_i64 fone = tcg_constant_i64(float64_one); + TCGv_i32 mone = tcg_constant_i32(-1); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_c); + gen_helper_fmaddd(d, tcg_env, fone, s1, s2, mone, op); } /* Use muladd to compute -((1 * src1) + src2 / 2) with one rounding. */ static void gen_op_fnhadds(TCGv_i32 d, TCGv_i32 s1, TCGv_i32 s2) { - TCGv_i32 one = tcg_constant_i32(float32_one); - int op = float_muladd_negate_result | float_muladd_halve_result; - gen_helper_fmadds(d, tcg_env, one, s1, s2, tcg_constant_i32(op)); + TCGv_i32 fone = tcg_constant_i32(float32_one); + TCGv_i32 mone = tcg_constant_i32(-1); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_result); + gen_helper_fmadds(d, tcg_env, fone, s1, s2, mone, op); } static void gen_op_fnhaddd(TCGv_i64 d, TCGv_i64 s1, TCGv_i64 s2) { - TCGv_i64 one = tcg_constant_i64(float64_one); - int op = float_muladd_negate_result | float_muladd_halve_result; - gen_helper_fmaddd(d, tcg_env, one, s1, s2, tcg_constant_i32(op)); + TCGv_i64 fone = tcg_constant_i64(float64_one); + TCGv_i32 mone = tcg_constant_i32(-1); + TCGv_i32 op = tcg_constant_i32(float_muladd_negate_result); + gen_helper_fmaddd(d, tcg_env, fone, s1, s2, mone, op); } static void gen_op_fpexception_im(DisasContext *dc, int ftt) -- 2.43.0 All uses have been convered to float*_muladd_scalbn. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/fpu/softfloat.h | 3 --- fpu/softfloat.c | 6 ------ fpu/softfloat-parts.c.inc | 4 ---- 3 files changed, 13 deletions(-) diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat.h +++ b/include/fpu/softfloat.h @@ -XXX,XX +XXX,XX @@ bfloat16 bfloat16_squash_input_denormal(bfloat16 a, float_status *status); | Using these differs from negating an input or output before calling | the muladd function in that this means that a NaN doesn't have its | sign bit inverted before it is propagated. -| We also support halving the result before rounding, as a special -| case to support the ARM fused-sqrt-step instruction FRSQRTS. *----------------------------------------------------------------------------*/ enum { float_muladd_negate_c = 1, float_muladd_negate_product = 2, float_muladd_negate_result = 4, - float_muladd_halve_result = 8, }; /*---------------------------------------------------------------------------- diff --git a/fpu/softfloat.c b/fpu/softfloat.c index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -XXX,XX +XXX,XX @@ float32_muladd(float32 xa, float32 xb, float32 xc, int flags, float_status *s) if (unlikely(!can_use_fpu(s))) { goto soft; } - if (unlikely(flags & float_muladd_halve_result)) { - goto soft; - } float32_input_flush3(&ua.s, &ub.s, &uc.s, s); if (unlikely(!f32_is_zon3(ua, ub, uc))) { @@ -XXX,XX +XXX,XX @@ float64_muladd(float64 xa, float64 xb, float64 xc, int flags, float_status *s) if (unlikely(!can_use_fpu(s))) { goto soft; } - if (unlikely(flags & float_muladd_halve_result)) { - goto soft; - } float64_input_flush3(&ua.s, &ub.s, &uc.s, s); if (unlikely(!f64_is_zon3(ua, ub, uc))) { diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(muladd_scalbn)(FloatPartsN *a, FloatPartsN *b, a->exp = p_widen.exp; return_normal: - /* TODO: Replace all use of float_muladd_halve_result with scale. */ - if (flags & float_muladd_halve_result) { - a->exp -= 1; - } a->exp += scale; finish_sign: if (flags & float_muladd_negate_result) { -- 2.43.0 This rounding mode is used by Hexagon. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/fpu/softfloat-types.h | 2 ++ fpu/softfloat-parts.c.inc | 3 +++ 2 files changed, 5 insertions(+) diff --git a/include/fpu/softfloat-types.h b/include/fpu/softfloat-types.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat-types.h +++ b/include/fpu/softfloat-types.h @@ -XXX,XX +XXX,XX @@ typedef enum __attribute__((__packed__)) { float_round_to_odd = 5, /* Not an IEEE rounding mode: round to closest odd, overflow to inf */ float_round_to_odd_inf = 6, + /* Not an IEEE rounding mode: round to nearest even, overflow to max */ + float_round_nearest_even_max = 7, } FloatRoundMode; /* diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static void partsN(uncanon_normal)(FloatPartsN *p, float_status *s, int exp, flags = 0; switch (s->float_rounding_mode) { + case float_round_nearest_even_max: + overflow_norm = true; + /* fall through */ case float_round_nearest_even: if (N > 64 && frac_lsb == 0) { inc = ((p->frac_hi & 1) || (p->frac_lo & round_mask) != frac_lsbm1 -- 2.43.0 Certain Hexagon instructions suppress changes to the result when the product of fma() is a true zero. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/fpu/softfloat.h | 5 +++++ fpu/softfloat.c | 3 +++ fpu/softfloat-parts.c.inc | 4 +++- 3 files changed, 11 insertions(+), 1 deletion(-) diff --git a/include/fpu/softfloat.h b/include/fpu/softfloat.h index XXXXXXX..XXXXXXX 100644 --- a/include/fpu/softfloat.h +++ b/include/fpu/softfloat.h @@ -XXX,XX +XXX,XX @@ bfloat16 bfloat16_squash_input_denormal(bfloat16 a, float_status *status); | Using these differs from negating an input or output before calling | the muladd function in that this means that a NaN doesn't have its | sign bit inverted before it is propagated. +| +| With float_muladd_suppress_add_product_zero, if A or B is zero +| such that the product is a true zero, then return C without addition. +| This preserves the sign of C when C is +/- 0. Used for Hexagon. *----------------------------------------------------------------------------*/ enum { float_muladd_negate_c = 1, float_muladd_negate_product = 2, float_muladd_negate_result = 4, + float_muladd_suppress_add_product_zero = 8, }; /*---------------------------------------------------------------------------- diff --git a/fpu/softfloat.c b/fpu/softfloat.c index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat.c +++ b/fpu/softfloat.c @@ -XXX,XX +XXX,XX @@ float32_muladd(float32 xa, float32 xb, float32 xc, int flags, float_status *s) if (unlikely(!can_use_fpu(s))) { goto soft; } + if (unlikely(flags & float_muladd_suppress_add_product_zero)) { + goto soft; + } float32_input_flush3(&ua.s, &ub.s, &uc.s, s); if (unlikely(!f32_is_zon3(ua, ub, uc))) { diff --git a/fpu/softfloat-parts.c.inc b/fpu/softfloat-parts.c.inc index XXXXXXX..XXXXXXX 100644 --- a/fpu/softfloat-parts.c.inc +++ b/fpu/softfloat-parts.c.inc @@ -XXX,XX +XXX,XX @@ static FloatPartsN *partsN(muladd_scalbn)(FloatPartsN *a, FloatPartsN *b, goto return_normal; } if (c->cls == float_class_zero) { - if (a->sign != c->sign) { + if (flags & float_muladd_suppress_add_product_zero) { + a->sign = c->sign; + } else if (a->sign != c->sign) { goto return_sub_zero; } goto return_zero; -- 2.43.0 There are no special cases for this instruction. Remove internal_mpyf as unused. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/fma_emu.h | 1 - target/hexagon/fma_emu.c | 8 -------- target/hexagon/op_helper.c | 2 +- 3 files changed, 1 insertion(+), 10 deletions(-) diff --git a/target/hexagon/fma_emu.h b/target/hexagon/fma_emu.h index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.h +++ b/target/hexagon/fma_emu.h @@ -XXX,XX +XXX,XX @@ int32_t float32_getexp(float32 f32); float32 infinite_float32(uint8_t sign); float32 internal_fmafx(float32 a, float32 b, float32 c, int scale, float_status *fp_status); -float32 internal_mpyf(float32 a, float32 b, float_status *fp_status); float64 internal_mpyhh(float64 a, float64 b, unsigned long long int accumulated, float_status *fp_status); diff --git a/target/hexagon/fma_emu.c b/target/hexagon/fma_emu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.c +++ b/target/hexagon/fma_emu.c @@ -XXX,XX +XXX,XX @@ float32 internal_fmafx(float32 a, float32 b, float32 c, int scale, return accum_round_float32(result, fp_status); } -float32 internal_mpyf(float32 a, float32 b, float_status *fp_status) -{ - if (float32_is_zero(a) || float32_is_zero(b)) { - return float32_mul(a, b, fp_status); - } - return internal_fmafx(a, b, float32_zero, 0, fp_status); -} - float64 internal_mpyhh(float64 a, float64 b, unsigned long long int accumulated, float_status *fp_status) diff --git a/target/hexagon/op_helper.c b/target/hexagon/op_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/op_helper.c +++ b/target/hexagon/op_helper.c @@ -XXX,XX +XXX,XX @@ float32 HELPER(sfmpy)(CPUHexagonState *env, float32 RsV, float32 RtV) { float32 RdV; arch_fpop_start(env); - RdV = internal_mpyf(RsV, RtV, &env->fp_status); + RdV = float32_mul(RsV, RtV, &env->fp_status); arch_fpop_end(env); return RdV; } -- 2.43.0 There are no special cases for this instruction. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/op_helper.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/target/hexagon/op_helper.c b/target/hexagon/op_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/op_helper.c +++ b/target/hexagon/op_helper.c @@ -XXX,XX +XXX,XX @@ float32 HELPER(sffma)(CPUHexagonState *env, float32 RxV, float32 RsV, float32 RtV) { arch_fpop_start(env); - RxV = internal_fmafx(RsV, RtV, RxV, 0, &env->fp_status); + RxV = float32_muladd(RsV, RtV, RxV, 0, &env->fp_status); arch_fpop_end(env); return RxV; } -- 2.43.0 There are no special cases for this instruction. Since hexagon always uses default-nan mode, explicitly negating the first input is unnecessary. Use float_muladd_negate_product instead. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/op_helper.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-) diff --git a/target/hexagon/op_helper.c b/target/hexagon/op_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/op_helper.c +++ b/target/hexagon/op_helper.c @@ -XXX,XX +XXX,XX @@ float32 HELPER(sffma_sc)(CPUHexagonState *env, float32 RxV, float32 HELPER(sffms)(CPUHexagonState *env, float32 RxV, float32 RsV, float32 RtV) { - float32 neg_RsV; arch_fpop_start(env); - neg_RsV = float32_set_sign(RsV, float32_is_neg(RsV) ? 0 : 1); - RxV = internal_fmafx(neg_RsV, RtV, RxV, 0, &env->fp_status); + RxV = float32_muladd(RsV, RtV, RxV, float_muladd_negate_product, + &env->fp_status); arch_fpop_end(env); return RxV; } -- 2.43.0 This instruction has a special case that 0 * x + c returns c without the normal sign folding that comes with 0 + -0. Use the new float_muladd_suppress_add_product_zero to describe this. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/op_helper.c | 11 +++-------- 1 file changed, 3 insertions(+), 8 deletions(-) diff --git a/target/hexagon/op_helper.c b/target/hexagon/op_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/op_helper.c +++ b/target/hexagon/op_helper.c @@ -XXX,XX +XXX,XX @@ static float32 check_nan(float32 dst, float32 x, float_status *fp_status) float32 HELPER(sffma_sc)(CPUHexagonState *env, float32 RxV, float32 RsV, float32 RtV, float32 PuV) { - size4s_t tmp; arch_fpop_start(env); - RxV = check_nan(RxV, RxV, &env->fp_status); - RxV = check_nan(RxV, RsV, &env->fp_status); - RxV = check_nan(RxV, RtV, &env->fp_status); - tmp = internal_fmafx(RsV, RtV, RxV, fSXTN(8, 64, PuV), &env->fp_status); - if (!(float32_is_zero(RxV) && is_zero_prod(RsV, RtV))) { - RxV = tmp; - } + RxV = float32_muladd_scalbn(RsV, RtV, RxV, fSXTN(8, 64, PuV), + float_muladd_suppress_add_product_zero, + &env->fp_status); arch_fpop_end(env); return RxV; } -- 2.43.0 There are multiple special cases for this instruction. (1) The saturate to normal maximum instead of overflow to infinity is handled by the new float_round_nearest_even_max rounding mode. (2) The 0 * n + c special case is handled by the new float_muladd_suppress_add_product_zero flag. (3) The Inf - Inf -> 0 special case can be detected after the fact by examining float_flag_invalid_isi. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/op_helper.c | 105 +++++++++---------------------------- 1 file changed, 26 insertions(+), 79 deletions(-) diff --git a/target/hexagon/op_helper.c b/target/hexagon/op_helper.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/op_helper.c +++ b/target/hexagon/op_helper.c @@ -XXX,XX +XXX,XX @@ float32 HELPER(sffma)(CPUHexagonState *env, float32 RxV, return RxV; } -static bool is_zero_prod(float32 a, float32 b) -{ - return ((float32_is_zero(a) && is_finite(b)) || - (float32_is_zero(b) && is_finite(a))); -} - -static float32 check_nan(float32 dst, float32 x, float_status *fp_status) -{ - float32 ret = dst; - if (float32_is_any_nan(x)) { - if (extract32(x, 22, 1) == 0) { - float_raise(float_flag_invalid, fp_status); - } - ret = make_float32(0xffffffff); /* nan */ - } - return ret; -} - float32 HELPER(sffma_sc)(CPUHexagonState *env, float32 RxV, float32 RsV, float32 RtV, float32 PuV) { @@ -XXX,XX +XXX,XX @@ float32 HELPER(sffms)(CPUHexagonState *env, float32 RxV, return RxV; } -static bool is_inf_prod(int32_t a, int32_t b) +static float32 do_sffma_lib(CPUHexagonState *env, float32 RxV, + float32 RsV, float32 RtV, int negate) { - return (float32_is_infinity(a) && float32_is_infinity(b)) || - (float32_is_infinity(a) && is_finite(b) && !float32_is_zero(b)) || - (float32_is_infinity(b) && is_finite(a) && !float32_is_zero(a)); + int flags; + + arch_fpop_start(env); + + set_float_rounding_mode(float_round_nearest_even_max, &env->fp_status); + RxV = float32_muladd(RsV, RtV, RxV, + negate | float_muladd_suppress_add_product_zero, + &env->fp_status); + + flags = get_float_exception_flags(&env->fp_status); + if (flags) { + /* Flags are suppressed by this instruction. */ + set_float_exception_flags(0, &env->fp_status); + + /* Return 0 for Inf - Inf. */ + if (flags & float_flag_invalid_isi) { + RxV = 0; + } + } + + arch_fpop_end(env); + return RxV; } float32 HELPER(sffma_lib)(CPUHexagonState *env, float32 RxV, float32 RsV, float32 RtV) { - bool infinp; - bool infminusinf; - float32 tmp; - - arch_fpop_start(env); - set_float_rounding_mode(float_round_nearest_even, &env->fp_status); - infminusinf = float32_is_infinity(RxV) && - is_inf_prod(RsV, RtV) && - (fGETBIT(31, RsV ^ RxV ^ RtV) != 0); - infinp = float32_is_infinity(RxV) || - float32_is_infinity(RtV) || - float32_is_infinity(RsV); - RxV = check_nan(RxV, RxV, &env->fp_status); - RxV = check_nan(RxV, RsV, &env->fp_status); - RxV = check_nan(RxV, RtV, &env->fp_status); - tmp = internal_fmafx(RsV, RtV, RxV, 0, &env->fp_status); - if (!(float32_is_zero(RxV) && is_zero_prod(RsV, RtV))) { - RxV = tmp; - } - set_float_exception_flags(0, &env->fp_status); - if (float32_is_infinity(RxV) && !infinp) { - RxV = RxV - 1; - } - if (infminusinf) { - RxV = 0; - } - arch_fpop_end(env); - return RxV; + return do_sffma_lib(env, RxV, RsV, RtV, 0); } float32 HELPER(sffms_lib)(CPUHexagonState *env, float32 RxV, float32 RsV, float32 RtV) { - bool infinp; - bool infminusinf; - float32 tmp; - - arch_fpop_start(env); - set_float_rounding_mode(float_round_nearest_even, &env->fp_status); - infminusinf = float32_is_infinity(RxV) && - is_inf_prod(RsV, RtV) && - (fGETBIT(31, RsV ^ RxV ^ RtV) == 0); - infinp = float32_is_infinity(RxV) || - float32_is_infinity(RtV) || - float32_is_infinity(RsV); - RxV = check_nan(RxV, RxV, &env->fp_status); - RxV = check_nan(RxV, RsV, &env->fp_status); - RxV = check_nan(RxV, RtV, &env->fp_status); - float32 minus_RsV = float32_sub(float32_zero, RsV, &env->fp_status); - tmp = internal_fmafx(minus_RsV, RtV, RxV, 0, &env->fp_status); - if (!(float32_is_zero(RxV) && is_zero_prod(RsV, RtV))) { - RxV = tmp; - } - set_float_exception_flags(0, &env->fp_status); - if (float32_is_infinity(RxV) && !infinp) { - RxV = RxV - 1; - } - if (infminusinf) { - RxV = 0; - } - arch_fpop_end(env); - return RxV; + return do_sffma_lib(env, RxV, RsV, RtV, float_muladd_negate_product); } float64 HELPER(dfmpyfix)(CPUHexagonState *env, float64 RssV, float64 RttV) -- 2.43.0 The function is now unused. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/fma_emu.h | 2 - target/hexagon/fma_emu.c | 171 --------------------------------------- 2 files changed, 173 deletions(-) diff --git a/target/hexagon/fma_emu.h b/target/hexagon/fma_emu.h index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.h +++ b/target/hexagon/fma_emu.h @@ -XXX,XX +XXX,XX @@ static inline uint32_t float32_getexp_raw(float32 f32) } int32_t float32_getexp(float32 f32); float32 infinite_float32(uint8_t sign); -float32 internal_fmafx(float32 a, float32 b, float32 c, - int scale, float_status *fp_status); float64 internal_mpyhh(float64 a, float64 b, unsigned long long int accumulated, float_status *fp_status); diff --git a/target/hexagon/fma_emu.c b/target/hexagon/fma_emu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.c +++ b/target/hexagon/fma_emu.c @@ -XXX,XX +XXX,XX @@ int32_t float64_getexp(float64 f64) return -1; } -static uint64_t float32_getmant(float32 f32) -{ - Float a = { .i = f32 }; - if (float32_is_normal(f32)) { - return a.mant | 1ULL << 23; - } - if (float32_is_zero(f32)) { - return 0; - } - if (float32_is_denormal(f32)) { - return a.mant; - } - return ~0ULL; -} - int32_t float32_getexp(float32 f32) { Float a = { .i = f32 }; @@ -XXX,XX +XXX,XX @@ float32 infinite_float32(uint8_t sign) } /* Return a maximum finite value with the requested sign */ -static float32 maxfinite_float32(uint8_t sign) -{ - if (sign) { - return make_float32(SF_MINUS_MAXF); - } else { - return make_float32(SF_MAXF); - } -} - -/* Return a zero value with requested sign */ -static float32 zero_float32(uint8_t sign) -{ - if (sign) { - return make_float32(0x80000000); - } else { - return float32_zero; - } -} - #define GEN_XF_ROUND(SUFFIX, MANTBITS, INF_EXP, INTERNAL_TYPE) \ static SUFFIX accum_round_##SUFFIX(Accum a, float_status * fp_status) \ { \ @@ -XXX,XX +XXX,XX @@ static SUFFIX accum_round_##SUFFIX(Accum a, float_status * fp_status) \ } GEN_XF_ROUND(float64, DF_MANTBITS, DF_INF_EXP, Double) -GEN_XF_ROUND(float32, SF_MANTBITS, SF_INF_EXP, Float) - -static bool is_inf_prod(float64 a, float64 b) -{ - return ((float64_is_infinity(a) && float64_is_infinity(b)) || - (float64_is_infinity(a) && is_finite(b) && (!float64_is_zero(b))) || - (float64_is_infinity(b) && is_finite(a) && (!float64_is_zero(a)))); -} - -static float64 special_fma(float64 a, float64 b, float64 c, - float_status *fp_status) -{ - float64 ret = make_float64(0); - - /* - * If A multiplied by B is an exact infinity and C is also an infinity - * but with the opposite sign, FMA returns NaN and raises invalid. - */ - uint8_t a_sign = float64_is_neg(a); - uint8_t b_sign = float64_is_neg(b); - uint8_t c_sign = float64_is_neg(c); - if (is_inf_prod(a, b) && float64_is_infinity(c)) { - if ((a_sign ^ b_sign) != c_sign) { - ret = make_float64(DF_NAN); - float_raise(float_flag_invalid, fp_status); - return ret; - } - } - if ((float64_is_infinity(a) && float64_is_zero(b)) || - (float64_is_zero(a) && float64_is_infinity(b))) { - ret = make_float64(DF_NAN); - float_raise(float_flag_invalid, fp_status); - return ret; - } - /* - * If none of the above checks are true and C is a NaN, - * a NaN shall be returned - * If A or B are NaN, a NAN shall be returned. - */ - if (float64_is_any_nan(a) || - float64_is_any_nan(b) || - float64_is_any_nan(c)) { - if (float64_is_any_nan(a) && (fGETBIT(51, a) == 0)) { - float_raise(float_flag_invalid, fp_status); - } - if (float64_is_any_nan(b) && (fGETBIT(51, b) == 0)) { - float_raise(float_flag_invalid, fp_status); - } - if (float64_is_any_nan(c) && (fGETBIT(51, c) == 0)) { - float_raise(float_flag_invalid, fp_status); - } - ret = make_float64(DF_NAN); - return ret; - } - /* - * We have checked for adding opposite-signed infinities. - * Other infinities return infinity with the correct sign - */ - if (float64_is_infinity(c)) { - ret = infinite_float64(c_sign); - return ret; - } - if (float64_is_infinity(a) || float64_is_infinity(b)) { - ret = infinite_float64(a_sign ^ b_sign); - return ret; - } - g_assert_not_reached(); -} - -static float32 special_fmaf(float32 a, float32 b, float32 c, - float_status *fp_status) -{ - float64 aa, bb, cc; - aa = float32_to_float64(a, fp_status); - bb = float32_to_float64(b, fp_status); - cc = float32_to_float64(c, fp_status); - return float64_to_float32(special_fma(aa, bb, cc, fp_status), fp_status); -} - -float32 internal_fmafx(float32 a, float32 b, float32 c, int scale, - float_status *fp_status) -{ - Accum prod; - Accum acc; - Accum result; - accum_init(&prod); - accum_init(&acc); - accum_init(&result); - - uint8_t a_sign = float32_is_neg(a); - uint8_t b_sign = float32_is_neg(b); - uint8_t c_sign = float32_is_neg(c); - if (float32_is_infinity(a) || - float32_is_infinity(b) || - float32_is_infinity(c)) { - return special_fmaf(a, b, c, fp_status); - } - if (float32_is_any_nan(a) || - float32_is_any_nan(b) || - float32_is_any_nan(c)) { - return special_fmaf(a, b, c, fp_status); - } - if ((scale == 0) && (float32_is_zero(a) || float32_is_zero(b))) { - float32 tmp = float32_mul(a, b, fp_status); - tmp = float32_add(tmp, c, fp_status); - return tmp; - } - - /* (a * 2**b) * (c * 2**d) == a*c * 2**(b+d) */ - prod.mant = int128_mul_6464(float32_getmant(a), float32_getmant(b)); - - /* - * Note: extracting the mantissa into an int is multiplying by - * 2**23, so adjust here - */ - prod.exp = float32_getexp(a) + float32_getexp(b) - SF_BIAS - 23; - prod.sign = a_sign ^ b_sign; - if (float32_is_zero(a) || float32_is_zero(b)) { - prod.exp = -2 * WAY_BIG_EXP; - } - if ((scale > 0) && float32_is_denormal(c)) { - acc.mant = int128_mul_6464(0, 0); - acc.exp = -WAY_BIG_EXP; - acc.sign = c_sign; - acc.sticky = 1; - result = accum_add(prod, acc); - } else if (!float32_is_zero(c)) { - acc.mant = int128_mul_6464(float32_getmant(c), 1); - acc.exp = float32_getexp(c); - acc.sign = c_sign; - result = accum_add(prod, acc); - } else { - result = prod; - } - result.exp += scale; - return accum_round_float32(result, fp_status); -} float64 internal_mpyhh(float64 a, float64 b, unsigned long long int accumulated, -- 2.43.0 This massive macro is now only used once. Expand it for use only by float64. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/fma_emu.c | 255 +++++++++++++++++++-------------------- 1 file changed, 127 insertions(+), 128 deletions(-) diff --git a/target/hexagon/fma_emu.c b/target/hexagon/fma_emu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.c +++ b/target/hexagon/fma_emu.c @@ -XXX,XX +XXX,XX @@ float32 infinite_float32(uint8_t sign) } /* Return a maximum finite value with the requested sign */ -#define GEN_XF_ROUND(SUFFIX, MANTBITS, INF_EXP, INTERNAL_TYPE) \ -static SUFFIX accum_round_##SUFFIX(Accum a, float_status * fp_status) \ -{ \ - if ((int128_gethi(a.mant) == 0) && (int128_getlo(a.mant) == 0) \ - && ((a.guard | a.round | a.sticky) == 0)) { \ - /* result zero */ \ - switch (fp_status->float_rounding_mode) { \ - case float_round_down: \ - return zero_##SUFFIX(1); \ - default: \ - return zero_##SUFFIX(0); \ - } \ - } \ - /* Normalize right */ \ - /* We want MANTBITS bits of mantissa plus the leading one. */ \ - /* That means that we want MANTBITS+1 bits, or 0x000000000000FF_FFFF */ \ - /* So we need to normalize right while the high word is non-zero and \ - * while the low word is nonzero when masked with 0xffe0_0000_0000_0000 */ \ - while ((int128_gethi(a.mant) != 0) || \ - ((int128_getlo(a.mant) >> (MANTBITS + 1)) != 0)) { \ - a = accum_norm_right(a, 1); \ - } \ - /* \ - * OK, now normalize left \ - * We want to normalize left until we have a leading one in bit 24 \ - * Theoretically, we only need to shift a maximum of one to the left if we \ - * shifted out lots of bits from B, or if we had no shift / 1 shift sticky \ - * should be 0 \ - */ \ - while ((int128_getlo(a.mant) & (1ULL << MANTBITS)) == 0) { \ - a = accum_norm_left(a); \ - } \ - /* \ - * OK, now we might need to denormalize because of potential underflow. \ - * We need to do this before rounding, and rounding might make us normal \ - * again \ - */ \ - while (a.exp <= 0) { \ - a = accum_norm_right(a, 1 - a.exp); \ - /* \ - * Do we have underflow? \ - * That's when we get an inexact answer because we ran out of bits \ - * in a denormal. \ - */ \ - if (a.guard || a.round || a.sticky) { \ - float_raise(float_flag_underflow, fp_status); \ - } \ - } \ - /* OK, we're relatively canonical... now we need to round */ \ - if (a.guard || a.round || a.sticky) { \ - float_raise(float_flag_inexact, fp_status); \ - switch (fp_status->float_rounding_mode) { \ - case float_round_to_zero: \ - /* Chop and we're done */ \ - break; \ - case float_round_up: \ - if (a.sign == 0) { \ - a.mant = int128_add(a.mant, int128_one()); \ - } \ - break; \ - case float_round_down: \ - if (a.sign != 0) { \ - a.mant = int128_add(a.mant, int128_one()); \ - } \ - break; \ - default: \ - if (a.round || a.sticky) { \ - /* round up if guard is 1, down if guard is zero */ \ - a.mant = int128_add(a.mant, int128_make64(a.guard)); \ - } else if (a.guard) { \ - /* exactly .5, round up if odd */ \ - a.mant = int128_add(a.mant, int128_and(a.mant, int128_one())); \ - } \ - break; \ - } \ - } \ - /* \ - * OK, now we might have carried all the way up. \ - * So we might need to shr once \ - * at least we know that the lsb should be zero if we rounded and \ - * got a carry out... \ - */ \ - if ((int128_getlo(a.mant) >> (MANTBITS + 1)) != 0) { \ - a = accum_norm_right(a, 1); \ - } \ - /* Overflow? */ \ - if (a.exp >= INF_EXP) { \ - /* Yep, inf result */ \ - float_raise(float_flag_overflow, fp_status); \ - float_raise(float_flag_inexact, fp_status); \ - switch (fp_status->float_rounding_mode) { \ - case float_round_to_zero: \ - return maxfinite_##SUFFIX(a.sign); \ - case float_round_up: \ - if (a.sign == 0) { \ - return infinite_##SUFFIX(a.sign); \ - } else { \ - return maxfinite_##SUFFIX(a.sign); \ - } \ - case float_round_down: \ - if (a.sign != 0) { \ - return infinite_##SUFFIX(a.sign); \ - } else { \ - return maxfinite_##SUFFIX(a.sign); \ - } \ - default: \ - return infinite_##SUFFIX(a.sign); \ - } \ - } \ - /* Underflow? */ \ - if (int128_getlo(a.mant) & (1ULL << MANTBITS)) { \ - /* Leading one means: No, we're normal. So, we should be done... */ \ - INTERNAL_TYPE ret; \ - ret.i = 0; \ - ret.sign = a.sign; \ - ret.exp = a.exp; \ - ret.mant = int128_getlo(a.mant); \ - return ret.i; \ - } \ - assert(a.exp == 1); \ - INTERNAL_TYPE ret; \ - ret.i = 0; \ - ret.sign = a.sign; \ - ret.exp = 0; \ - ret.mant = int128_getlo(a.mant); \ - return ret.i; \ +static float64 accum_round_float64(Accum a, float_status *fp_status) +{ + if ((int128_gethi(a.mant) == 0) && (int128_getlo(a.mant) == 0) + && ((a.guard | a.round | a.sticky) == 0)) { + /* result zero */ + switch (fp_status->float_rounding_mode) { + case float_round_down: + return zero_float64(1); + default: + return zero_float64(0); + } + } + /* + * Normalize right + * We want DF_MANTBITS bits of mantissa plus the leading one. + * That means that we want DF_MANTBITS+1 bits, or 0x000000000000FF_FFFF + * So we need to normalize right while the high word is non-zero and + * while the low word is nonzero when masked with 0xffe0_0000_0000_0000 + */ + while ((int128_gethi(a.mant) != 0) || + ((int128_getlo(a.mant) >> (DF_MANTBITS + 1)) != 0)) { + a = accum_norm_right(a, 1); + } + /* + * OK, now normalize left + * We want to normalize left until we have a leading one in bit 24 + * Theoretically, we only need to shift a maximum of one to the left if we + * shifted out lots of bits from B, or if we had no shift / 1 shift sticky + * should be 0 + */ + while ((int128_getlo(a.mant) & (1ULL << DF_MANTBITS)) == 0) { + a = accum_norm_left(a); + } + /* + * OK, now we might need to denormalize because of potential underflow. + * We need to do this before rounding, and rounding might make us normal + * again + */ + while (a.exp <= 0) { + a = accum_norm_right(a, 1 - a.exp); + /* + * Do we have underflow? + * That's when we get an inexact answer because we ran out of bits + * in a denormal. + */ + if (a.guard || a.round || a.sticky) { + float_raise(float_flag_underflow, fp_status); + } + } + /* OK, we're relatively canonical... now we need to round */ + if (a.guard || a.round || a.sticky) { + float_raise(float_flag_inexact, fp_status); + switch (fp_status->float_rounding_mode) { + case float_round_to_zero: + /* Chop and we're done */ + break; + case float_round_up: + if (a.sign == 0) { + a.mant = int128_add(a.mant, int128_one()); + } + break; + case float_round_down: + if (a.sign != 0) { + a.mant = int128_add(a.mant, int128_one()); + } + break; + default: + if (a.round || a.sticky) { + /* round up if guard is 1, down if guard is zero */ + a.mant = int128_add(a.mant, int128_make64(a.guard)); + } else if (a.guard) { + /* exactly .5, round up if odd */ + a.mant = int128_add(a.mant, int128_and(a.mant, int128_one())); + } + break; + } + } + /* + * OK, now we might have carried all the way up. + * So we might need to shr once + * at least we know that the lsb should be zero if we rounded and + * got a carry out... + */ + if ((int128_getlo(a.mant) >> (DF_MANTBITS + 1)) != 0) { + a = accum_norm_right(a, 1); + } + /* Overflow? */ + if (a.exp >= DF_INF_EXP) { + /* Yep, inf result */ + float_raise(float_flag_overflow, fp_status); + float_raise(float_flag_inexact, fp_status); + switch (fp_status->float_rounding_mode) { + case float_round_to_zero: + return maxfinite_float64(a.sign); + case float_round_up: + if (a.sign == 0) { + return infinite_float64(a.sign); + } else { + return maxfinite_float64(a.sign); + } + case float_round_down: + if (a.sign != 0) { + return infinite_float64(a.sign); + } else { + return maxfinite_float64(a.sign); + } + default: + return infinite_float64(a.sign); + } + } + /* Underflow? */ + if (int128_getlo(a.mant) & (1ULL << DF_MANTBITS)) { + /* Leading one means: No, we're normal. So, we should be done... */ + Double ret; + ret.i = 0; + ret.sign = a.sign; + ret.exp = a.exp; + ret.mant = int128_getlo(a.mant); + return ret.i; + } + assert(a.exp == 1); + Double ret; + ret.i = 0; + ret.sign = a.sign; + ret.exp = 0; + ret.mant = int128_getlo(a.mant); + return ret.i; } -GEN_XF_ROUND(float64, DF_MANTBITS, DF_INF_EXP, Double) - float64 internal_mpyhh(float64 a, float64 b, unsigned long long int accumulated, float_status *fp_status) -- 2.43.0 This structure, with bitfields, is incorrect for big-endian. Use the existing float32_getexp_raw which uses extract32. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/fma_emu.c | 16 +++------------- 1 file changed, 3 insertions(+), 13 deletions(-) diff --git a/target/hexagon/fma_emu.c b/target/hexagon/fma_emu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.c +++ b/target/hexagon/fma_emu.c @@ -XXX,XX +XXX,XX @@ typedef union { }; } Double; -typedef union { - float f; - uint32_t i; - struct { - uint32_t mant:23; - uint32_t exp:8; - uint32_t sign:1; - }; -} Float; - static uint64_t float64_getmant(float64 f64) { Double a = { .i = f64 }; @@ -XXX,XX +XXX,XX @@ int32_t float64_getexp(float64 f64) int32_t float32_getexp(float32 f32) { - Float a = { .i = f32 }; + int exp = float32_getexp_raw(f32); if (float32_is_normal(f32)) { - return a.exp; + return exp; } if (float32_is_denormal(f32)) { - return a.exp + 1; + return exp + 1; } return -1; } -- 2.43.0 This structure, with bitfields, is incorrect for big-endian. Use extract64 and deposit64 instead. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/fma_emu.c | 46 ++++++++++++++-------------------------- 1 file changed, 16 insertions(+), 30 deletions(-) diff --git a/target/hexagon/fma_emu.c b/target/hexagon/fma_emu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.c +++ b/target/hexagon/fma_emu.c @@ -XXX,XX +XXX,XX @@ #define WAY_BIG_EXP 4096 -typedef union { - double f; - uint64_t i; - struct { - uint64_t mant:52; - uint64_t exp:11; - uint64_t sign:1; - }; -} Double; - static uint64_t float64_getmant(float64 f64) { - Double a = { .i = f64 }; + uint64_t mant = extract64(f64, 0, 52); if (float64_is_normal(f64)) { - return a.mant | 1ULL << 52; + return mant | 1ULL << 52; } if (float64_is_zero(f64)) { return 0; } if (float64_is_denormal(f64)) { - return a.mant; + return mant; } return ~0ULL; } int32_t float64_getexp(float64 f64) { - Double a = { .i = f64 }; + int exp = extract64(f64, 52, 11); if (float64_is_normal(f64)) { - return a.exp; + return exp; } if (float64_is_denormal(f64)) { - return a.exp + 1; + return exp + 1; } return -1; } @@ -XXX,XX +XXX,XX @@ float32 infinite_float32(uint8_t sign) /* Return a maximum finite value with the requested sign */ static float64 accum_round_float64(Accum a, float_status *fp_status) { + uint64_t ret; + if ((int128_gethi(a.mant) == 0) && (int128_getlo(a.mant) == 0) && ((a.guard | a.round | a.sticky) == 0)) { /* result zero */ @@ -XXX,XX +XXX,XX @@ static float64 accum_round_float64(Accum a, float_status *fp_status) } } /* Underflow? */ - if (int128_getlo(a.mant) & (1ULL << DF_MANTBITS)) { + ret = int128_getlo(a.mant); + if (ret & (1ULL << DF_MANTBITS)) { /* Leading one means: No, we're normal. So, we should be done... */ - Double ret; - ret.i = 0; - ret.sign = a.sign; - ret.exp = a.exp; - ret.mant = int128_getlo(a.mant); - return ret.i; + ret = deposit64(ret, 52, 11, a.exp); + } else { + assert(a.exp == 1); + ret = deposit64(ret, 52, 11, 0); } - assert(a.exp == 1); - Double ret; - ret.i = 0; - ret.sign = a.sign; - ret.exp = 0; - ret.mant = int128_getlo(a.mant); - return ret.i; + ret = deposit64(ret, 63, 1, a.sign); + return ret; } float64 internal_mpyhh(float64 a, float64 b, -- 2.43.0 No need to open-code 64x64->128-bit multiplication. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/fma_emu.c | 32 +++----------------------------- 1 file changed, 3 insertions(+), 29 deletions(-) diff --git a/target/hexagon/fma_emu.c b/target/hexagon/fma_emu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.c +++ b/target/hexagon/fma_emu.c @@ -XXX,XX +XXX,XX @@ int32_t float32_getexp(float32 f32) return -1; } -static uint32_t int128_getw0(Int128 x) -{ - return int128_getlo(x); -} - -static uint32_t int128_getw1(Int128 x) -{ - return int128_getlo(x) >> 32; -} - static Int128 int128_mul_6464(uint64_t ai, uint64_t bi) { - Int128 a, b; - uint64_t pp0, pp1a, pp1b, pp1s, pp2; + uint64_t l, h; - a = int128_make64(ai); - b = int128_make64(bi); - pp0 = (uint64_t)int128_getw0(a) * (uint64_t)int128_getw0(b); - pp1a = (uint64_t)int128_getw1(a) * (uint64_t)int128_getw0(b); - pp1b = (uint64_t)int128_getw1(b) * (uint64_t)int128_getw0(a); - pp2 = (uint64_t)int128_getw1(a) * (uint64_t)int128_getw1(b); - - pp1s = pp1a + pp1b; - if ((pp1s < pp1a) || (pp1s < pp1b)) { - pp2 += (1ULL << 32); - } - uint64_t ret_low = pp0 + (pp1s << 32); - if ((ret_low < pp0) || (ret_low < (pp1s << 32))) { - pp2 += 1; - } - - return int128_make128(ret_low, pp2 + (pp1s >> 32)); + mulu64(&l, &h, ai, bi); + return int128_make128(l, h); } static Int128 int128_sub_borrow(Int128 a, Int128 b, int borrow) -- 2.43.0 Initialize x with accumulated via direct assignment, rather than multiplying by 1. Reviewed-by: Brian Cain <brian.cain@oss.qualcomm.com> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/hexagon/fma_emu.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/target/hexagon/fma_emu.c b/target/hexagon/fma_emu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/fma_emu.c +++ b/target/hexagon/fma_emu.c @@ -XXX,XX +XXX,XX @@ float64 internal_mpyhh(float64 a, float64 b, float64_is_infinity(b)) { return float64_mul(a, b, fp_status); } - x.mant = int128_mul_6464(accumulated, 1); + x.mant = int128_make64(accumulated); x.sticky = sticky; prod = fGETUWORD(1, float64_getmant(a)) * fGETUWORD(1, float64_getmant(b)); x.mant = int128_add(x.mant, int128_mul_6464(prod, 0x100000000ULL)); -- 2.43.0 Convert all targets simultaneously, as the gen_intermediate_code function disappears from the target. While there are possible workarounds, they're larger than simply performing the conversion. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/exec/translator.h | 14 -------------- include/hw/core/tcg-cpu-ops.h | 13 +++++++++++++ target/alpha/cpu.h | 2 ++ target/arm/internals.h | 2 ++ target/avr/cpu.h | 2 ++ target/hexagon/cpu.h | 2 ++ target/hppa/cpu.h | 2 ++ target/i386/tcg/helper-tcg.h | 2 ++ target/loongarch/internals.h | 2 ++ target/m68k/cpu.h | 2 ++ target/microblaze/cpu.h | 2 ++ target/mips/tcg/tcg-internal.h | 2 ++ target/openrisc/cpu.h | 2 ++ target/ppc/cpu.h | 2 ++ target/riscv/cpu.h | 3 +++ target/rx/cpu.h | 2 ++ target/s390x/s390x-internal.h | 2 ++ target/sh4/cpu.h | 2 ++ target/sparc/cpu.h | 2 ++ target/tricore/cpu.h | 2 ++ target/xtensa/cpu.h | 2 ++ accel/tcg/cpu-exec.c | 8 +++++--- accel/tcg/translate-all.c | 8 +++++--- target/alpha/cpu.c | 1 + target/alpha/translate.c | 4 ++-- target/arm/cpu.c | 1 + target/arm/tcg/cpu-v7m.c | 1 + target/arm/tcg/translate.c | 5 ++--- target/avr/cpu.c | 1 + target/avr/translate.c | 6 +++--- target/hexagon/cpu.c | 1 + target/hexagon/translate.c | 4 ++-- target/hppa/cpu.c | 1 + target/hppa/translate.c | 4 ++-- target/i386/tcg/tcg-cpu.c | 1 + target/i386/tcg/translate.c | 5 ++--- target/loongarch/cpu.c | 1 + target/loongarch/tcg/translate.c | 4 ++-- target/m68k/cpu.c | 1 + target/m68k/translate.c | 4 ++-- target/microblaze/cpu.c | 1 + target/microblaze/translate.c | 4 ++-- target/mips/cpu.c | 1 + target/mips/tcg/translate.c | 4 ++-- target/openrisc/cpu.c | 1 + target/openrisc/translate.c | 4 ++-- target/ppc/cpu_init.c | 1 + target/ppc/translate.c | 4 ++-- target/riscv/tcg/tcg-cpu.c | 1 + target/riscv/translate.c | 4 ++-- target/rx/cpu.c | 1 + target/rx/translate.c | 4 ++-- target/s390x/cpu.c | 1 + target/s390x/tcg/translate.c | 4 ++-- target/sh4/cpu.c | 1 + target/sh4/translate.c | 4 ++-- target/sparc/cpu.c | 1 + target/sparc/translate.c | 4 ++-- target/tricore/cpu.c | 1 + target/tricore/translate.c | 5 ++--- target/xtensa/cpu.c | 1 + target/xtensa/translate.c | 4 ++-- 62 files changed, 121 insertions(+), 62 deletions(-) diff --git a/include/exec/translator.h b/include/exec/translator.h index XXXXXXX..XXXXXXX 100644 --- a/include/exec/translator.h +++ b/include/exec/translator.h @@ -XXX,XX +XXX,XX @@ #include "qemu/bswap.h" #include "exec/vaddr.h" -/** - * gen_intermediate_code - * @cpu: cpu context - * @tb: translation block - * @max_insns: max number of instructions to translate - * @pc: guest virtual program counter address - * @host_pc: host physical program counter address - * - * This function must be provided by the target, which should create - * the target-specific DisasContext, and then invoke translator_loop. - */ -void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc); - /** * DisasJumpType: * @DISAS_NEXT: Next instruction in program order. diff --git a/include/hw/core/tcg-cpu-ops.h b/include/hw/core/tcg-cpu-ops.h index XXXXXXX..XXXXXXX 100644 --- a/include/hw/core/tcg-cpu-ops.h +++ b/include/hw/core/tcg-cpu-ops.h @@ -XXX,XX +XXX,XX @@ struct TCGCPUOps { * Called when the first CPU is realized. */ void (*initialize)(void); + /** + * @translate_code: Translate guest instructions to TCGOps + * @cpu: cpu context + * @tb: translation block + * @max_insns: max number of instructions to translate + * @pc: guest virtual program counter address + * @host_pc: host physical program counter address + * + * This function must be provided by the target, which should create + * the target-specific DisasContext, and then invoke translator_loop. + */ + void (*translate_code)(CPUState *cpu, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); /** * @synchronize_from_tb: Synchronize state from a TCG #TranslationBlock * diff --git a/target/alpha/cpu.h b/target/alpha/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/alpha/cpu.h +++ b/target/alpha/cpu.h @@ -XXX,XX +XXX,XX @@ enum { }; void alpha_translate_init(void); +void alpha_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); #define CPU_RESOLVING_TYPE TYPE_ALPHA_CPU diff --git a/target/arm/internals.h b/target/arm/internals.h index XXXXXXX..XXXXXXX 100644 --- a/target/arm/internals.h +++ b/target/arm/internals.h @@ -XXX,XX +XXX,XX @@ void init_cpreg_list(ARMCPU *cpu); void arm_cpu_register_gdb_regs_for_features(ARMCPU *cpu); void arm_translate_init(void); +void arm_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); void arm_cpu_register_gdb_commands(ARMCPU *cpu); void aarch64_cpu_register_gdb_commands(ARMCPU *cpu, GString *, diff --git a/target/avr/cpu.h b/target/avr/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/avr/cpu.h +++ b/target/avr/cpu.h @@ -XXX,XX +XXX,XX @@ static inline void set_avr_feature(CPUAVRState *env, int feature) } void avr_cpu_tcg_init(void); +void avr_cpu_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); int cpu_avr_exec(CPUState *cpu); diff --git a/target/hexagon/cpu.h b/target/hexagon/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/cpu.h +++ b/target/hexagon/cpu.h @@ -XXX,XX +XXX,XX @@ static inline void cpu_get_tb_cpu_state(CPUHexagonState *env, vaddr *pc, typedef HexagonCPU ArchCPU; void hexagon_translate_init(void); +void hexagon_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); #include "exec/cpu-all.h" diff --git a/target/hppa/cpu.h b/target/hppa/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/hppa/cpu.h +++ b/target/hppa/cpu.h @@ -XXX,XX +XXX,XX @@ static inline int HPPA_BTLB_ENTRIES(CPUHPPAState *env) } void hppa_translate_init(void); +void hppa_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); #define CPU_RESOLVING_TYPE TYPE_HPPA_CPU diff --git a/target/i386/tcg/helper-tcg.h b/target/i386/tcg/helper-tcg.h index XXXXXXX..XXXXXXX 100644 --- a/target/i386/tcg/helper-tcg.h +++ b/target/i386/tcg/helper-tcg.h @@ -XXX,XX +XXX,XX @@ static inline target_long lshift(target_long x, int n) /* translate.c */ void tcg_x86_init(void); +void x86_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); /* excp_helper.c */ G_NORETURN void raise_exception(CPUX86State *env, int exception_index); diff --git a/target/loongarch/internals.h b/target/loongarch/internals.h index XXXXXXX..XXXXXXX 100644 --- a/target/loongarch/internals.h +++ b/target/loongarch/internals.h @@ -XXX,XX +XXX,XX @@ #define TARGET_VIRT_MASK MAKE_64BIT_MASK(0, TARGET_VIRT_ADDR_SPACE_BITS) void loongarch_translate_init(void); +void loongarch_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); void G_NORETURN do_raise_exception(CPULoongArchState *env, uint32_t exception, diff --git a/target/m68k/cpu.h b/target/m68k/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/m68k/cpu.h +++ b/target/m68k/cpu.h @@ -XXX,XX +XXX,XX @@ int m68k_cpu_gdb_read_register(CPUState *cpu, GByteArray *buf, int reg); int m68k_cpu_gdb_write_register(CPUState *cpu, uint8_t *buf, int reg); void m68k_tcg_init(void); +void m68k_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); void m68k_cpu_init_gdb(M68kCPU *cpu); uint32_t cpu_m68k_get_ccr(CPUM68KState *env); void cpu_m68k_set_ccr(CPUM68KState *env, uint32_t); diff --git a/target/microblaze/cpu.h b/target/microblaze/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/microblaze/cpu.h +++ b/target/microblaze/cpu.h @@ -XXX,XX +XXX,XX @@ static inline void mb_cpu_write_msr(CPUMBState *env, uint32_t val) } void mb_tcg_init(void); +void mb_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); #define CPU_RESOLVING_TYPE TYPE_MICROBLAZE_CPU diff --git a/target/mips/tcg/tcg-internal.h b/target/mips/tcg/tcg-internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/mips/tcg/tcg-internal.h +++ b/target/mips/tcg/tcg-internal.h @@ -XXX,XX +XXX,XX @@ #include "cpu.h" void mips_tcg_init(void); +void mips_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); void mips_cpu_synchronize_from_tb(CPUState *cs, const TranslationBlock *tb); G_NORETURN void mips_cpu_do_unaligned_access(CPUState *cpu, vaddr addr, diff --git a/target/openrisc/cpu.h b/target/openrisc/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/openrisc/cpu.h +++ b/target/openrisc/cpu.h @@ -XXX,XX +XXX,XX @@ void openrisc_cpu_dump_state(CPUState *cpu, FILE *f, int flags); int openrisc_cpu_gdb_read_register(CPUState *cpu, GByteArray *buf, int reg); int openrisc_cpu_gdb_write_register(CPUState *cpu, uint8_t *buf, int reg); void openrisc_translate_init(void); +void openrisc_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); int print_insn_or1k(bfd_vma addr, disassemble_info *info); #ifndef CONFIG_USER_ONLY diff --git a/target/ppc/cpu.h b/target/ppc/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/ppc/cpu.h +++ b/target/ppc/cpu.h @@ -XXX,XX +XXX,XX @@ extern const VMStateDescription vmstate_ppc_cpu; /*****************************************************************************/ void ppc_translate_init(void); +void ppc_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); #if !defined(CONFIG_USER_ONLY) void ppc_store_sdr1(CPUPPCState *env, target_ulong value); diff --git a/target/riscv/cpu.h b/target/riscv/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/riscv/cpu.h +++ b/target/riscv/cpu.h @@ -XXX,XX +XXX,XX @@ RISCVException smstateen_acc_ok(CPURISCVState *env, int index, uint64_t bit); void riscv_cpu_set_mode(CPURISCVState *env, target_ulong newpriv, bool virt_en); void riscv_translate_init(void); +void riscv_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); + G_NORETURN void riscv_raise_exception(CPURISCVState *env, uint32_t exception, uintptr_t pc); diff --git a/target/rx/cpu.h b/target/rx/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/rx/cpu.h +++ b/target/rx/cpu.h @@ -XXX,XX +XXX,XX @@ int rx_cpu_gdb_read_register(CPUState *cpu, GByteArray *buf, int reg); int rx_cpu_gdb_write_register(CPUState *cpu, uint8_t *buf, int reg); void rx_translate_init(void); +void rx_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); void rx_cpu_unpack_psw(CPURXState *env, uint32_t psw, int rte); #include "exec/cpu-all.h" diff --git a/target/s390x/s390x-internal.h b/target/s390x/s390x-internal.h index XXXXXXX..XXXXXXX 100644 --- a/target/s390x/s390x-internal.h +++ b/target/s390x/s390x-internal.h @@ -XXX,XX +XXX,XX @@ void handle_diag_308(CPUS390XState *env, uint64_t r1, uint64_t r3, /* translate.c */ void s390x_translate_init(void); +void s390x_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); void s390x_restore_state_to_opc(CPUState *cs, const TranslationBlock *tb, const uint64_t *data); diff --git a/target/sh4/cpu.h b/target/sh4/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/sh4/cpu.h +++ b/target/sh4/cpu.h @@ -XXX,XX +XXX,XX @@ G_NORETURN void superh_cpu_do_unaligned_access(CPUState *cpu, vaddr addr, uintptr_t retaddr); void sh4_translate_init(void); +void sh4_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); #if !defined(CONFIG_USER_ONLY) hwaddr superh_cpu_get_phys_page_debug(CPUState *cpu, vaddr addr); diff --git a/target/sparc/cpu.h b/target/sparc/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/cpu.h +++ b/target/sparc/cpu.h @@ -XXX,XX +XXX,XX @@ int sparc_cpu_memory_rw_debug(CPUState *cpu, vaddr addr, /* translate.c */ void sparc_tcg_init(void); +void sparc_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); /* fop_helper.c */ target_ulong cpu_get_fsr(CPUSPARCState *); diff --git a/target/tricore/cpu.h b/target/tricore/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/tricore/cpu.h +++ b/target/tricore/cpu.h @@ -XXX,XX +XXX,XX @@ FIELD(TB_FLAGS, PRIV, 0, 2) void cpu_state_reset(CPUTriCoreState *s); void tricore_tcg_init(void); +void tricore_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); static inline void cpu_get_tb_cpu_state(CPUTriCoreState *env, vaddr *pc, uint64_t *cs_base, uint32_t *flags) diff --git a/target/xtensa/cpu.h b/target/xtensa/cpu.h index XXXXXXX..XXXXXXX 100644 --- a/target/xtensa/cpu.h +++ b/target/xtensa/cpu.h @@ -XXX,XX +XXX,XX @@ G_NORETURN void xtensa_cpu_do_unaligned_access(CPUState *cpu, vaddr addr, void xtensa_collect_sr_names(const XtensaConfig *config); void xtensa_translate_init(void); +void xtensa_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc); void **xtensa_get_regfile_by_name(const char *name, int entries, int bits); void xtensa_breakpoint_handler(CPUState *cs); void xtensa_register_core(XtensaConfigList *node); diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/cpu-exec.c +++ b/accel/tcg/cpu-exec.c @@ -XXX,XX +XXX,XX @@ bool tcg_exec_realizefn(CPUState *cpu, Error **errp) if (!tcg_target_initialized) { /* Check mandatory TCGCPUOps handlers */ + const TCGCPUOps *tcg_ops = cpu->cc->tcg_ops; #ifndef CONFIG_USER_ONLY - assert(cpu->cc->tcg_ops->cpu_exec_halt); - assert(cpu->cc->tcg_ops->cpu_exec_interrupt); + assert(tcg_ops->cpu_exec_halt); + assert(tcg_ops->cpu_exec_interrupt); #endif /* !CONFIG_USER_ONLY */ - cpu->cc->tcg_ops->initialize(); + assert(tcg_ops->translate_code); + tcg_ops->initialize(); tcg_target_initialized = true; } diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/translate-all.c +++ b/accel/tcg/translate-all.c @@ -XXX,XX +XXX,XX @@ static int setjmp_gen_code(CPUArchState *env, TranslationBlock *tb, tcg_func_start(tcg_ctx); - tcg_ctx->cpu = env_cpu(env); - gen_intermediate_code(env_cpu(env), tb, max_insns, pc, host_pc); + CPUState *cs = env_cpu(env); + tcg_ctx->cpu = cs; + cs->cc->tcg_ops->translate_code(cs, tb, max_insns, pc, host_pc); + assert(tb->size != 0); tcg_ctx->cpu = NULL; *max_insns = tb->icount; @@ -XXX,XX +XXX,XX @@ TranslationBlock *tb_gen_code(CPUState *cpu, /* * Overflow of code_gen_buffer, or the current slice of it. * - * TODO: We don't need to re-do gen_intermediate_code, nor + * TODO: We don't need to re-do tcg_ops->translate_code, nor * should we re-do the tcg optimization currently hidden * inside tcg_gen_code. All that should be required is to * flush the TBs, allocate a new TB, re-initialize it per diff --git a/target/alpha/cpu.c b/target/alpha/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/alpha/cpu.c +++ b/target/alpha/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps alpha_sysemu_ops = { static const TCGCPUOps alpha_tcg_ops = { .initialize = alpha_translate_init, + .translate_code = alpha_translate_code, .synchronize_from_tb = alpha_cpu_synchronize_from_tb, .restore_state_to_opc = alpha_restore_state_to_opc, diff --git a/target/alpha/translate.c b/target/alpha/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/alpha/translate.c +++ b/target/alpha/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps alpha_tr_ops = { .tb_stop = alpha_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void alpha_translate_code(CPUState *cpu, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; translator_loop(cpu, tb, max_insns, pc, host_pc, &alpha_tr_ops, &dc.base); diff --git a/target/arm/cpu.c b/target/arm/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/cpu.c +++ b/target/arm/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps arm_sysemu_ops = { #ifdef CONFIG_TCG static const TCGCPUOps arm_tcg_ops = { .initialize = arm_translate_init, + .translate_code = arm_translate_code, .synchronize_from_tb = arm_cpu_synchronize_from_tb, .debug_excp_handler = arm_debug_excp_handler, .restore_state_to_opc = arm_restore_state_to_opc, diff --git a/target/arm/tcg/cpu-v7m.c b/target/arm/tcg/cpu-v7m.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/cpu-v7m.c +++ b/target/arm/tcg/cpu-v7m.c @@ -XXX,XX +XXX,XX @@ static void cortex_m55_initfn(Object *obj) static const TCGCPUOps arm_v7m_tcg_ops = { .initialize = arm_translate_init, + .translate_code = arm_translate_code, .synchronize_from_tb = arm_cpu_synchronize_from_tb, .debug_excp_handler = arm_debug_excp_handler, .restore_state_to_opc = arm_restore_state_to_opc, diff --git a/target/arm/tcg/translate.c b/target/arm/tcg/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/arm/tcg/translate.c +++ b/target/arm/tcg/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps thumb_translator_ops = { .tb_stop = arm_tr_tb_stop, }; -/* generate intermediate code for basic block 'tb'. */ -void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void arm_translate_code(CPUState *cpu, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc = { }; const TranslatorOps *ops = &arm_translator_ops; diff --git a/target/avr/cpu.c b/target/avr/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/avr/cpu.c +++ b/target/avr/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps avr_sysemu_ops = { static const TCGCPUOps avr_tcg_ops = { .initialize = avr_cpu_tcg_init, + .translate_code = avr_cpu_translate_code, .synchronize_from_tb = avr_cpu_synchronize_from_tb, .restore_state_to_opc = avr_restore_state_to_opc, .cpu_exec_interrupt = avr_cpu_exec_interrupt, diff --git a/target/avr/translate.c b/target/avr/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/avr/translate.c +++ b/target/avr/translate.c @@ -XXX,XX +XXX,XX @@ static bool trans_WDR(DisasContext *ctx, arg_WDR *a) * * - translate() * - canonicalize_skip() - * - gen_intermediate_code() + * - translate_code() * - restore_state_to_opc() * */ @@ -XXX,XX +XXX,XX @@ static const TranslatorOps avr_tr_ops = { .tb_stop = avr_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void avr_cpu_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc = { }; translator_loop(cs, tb, max_insns, pc, host_pc, &avr_tr_ops, &dc.base); diff --git a/target/hexagon/cpu.c b/target/hexagon/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/cpu.c +++ b/target/hexagon/cpu.c @@ -XXX,XX +XXX,XX @@ static void hexagon_cpu_init(Object *obj) static const TCGCPUOps hexagon_tcg_ops = { .initialize = hexagon_translate_init, + .translate_code = hexagon_translate_code, .synchronize_from_tb = hexagon_cpu_synchronize_from_tb, .restore_state_to_opc = hexagon_restore_state_to_opc, }; diff --git a/target/hexagon/translate.c b/target/hexagon/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/hexagon/translate.c +++ b/target/hexagon/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps hexagon_tr_ops = { .tb_stop = hexagon_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void hexagon_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; diff --git a/target/hppa/cpu.c b/target/hppa/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/hppa/cpu.c +++ b/target/hppa/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps hppa_sysemu_ops = { static const TCGCPUOps hppa_tcg_ops = { .initialize = hppa_translate_init, + .translate_code = hppa_translate_code, .synchronize_from_tb = hppa_cpu_synchronize_from_tb, .restore_state_to_opc = hppa_restore_state_to_opc, diff --git a/target/hppa/translate.c b/target/hppa/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/hppa/translate.c +++ b/target/hppa/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps hppa_tr_ops = { #endif }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void hppa_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx = { }; translator_loop(cs, tb, max_insns, pc, host_pc, &hppa_tr_ops, &ctx.base); diff --git a/target/i386/tcg/tcg-cpu.c b/target/i386/tcg/tcg-cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/i386/tcg/tcg-cpu.c +++ b/target/i386/tcg/tcg-cpu.c @@ -XXX,XX +XXX,XX @@ static bool x86_debug_check_breakpoint(CPUState *cs) static const TCGCPUOps x86_tcg_ops = { .initialize = tcg_x86_init, + .translate_code = x86_translate_code, .synchronize_from_tb = x86_cpu_synchronize_from_tb, .restore_state_to_opc = x86_restore_state_to_opc, .cpu_exec_enter = x86_cpu_exec_enter, diff --git a/target/i386/tcg/translate.c b/target/i386/tcg/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/i386/tcg/translate.c +++ b/target/i386/tcg/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps i386_tr_ops = { .tb_stop = i386_tr_tb_stop, }; -/* generate intermediate code for basic block 'tb'. */ -void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void x86_translate_code(CPUState *cpu, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; diff --git a/target/loongarch/cpu.c b/target/loongarch/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/loongarch/cpu.c +++ b/target/loongarch/cpu.c @@ -XXX,XX +XXX,XX @@ static void loongarch_cpu_dump_state(CPUState *cs, FILE *f, int flags) static const TCGCPUOps loongarch_tcg_ops = { .initialize = loongarch_translate_init, + .translate_code = loongarch_translate_code, .synchronize_from_tb = loongarch_cpu_synchronize_from_tb, .restore_state_to_opc = loongarch_restore_state_to_opc, diff --git a/target/loongarch/tcg/translate.c b/target/loongarch/tcg/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/loongarch/tcg/translate.c +++ b/target/loongarch/tcg/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps loongarch_tr_ops = { .tb_stop = loongarch_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void loongarch_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; diff --git a/target/m68k/cpu.c b/target/m68k/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/m68k/cpu.c +++ b/target/m68k/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps m68k_sysemu_ops = { static const TCGCPUOps m68k_tcg_ops = { .initialize = m68k_tcg_init, + .translate_code = m68k_translate_code, .restore_state_to_opc = m68k_restore_state_to_opc, #ifndef CONFIG_USER_ONLY diff --git a/target/m68k/translate.c b/target/m68k/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/m68k/translate.c +++ b/target/m68k/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps m68k_tr_ops = { .tb_stop = m68k_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void m68k_translate_code(CPUState *cpu, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; translator_loop(cpu, tb, max_insns, pc, host_pc, &m68k_tr_ops, &dc.base); diff --git a/target/microblaze/cpu.c b/target/microblaze/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/microblaze/cpu.c +++ b/target/microblaze/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps mb_sysemu_ops = { static const TCGCPUOps mb_tcg_ops = { .initialize = mb_tcg_init, + .translate_code = mb_translate_code, .synchronize_from_tb = mb_cpu_synchronize_from_tb, .restore_state_to_opc = mb_restore_state_to_opc, diff --git a/target/microblaze/translate.c b/target/microblaze/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/microblaze/translate.c +++ b/target/microblaze/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps mb_tr_ops = { .tb_stop = mb_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void mb_translate_code(CPUState *cpu, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; translator_loop(cpu, tb, max_insns, pc, host_pc, &mb_tr_ops, &dc.base); diff --git a/target/mips/cpu.c b/target/mips/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/mips/cpu.c +++ b/target/mips/cpu.c @@ -XXX,XX +XXX,XX @@ static const Property mips_cpu_properties[] = { #include "hw/core/tcg-cpu-ops.h" static const TCGCPUOps mips_tcg_ops = { .initialize = mips_tcg_init, + .translate_code = mips_translate_code, .synchronize_from_tb = mips_cpu_synchronize_from_tb, .restore_state_to_opc = mips_restore_state_to_opc, diff --git a/target/mips/tcg/translate.c b/target/mips/tcg/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/mips/tcg/translate.c +++ b/target/mips/tcg/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps mips_tr_ops = { .tb_stop = mips_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void mips_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; diff --git a/target/openrisc/cpu.c b/target/openrisc/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/openrisc/cpu.c +++ b/target/openrisc/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps openrisc_sysemu_ops = { static const TCGCPUOps openrisc_tcg_ops = { .initialize = openrisc_translate_init, + .translate_code = openrisc_translate_code, .synchronize_from_tb = openrisc_cpu_synchronize_from_tb, .restore_state_to_opc = openrisc_restore_state_to_opc, diff --git a/target/openrisc/translate.c b/target/openrisc/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/openrisc/translate.c +++ b/target/openrisc/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps openrisc_tr_ops = { .tb_stop = openrisc_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void openrisc_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; diff --git a/target/ppc/cpu_init.c b/target/ppc/cpu_init.c index XXXXXXX..XXXXXXX 100644 --- a/target/ppc/cpu_init.c +++ b/target/ppc/cpu_init.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps ppc_sysemu_ops = { static const TCGCPUOps ppc_tcg_ops = { .initialize = ppc_translate_init, + .translate_code = ppc_translate_code, .restore_state_to_opc = ppc_restore_state_to_opc, #ifdef CONFIG_USER_ONLY diff --git a/target/ppc/translate.c b/target/ppc/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/ppc/translate.c +++ b/target/ppc/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps ppc_tr_ops = { .tb_stop = ppc_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void ppc_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; diff --git a/target/riscv/tcg/tcg-cpu.c b/target/riscv/tcg/tcg-cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/riscv/tcg/tcg-cpu.c +++ b/target/riscv/tcg/tcg-cpu.c @@ -XXX,XX +XXX,XX @@ static void riscv_restore_state_to_opc(CPUState *cs, static const TCGCPUOps riscv_tcg_ops = { .initialize = riscv_translate_init, + .translate_code = riscv_translate_code, .synchronize_from_tb = riscv_cpu_synchronize_from_tb, .restore_state_to_opc = riscv_restore_state_to_opc, diff --git a/target/riscv/translate.c b/target/riscv/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/riscv/translate.c +++ b/target/riscv/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps riscv_tr_ops = { .tb_stop = riscv_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void riscv_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; diff --git a/target/rx/cpu.c b/target/rx/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/rx/cpu.c +++ b/target/rx/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps rx_sysemu_ops = { static const TCGCPUOps rx_tcg_ops = { .initialize = rx_translate_init, + .translate_code = rx_translate_code, .synchronize_from_tb = rx_cpu_synchronize_from_tb, .restore_state_to_opc = rx_restore_state_to_opc, .tlb_fill = rx_cpu_tlb_fill, diff --git a/target/rx/translate.c b/target/rx/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/rx/translate.c +++ b/target/rx/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps rx_tr_ops = { .tb_stop = rx_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void rx_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; diff --git a/target/s390x/cpu.c b/target/s390x/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/s390x/cpu.c +++ b/target/s390x/cpu.c @@ -XXX,XX +XXX,XX @@ void cpu_get_tb_cpu_state(CPUS390XState *env, vaddr *pc, static const TCGCPUOps s390_tcg_ops = { .initialize = s390x_translate_init, + .translate_code = s390x_translate_code, .restore_state_to_opc = s390x_restore_state_to_opc, #ifdef CONFIG_USER_ONLY diff --git a/target/s390x/tcg/translate.c b/target/s390x/tcg/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/s390x/tcg/translate.c +++ b/target/s390x/tcg/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps s390x_tr_ops = { .disas_log = s390x_tr_disas_log, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void s390x_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc; diff --git a/target/sh4/cpu.c b/target/sh4/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/sh4/cpu.c +++ b/target/sh4/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps sh4_sysemu_ops = { static const TCGCPUOps superh_tcg_ops = { .initialize = sh4_translate_init, + .translate_code = sh4_translate_code, .synchronize_from_tb = superh_cpu_synchronize_from_tb, .restore_state_to_opc = superh_restore_state_to_opc, diff --git a/target/sh4/translate.c b/target/sh4/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/sh4/translate.c +++ b/target/sh4/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps sh4_tr_ops = { .tb_stop = sh4_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void sh4_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/cpu.c +++ b/target/sparc/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps sparc_sysemu_ops = { static const TCGCPUOps sparc_tcg_ops = { .initialize = sparc_tcg_init, + .translate_code = sparc_translate_code, .synchronize_from_tb = sparc_cpu_synchronize_from_tb, .restore_state_to_opc = sparc_restore_state_to_opc, diff --git a/target/sparc/translate.c b/target/sparc/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/sparc/translate.c +++ b/target/sparc/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps sparc_tr_ops = { .tb_stop = sparc_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void sparc_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc = {}; diff --git a/target/tricore/cpu.c b/target/tricore/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/tricore/cpu.c +++ b/target/tricore/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps tricore_sysemu_ops = { static const TCGCPUOps tricore_tcg_ops = { .initialize = tricore_tcg_init, + .translate_code = tricore_translate_code, .synchronize_from_tb = tricore_cpu_synchronize_from_tb, .restore_state_to_opc = tricore_restore_state_to_opc, .tlb_fill = tricore_cpu_tlb_fill, diff --git a/target/tricore/translate.c b/target/tricore/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/tricore/translate.c +++ b/target/tricore/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps tricore_tr_ops = { .tb_stop = tricore_tr_tb_stop, }; - -void gen_intermediate_code(CPUState *cs, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void tricore_translate_code(CPUState *cs, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext ctx; translator_loop(cs, tb, max_insns, pc, host_pc, diff --git a/target/xtensa/cpu.c b/target/xtensa/cpu.c index XXXXXXX..XXXXXXX 100644 --- a/target/xtensa/cpu.c +++ b/target/xtensa/cpu.c @@ -XXX,XX +XXX,XX @@ static const struct SysemuCPUOps xtensa_sysemu_ops = { static const TCGCPUOps xtensa_tcg_ops = { .initialize = xtensa_translate_init, + .translate_code = xtensa_translate_code, .debug_excp_handler = xtensa_breakpoint_handler, .restore_state_to_opc = xtensa_restore_state_to_opc, diff --git a/target/xtensa/translate.c b/target/xtensa/translate.c index XXXXXXX..XXXXXXX 100644 --- a/target/xtensa/translate.c +++ b/target/xtensa/translate.c @@ -XXX,XX +XXX,XX @@ static const TranslatorOps xtensa_translator_ops = { .tb_stop = xtensa_tr_tb_stop, }; -void gen_intermediate_code(CPUState *cpu, TranslationBlock *tb, int *max_insns, - vaddr pc, void *host_pc) +void xtensa_translate_code(CPUState *cpu, TranslationBlock *tb, + int *max_insns, vaddr pc, void *host_pc) { DisasContext dc = {}; translator_loop(cpu, tb, max_insns, pc, host_pc, -- 2.43.0 Changes from v1: * Patch 10 is new, avoiding an overflow in probe_guest_base, visible with aarch64 host, --static --disable-pie, exposed by the placement of the host binary in the address space. r~ Emilio Cota (2): util: import GTree as QTree tcg: use QTree instead of GTree Richard Henderson (10): linux-user: Diagnose misaligned -R size accel/tcg: Pass last not end to page_set_flags accel/tcg: Pass last not end to page_reset_target_data accel/tcg: Pass last not end to PAGE_FOR_EACH_TB accel/tcg: Pass last not end to page_collection_lock accel/tcg: Pass last not end to tb_invalidate_phys_page_range__locked accel/tcg: Pass last not end to tb_invalidate_phys_range linux-user: Pass last not end to probe_guest_base include/exec: Change reserved_va semantics to last byte linux-user/arm: Take more care allocating commpage configure | 15 + meson.build | 4 + include/exec/cpu-all.h | 15 +- include/exec/exec-all.h | 2 +- include/qemu/qtree.h | 201 +++++ linux-user/arm/target_cpu.h | 2 +- accel/tcg/tb-maint.c | 112 +-- accel/tcg/translate-all.c | 2 +- accel/tcg/user-exec.c | 25 +- bsd-user/main.c | 10 +- bsd-user/mmap.c | 10 +- linux-user/elfload.c | 72 +- linux-user/flatload.c | 2 +- linux-user/main.c | 31 +- linux-user/mmap.c | 22 +- linux-user/syscall.c | 4 +- softmmu/physmem.c | 2 +- tcg/region.c | 19 +- tests/bench/qtree-bench.c | 286 +++++++ tests/unit/test-qtree.c | 333 +++++++++ util/qtree.c | 1390 +++++++++++++++++++++++++++++++++++ tests/bench/meson.build | 4 + tests/unit/meson.build | 1 + util/meson.build | 1 + 24 files changed, 2415 insertions(+), 150 deletions(-) create mode 100644 include/qemu/qtree.h create mode 100644 tests/bench/qtree-bench.c create mode 100644 tests/unit/test-qtree.c create mode 100644 util/qtree.c -- 2.34.1 From: Emilio Cota <cota@braap.org> The only reason to add this implementation is to control the memory allocator used. Some users (e.g. TCG) cannot work reliably in multi-threaded environments (e.g. forking in user-mode) with GTree's allocator, GSlice. See https://gitlab.com/qemu-project/qemu/-/issues/285 for details. Importing GTree is a temporary workaround until GTree migrates away from GSlice. This implementation is identical to that in glib v2.75.0, except that we don't import recent additions to the API nor deprecated API calls, none of which are used in QEMU. I've imported tests from glib and added a benchmark just to make sure that performance is similar. Note: it cannot be identical because (1) we are not using GSlice, (2) we use different compilation flags (e.g. -fPIC) and (3) we're linking statically. $ cat /proc/cpuinfo| grep 'model name' | head -1 model name : AMD Ryzen 7 PRO 5850U with Radeon Graphics $ echo '0' | sudo tee /sys/devices/system/cpu/cpufreq/boost $ tests/bench/qtree-bench Tree Op 32 1024 4096 131072 1048576 ------------------------------------------------------------------------------------------------ GTree Lookup 83.23 43.08 25.31 19.40 16.22 QTree Lookup 113.42 (1.36x) 53.83 (1.25x) 28.38 (1.12x) 17.64 (0.91x) 13.04 (0.80x) GTree Insert 44.23 29.37 25.83 19.49 17.03 QTree Insert 46.87 (1.06x) 25.62 (0.87x) 24.29 (0.94x) 16.83 (0.86x) 12.97 (0.76x) GTree Remove 53.27 35.15 31.43 24.64 16.70 QTree Remove 57.32 (1.08x) 41.76 (1.19x) 38.37 (1.22x) 29.30 (1.19x) 15.07 (0.90x) GTree RemoveAll 135.44 127.52 126.72 120.11 64.34 QTree RemoveAll 127.15 (0.94x) 110.37 (0.87x) 107.97 (0.85x) 97.13 (0.81x) 55.10 (0.86x) GTree Traverse 277.71 276.09 272.78 246.72 98.47 QTree Traverse 370.33 (1.33x) 411.97 (1.49x) 400.23 (1.47x) 262.82 (1.07x) 78.52 (0.80x) ------------------------------------------------------------------------------------------------ As a sanity check, the same benchmark when Glib's version is >= $glib_dropped_gslice_version (i.e. QTree == GTree): Tree Op 32 1024 4096 131072 1048576 ------------------------------------------------------------------------------------------------ GTree Lookup 82.72 43.09 24.18 19.73 16.09 QTree Lookup 81.82 (0.99x) 43.10 (1.00x) 24.20 (1.00x) 19.76 (1.00x) 16.26 (1.01x) GTree Insert 45.07 29.62 26.34 19.90 17.18 QTree Insert 45.72 (1.01x) 29.60 (1.00x) 26.38 (1.00x) 19.71 (0.99x) 17.20 (1.00x) GTree Remove 54.48 35.36 31.77 24.97 16.95 QTree Remove 54.46 (1.00x) 35.32 (1.00x) 31.77 (1.00x) 24.91 (1.00x) 17.15 (1.01x) GTree RemoveAll 140.68 127.36 125.43 121.45 68.20 QTree RemoveAll 140.65 (1.00x) 127.64 (1.00x) 125.01 (1.00x) 121.73 (1.00x) 67.06 (0.98x) GTree Traverse 278.68 276.05 266.75 251.65 104.93 QTree Traverse 278.31 (1.00x) 275.78 (1.00x) 266.42 (1.00x) 247.89 (0.99x) 104.58 (1.00x) ------------------------------------------------------------------------------------------------ Signed-off-by: Emilio Cota <cota@braap.org> Message-Id: <20230205163758.416992-2-cota@braap.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- configure | 15 + meson.build | 4 + include/qemu/qtree.h | 201 ++++++ tests/bench/qtree-bench.c | 286 ++++++++ tests/unit/test-qtree.c | 333 +++++++++ util/qtree.c | 1390 +++++++++++++++++++++++++++++++++++++ tests/bench/meson.build | 4 + tests/unit/meson.build | 1 + util/meson.build | 1 + 9 files changed, 2235 insertions(+) create mode 100644 include/qemu/qtree.h create mode 100644 tests/bench/qtree-bench.c create mode 100644 tests/unit/test-qtree.c create mode 100644 util/qtree.c diff --git a/configure b/configure index XXXXXXX..XXXXXXX 100755 --- a/configure +++ b/configure @@ -XXX,XX +XXX,XX @@ safe_stack="" use_containers="yes" gdb_bin=$(command -v "gdb-multiarch" || command -v "gdb") gdb_arches="" +glib_has_gslice="no" if test -e "$source_path/.git" then @@ -XXX,XX +XXX,XX @@ for i in $glib_modules; do fi done +# Check whether glib has gslice, which we have to avoid for correctness. +# TODO: remove this check and the corresponding workaround (qtree) when +# the minimum supported glib is >= $glib_dropped_gslice_version. +glib_dropped_gslice_version=2.75.3 +for i in $glib_modules; do + if ! $pkg_config --atleast-version=$glib_dropped_gslice_version $i; then + glib_has_gslice="yes" + break + fi +done + glib_bindir="$($pkg_config --variable=bindir glib-2.0)" if test -z "$glib_bindir" ; then glib_bindir="$($pkg_config --variable=prefix glib-2.0)"/bin @@ -XXX,XX +XXX,XX @@ echo "GLIB_CFLAGS=$glib_cflags" >> $config_host_mak echo "GLIB_LIBS=$glib_libs" >> $config_host_mak echo "GLIB_BINDIR=$glib_bindir" >> $config_host_mak echo "GLIB_VERSION=$($pkg_config --modversion glib-2.0)" >> $config_host_mak +if test "$glib_has_gslice" = "yes" ; then + echo "HAVE_GLIB_WITH_SLICE_ALLOCATOR=y" >> $config_host_mak +fi echo "QEMU_LDFLAGS=$QEMU_LDFLAGS" >> $config_host_mak echo "EXESUF=$EXESUF" >> $config_host_mak diff --git a/meson.build b/meson.build index XXXXXXX..XXXXXXX 100644 --- a/meson.build +++ b/meson.build @@ -XXX,XX +XXX,XX @@ glib = declare_dependency(compile_args: config_host['GLIB_CFLAGS'].split(), }) # override glib dep with the configure results (for subprojects) meson.override_dependency('glib-2.0', glib) +# pass down whether Glib has the slice allocator +if config_host.has_key('HAVE_GLIB_WITH_SLICE_ALLOCATOR') + config_host_data.set('HAVE_GLIB_WITH_SLICE_ALLOCATOR', true) +endif gio = not_found gdbus_codegen = not_found diff --git a/include/qemu/qtree.h b/include/qemu/qtree.h new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/include/qemu/qtree.h @@ -XXX,XX +XXX,XX @@ +/* + * GLIB - Library of useful routines for C programming + * Copyright (C) 1995-1997 Peter Mattis, Spencer Kimball and Josh MacDonald + * + * SPDX-License-Identifier: LGPL-2.1-or-later + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, see <http://www.gnu.org/licenses/>. + */ + +/* + * Modified by the GLib Team and others 1997-2000. See the AUTHORS + * file for a list of people on the GLib Team. See the ChangeLog + * files for a list of changes. These files are distributed with + * GLib at ftp://ftp.gtk.org/pub/gtk/. + */ + +/* + * QTree is a partial import of Glib's GTree. The parts excluded correspond + * to API calls either deprecated (e.g. g_tree_traverse) or recently added + * (e.g. g_tree_search_node, added in 2.68); neither have callers in QEMU. + * + * The reason for this import is to allow us to control the memory allocator + * used by the tree implementation. Until Glib 2.75.3, GTree uses Glib's + * slice allocator, which causes problems when forking in user-mode; + * see https://gitlab.com/qemu-project/qemu/-/issues/285 and glib's + * "45b5a6c1e gslice: Remove slice allocator and use malloc() instead". + * + * TODO: remove QTree when QEMU's minimum Glib version is >= 2.75.3. + */ + +#ifndef QEMU_QTREE_H +#define QEMU_QTREE_H + +#include "qemu/osdep.h" + +#ifdef HAVE_GLIB_WITH_SLICE_ALLOCATOR + +typedef struct _QTree QTree; + +typedef struct _QTreeNode QTreeNode; + +typedef gboolean (*QTraverseNodeFunc)(QTreeNode *node, + gpointer user_data); + +/* + * Balanced binary trees + */ +QTree *q_tree_new(GCompareFunc key_compare_func); +QTree *q_tree_new_with_data(GCompareDataFunc key_compare_func, + gpointer key_compare_data); +QTree *q_tree_new_full(GCompareDataFunc key_compare_func, + gpointer key_compare_data, + GDestroyNotify key_destroy_func, + GDestroyNotify value_destroy_func); +QTree *q_tree_ref(QTree *tree); +void q_tree_unref(QTree *tree); +void q_tree_destroy(QTree *tree); +void q_tree_insert(QTree *tree, + gpointer key, + gpointer value); +void q_tree_replace(QTree *tree, + gpointer key, + gpointer value); +gboolean q_tree_remove(QTree *tree, + gconstpointer key); +gboolean q_tree_steal(QTree *tree, + gconstpointer key); +gpointer q_tree_lookup(QTree *tree, + gconstpointer key); +gboolean q_tree_lookup_extended(QTree *tree, + gconstpointer lookup_key, + gpointer *orig_key, + gpointer *value); +void q_tree_foreach(QTree *tree, + GTraverseFunc func, + gpointer user_data); +gpointer q_tree_search(QTree *tree, + GCompareFunc search_func, + gconstpointer user_data); +gint q_tree_height(QTree *tree); +gint q_tree_nnodes(QTree *tree); + +#else /* !HAVE_GLIB_WITH_SLICE_ALLOCATOR */ + +typedef GTree QTree; +typedef GTreeNode QTreeNode; +typedef GTraverseNodeFunc QTraverseNodeFunc; + +static inline QTree *q_tree_new(GCompareFunc key_compare_func) +{ + return g_tree_new(key_compare_func); +} + +static inline QTree *q_tree_new_with_data(GCompareDataFunc key_compare_func, + gpointer key_compare_data) +{ + return g_tree_new_with_data(key_compare_func, key_compare_data); +} + +static inline QTree *q_tree_new_full(GCompareDataFunc key_compare_func, + gpointer key_compare_data, + GDestroyNotify key_destroy_func, + GDestroyNotify value_destroy_func) +{ + return g_tree_new_full(key_compare_func, key_compare_data, + key_destroy_func, value_destroy_func); +} + +static inline QTree *q_tree_ref(QTree *tree) +{ + return g_tree_ref(tree); +} + +static inline void q_tree_unref(QTree *tree) +{ + g_tree_unref(tree); +} + +static inline void q_tree_destroy(QTree *tree) +{ + g_tree_destroy(tree); +} + +static inline void q_tree_insert(QTree *tree, + gpointer key, + gpointer value) +{ + g_tree_insert(tree, key, value); +} + +static inline void q_tree_replace(QTree *tree, + gpointer key, + gpointer value) +{ + g_tree_replace(tree, key, value); +} + +static inline gboolean q_tree_remove(QTree *tree, + gconstpointer key) +{ + return g_tree_remove(tree, key); +} + +static inline gboolean q_tree_steal(QTree *tree, + gconstpointer key) +{ + return g_tree_steal(tree, key); +} + +static inline gpointer q_tree_lookup(QTree *tree, + gconstpointer key) +{ + return g_tree_lookup(tree, key); +} + +static inline gboolean q_tree_lookup_extended(QTree *tree, + gconstpointer lookup_key, + gpointer *orig_key, + gpointer *value) +{ + return g_tree_lookup_extended(tree, lookup_key, orig_key, value); +} + +static inline void q_tree_foreach(QTree *tree, + GTraverseFunc func, + gpointer user_data) +{ + return g_tree_foreach(tree, func, user_data); +} + +static inline gpointer q_tree_search(QTree *tree, + GCompareFunc search_func, + gconstpointer user_data) +{ + return g_tree_search(tree, search_func, user_data); +} + +static inline gint q_tree_height(QTree *tree) +{ + return g_tree_height(tree); +} + +static inline gint q_tree_nnodes(QTree *tree) +{ + return g_tree_nnodes(tree); +} + +#endif /* HAVE_GLIB_WITH_SLICE_ALLOCATOR */ + +#endif /* QEMU_QTREE_H */ diff --git a/tests/bench/qtree-bench.c b/tests/bench/qtree-bench.c new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/tests/bench/qtree-bench.c @@ -XXX,XX +XXX,XX @@ +/* SPDX-License-Identifier: GPL-2.0-or-later */ +#include "qemu/osdep.h" +#include "qemu/qtree.h" +#include "qemu/timer.h" + +enum tree_op { + OP_LOOKUP, + OP_INSERT, + OP_REMOVE, + OP_REMOVE_ALL, + OP_TRAVERSE, +}; + +struct benchmark { + const char * const name; + enum tree_op op; + bool fill_on_init; +}; + +enum impl_type { + IMPL_GTREE, + IMPL_QTREE, +}; + +struct tree_implementation { + const char * const name; + enum impl_type type; +}; + +static const struct benchmark benchmarks[] = { + { + .name = "Lookup", + .op = OP_LOOKUP, + .fill_on_init = true, + }, + { + .name = "Insert", + .op = OP_INSERT, + .fill_on_init = false, + }, + { + .name = "Remove", + .op = OP_REMOVE, + .fill_on_init = true, + }, + { + .name = "RemoveAll", + .op = OP_REMOVE_ALL, + .fill_on_init = true, + }, + { + .name = "Traverse", + .op = OP_TRAVERSE, + .fill_on_init = true, + }, +}; + +static const struct tree_implementation impls[] = { + { + .name = "GTree", + .type = IMPL_GTREE, + }, + { + .name = "QTree", + .type = IMPL_QTREE, + }, +}; + +static int compare_func(const void *ap, const void *bp) +{ + const size_t *a = ap; + const size_t *b = bp; + + return *a - *b; +} + +static void init_empty_tree_and_keys(enum impl_type impl, + void **ret_tree, size_t **ret_keys, + size_t n_elems) +{ + size_t *keys = g_malloc_n(n_elems, sizeof(*keys)); + for (size_t i = 0; i < n_elems; i++) { + keys[i] = i; + } + + void *tree; + switch (impl) { + case IMPL_GTREE: + tree = g_tree_new(compare_func); + break; + case IMPL_QTREE: + tree = q_tree_new(compare_func); + break; + default: + g_assert_not_reached(); + } + + *ret_tree = tree; + *ret_keys = keys; +} + +static gboolean traverse_func(gpointer key, gpointer value, gpointer data) +{ + return FALSE; +} + +static inline void remove_all(void *tree, enum impl_type impl) +{ + switch (impl) { + case IMPL_GTREE: + g_tree_destroy(tree); + break; + case IMPL_QTREE: + q_tree_destroy(tree); + break; + default: + g_assert_not_reached(); + } +} + +static int64_t run_benchmark(const struct benchmark *bench, + enum impl_type impl, + size_t n_elems) +{ + void *tree; + size_t *keys; + + init_empty_tree_and_keys(impl, &tree, &keys, n_elems); + if (bench->fill_on_init) { + for (size_t i = 0; i < n_elems; i++) { + switch (impl) { + case IMPL_GTREE: + g_tree_insert(tree, &keys[i], &keys[i]); + break; + case IMPL_QTREE: + q_tree_insert(tree, &keys[i], &keys[i]); + break; + default: + g_assert_not_reached(); + } + } + } + + int64_t start_ns = get_clock(); + switch (bench->op) { + case OP_LOOKUP: + for (size_t i = 0; i < n_elems; i++) { + void *value; + switch (impl) { + case IMPL_GTREE: + value = g_tree_lookup(tree, &keys[i]); + break; + case IMPL_QTREE: + value = q_tree_lookup(tree, &keys[i]); + break; + default: + g_assert_not_reached(); + } + (void)value; + } + break; + case OP_INSERT: + for (size_t i = 0; i < n_elems; i++) { + switch (impl) { + case IMPL_GTREE: + g_tree_insert(tree, &keys[i], &keys[i]); + break; + case IMPL_QTREE: + q_tree_insert(tree, &keys[i], &keys[i]); + break; + default: + g_assert_not_reached(); + } + } + break; + case OP_REMOVE: + for (size_t i = 0; i < n_elems; i++) { + switch (impl) { + case IMPL_GTREE: + g_tree_remove(tree, &keys[i]); + break; + case IMPL_QTREE: + q_tree_remove(tree, &keys[i]); + break; + default: + g_assert_not_reached(); + } + } + break; + case OP_REMOVE_ALL: + remove_all(tree, impl); + break; + case OP_TRAVERSE: + switch (impl) { + case IMPL_GTREE: + g_tree_foreach(tree, traverse_func, NULL); + break; + case IMPL_QTREE: + q_tree_foreach(tree, traverse_func, NULL); + break; + default: + g_assert_not_reached(); + } + break; + default: + g_assert_not_reached(); + } + int64_t ns = get_clock() - start_ns; + + if (bench->op != OP_REMOVE_ALL) { + remove_all(tree, impl); + } + g_free(keys); + + return ns; +} + +int main(int argc, char *argv[]) +{ + size_t sizes[] = { + 32, + 1024, + 1024 * 4, + 1024 * 128, + 1024 * 1024, + }; + + double res[ARRAY_SIZE(benchmarks)][ARRAY_SIZE(impls)][ARRAY_SIZE(sizes)]; + for (int i = 0; i < ARRAY_SIZE(sizes); i++) { + size_t size = sizes[i]; + for (int j = 0; j < ARRAY_SIZE(impls); j++) { + const struct tree_implementation *impl = &impls[j]; + for (int k = 0; k < ARRAY_SIZE(benchmarks); k++) { + const struct benchmark *bench = &benchmarks[k]; + + /* warm-up run */ + run_benchmark(bench, impl->type, size); + + int64_t total_ns = 0; + int64_t n_runs = 0; + while (total_ns < 2e8 || n_runs < 5) { + total_ns += run_benchmark(bench, impl->type, size); + n_runs++; + } + double ns_per_run = (double)total_ns / n_runs; + + /* Throughput, in Mops/s */ + res[k][j][i] = size / ns_per_run * 1e3; + } + } + } + + printf("# Results' breakdown: Tree, Op and #Elements. Units: Mops/s\n"); + printf("%5s %10s ", "Tree", "Op"); + for (int i = 0; i < ARRAY_SIZE(sizes); i++) { + printf("%7zu ", sizes[i]); + } + printf("\n"); + char separator[97]; + for (int i = 0; i < ARRAY_SIZE(separator) - 1; i++) { + separator[i] = '-'; + } + separator[ARRAY_SIZE(separator) - 1] = '\0'; + printf("%s\n", separator); + for (int i = 0; i < ARRAY_SIZE(benchmarks); i++) { + for (int j = 0; j < ARRAY_SIZE(impls); j++) { + printf("%5s %10s ", impls[j].name, benchmarks[i].name); + for (int k = 0; k < ARRAY_SIZE(sizes); k++) { + printf("%7.2f ", res[i][j][k]); + if (j == 0) { + printf(" "); + } else { + if (res[i][0][k] != 0) { + double speedup = res[i][j][k] / res[i][0][k]; + printf("(%4.2fx) ", speedup); + } else { + printf("( ) "); + } + } + } + printf("\n"); + } + } + printf("%s\n", separator); + return 0; +} diff --git a/tests/unit/test-qtree.c b/tests/unit/test-qtree.c new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/tests/unit/test-qtree.c @@ -XXX,XX +XXX,XX @@ +/* + * SPDX-License-Identifier: LGPL-2.1-or-later + * + * Tests for QTree. + * Original source: glib + * https://gitlab.gnome.org/GNOME/glib/-/blob/main/glib/tests/tree.c + * LGPL license. + * Copyright (C) 1995-1997 Peter Mattis, Spencer Kimball and Josh MacDonald + */ + +#include "qemu/osdep.h" +#include "qemu/qtree.h" + +static gint my_compare(gconstpointer a, gconstpointer b) +{ + const char *cha = a; + const char *chb = b; + + return *cha - *chb; +} + +static gint my_compare_with_data(gconstpointer a, + gconstpointer b, + gpointer user_data) +{ + const char *cha = a; + const char *chb = b; + + /* just check that we got the right data */ + g_assert(GPOINTER_TO_INT(user_data) == 123); + + return *cha - *chb; +} + +static gint my_search(gconstpointer a, gconstpointer b) +{ + return my_compare(b, a); +} + +static gpointer destroyed_key; +static gpointer destroyed_value; +static guint destroyed_key_count; +static guint destroyed_value_count; + +static void my_key_destroy(gpointer key) +{ + destroyed_key = key; + destroyed_key_count++; +} + +static void my_value_destroy(gpointer value) +{ + destroyed_value = value; + destroyed_value_count++; +} + +static gint my_traverse(gpointer key, gpointer value, gpointer data) +{ + char *ch = key; + + g_assert((*ch) > 0); + + if (*ch == 'd') { + return TRUE; + } + + return FALSE; +} + +char chars[] = + "0123456789" + "ABCDEFGHIJKLMNOPQRSTUVWXYZ" + "abcdefghijklmnopqrstuvwxyz"; + +char chars2[] = + "0123456789" + "abcdefghijklmnopqrstuvwxyz"; + +static gint check_order(gpointer key, gpointer value, gpointer data) +{ + char **p = data; + char *ch = key; + + g_assert(**p == *ch); + + (*p)++; + + return FALSE; +} + +static void test_tree_search(void) +{ + gint i; + QTree *tree; + gboolean removed; + gchar c; + gchar *p, *d; + + tree = q_tree_new_with_data(my_compare_with_data, GINT_TO_POINTER(123)); + + for (i = 0; chars[i]; i++) { + q_tree_insert(tree, &chars[i], &chars[i]); + } + + q_tree_foreach(tree, my_traverse, NULL); + + g_assert(q_tree_nnodes(tree) == strlen(chars)); + g_assert(q_tree_height(tree) == 6); + + p = chars; + q_tree_foreach(tree, check_order, &p); + + for (i = 0; i < 26; i++) { + removed = q_tree_remove(tree, &chars[i + 10]); + g_assert(removed); + } + + c = '\0'; + removed = q_tree_remove(tree, &c); + g_assert(!removed); + + q_tree_foreach(tree, my_traverse, NULL); + + g_assert(q_tree_nnodes(tree) == strlen(chars2)); + g_assert(q_tree_height(tree) == 6); + + p = chars2; + q_tree_foreach(tree, check_order, &p); + + for (i = 25; i >= 0; i--) { + q_tree_insert(tree, &chars[i + 10], &chars[i + 10]); + } + + p = chars; + q_tree_foreach(tree, check_order, &p); + + c = '0'; + p = q_tree_lookup(tree, &c); + g_assert(p && *p == c); + g_assert(q_tree_lookup_extended(tree, &c, (gpointer *)&d, (gpointer *)&p)); + g_assert(c == *d && c == *p); + + c = 'A'; + p = q_tree_lookup(tree, &c); + g_assert(p && *p == c); + + c = 'a'; + p = q_tree_lookup(tree, &c); + g_assert(p && *p == c); + + c = 'z'; + p = q_tree_lookup(tree, &c); + g_assert(p && *p == c); + + c = '!'; + p = q_tree_lookup(tree, &c); + g_assert(p == NULL); + + c = '='; + p = q_tree_lookup(tree, &c); + g_assert(p == NULL); + + c = '|'; + p = q_tree_lookup(tree, &c); + g_assert(p == NULL); + + c = '0'; + p = q_tree_search(tree, my_search, &c); + g_assert(p && *p == c); + + c = 'A'; + p = q_tree_search(tree, my_search, &c); + g_assert(p && *p == c); + + c = 'a'; + p = q_tree_search(tree, my_search, &c); + g_assert(p && *p == c); + + c = 'z'; + p = q_tree_search(tree, my_search, &c); + g_assert(p && *p == c); + + c = '!'; + p = q_tree_search(tree, my_search, &c); + g_assert(p == NULL); + + c = '='; + p = q_tree_search(tree, my_search, &c); + g_assert(p == NULL); + + c = '|'; + p = q_tree_search(tree, my_search, &c); + g_assert(p == NULL); + + q_tree_destroy(tree); +} + +static void test_tree_remove(void) +{ + QTree *tree; + char c, d; + gint i; + gboolean removed; + + tree = q_tree_new_full((GCompareDataFunc)my_compare, NULL, + my_key_destroy, + my_value_destroy); + + for (i = 0; chars[i]; i++) { + q_tree_insert(tree, &chars[i], &chars[i]); + } + + c = '0'; + q_tree_insert(tree, &c, &c); + g_assert(destroyed_key == &c); + g_assert(destroyed_value == &chars[0]); + destroyed_key = NULL; + destroyed_value = NULL; + + d = '1'; + q_tree_replace(tree, &d, &d); + g_assert(destroyed_key == &chars[1]); + g_assert(destroyed_value == &chars[1]); + destroyed_key = NULL; + destroyed_value = NULL; + + c = '2'; + removed = q_tree_remove(tree, &c); + g_assert(removed); + g_assert(destroyed_key == &chars[2]); + g_assert(destroyed_value == &chars[2]); + destroyed_key = NULL; + destroyed_value = NULL; + + c = '3'; + removed = q_tree_steal(tree, &c); + g_assert(removed); + g_assert(destroyed_key == NULL); + g_assert(destroyed_value == NULL); + + const gchar *remove = "omkjigfedba"; + for (i = 0; remove[i]; i++) { + removed = q_tree_remove(tree, &remove[i]); + g_assert(removed); + } + + q_tree_destroy(tree); +} + +static void test_tree_destroy(void) +{ + QTree *tree; + gint i; + + tree = q_tree_new(my_compare); + + for (i = 0; chars[i]; i++) { + q_tree_insert(tree, &chars[i], &chars[i]); + } + + g_assert(q_tree_nnodes(tree) == strlen(chars)); + + g_test_message("nnodes: %d", q_tree_nnodes(tree)); + q_tree_ref(tree); + q_tree_destroy(tree); + + g_test_message("nnodes: %d", q_tree_nnodes(tree)); + g_assert(q_tree_nnodes(tree) == 0); + + q_tree_unref(tree); +} + +static void test_tree_insert(void) +{ + QTree *tree; + gchar *p; + gint i; + gchar *scrambled; + + tree = q_tree_new(my_compare); + + for (i = 0; chars[i]; i++) { + q_tree_insert(tree, &chars[i], &chars[i]); + } + p = chars; + q_tree_foreach(tree, check_order, &p); + + q_tree_unref(tree); + tree = q_tree_new(my_compare); + + for (i = strlen(chars) - 1; i >= 0; i--) { + q_tree_insert(tree, &chars[i], &chars[i]); + } + p = chars; + q_tree_foreach(tree, check_order, &p); + + q_tree_unref(tree); + tree = q_tree_new(my_compare); + + scrambled = g_strdup(chars); + + for (i = 0; i < 30; i++) { + gchar tmp; + gint a, b; + + a = g_random_int_range(0, strlen(scrambled)); + b = g_random_int_range(0, strlen(scrambled)); + tmp = scrambled[a]; + scrambled[a] = scrambled[b]; + scrambled[b] = tmp; + } + + for (i = 0; scrambled[i]; i++) { + q_tree_insert(tree, &scrambled[i], &scrambled[i]); + } + p = chars; + q_tree_foreach(tree, check_order, &p); + + g_free(scrambled); + q_tree_unref(tree); +} + +int main(int argc, char *argv[]) +{ + g_test_init(&argc, &argv, NULL); + + g_test_add_func("/qtree/search", test_tree_search); + g_test_add_func("/qtree/remove", test_tree_remove); + g_test_add_func("/qtree/destroy", test_tree_destroy); + g_test_add_func("/qtree/insert", test_tree_insert); + + return g_test_run(); +} diff --git a/util/qtree.c b/util/qtree.c new file mode 100644 index XXXXXXX..XXXXXXX --- /dev/null +++ b/util/qtree.c @@ -XXX,XX +XXX,XX @@ +/* + * GLIB - Library of useful routines for C programming + * Copyright (C) 1995-1997 Peter Mattis, Spencer Kimball and Josh MacDonald + * + * SPDX-License-Identifier: LGPL-2.1-or-later + * + * This library is free software; you can redistribute it and/or + * modify it under the terms of the GNU Lesser General Public + * License as published by the Free Software Foundation; either + * version 2.1 of the License, or (at your option) any later version. + * + * This library is distributed in the hope that it will be useful, + * but WITHOUT ANY WARRANTY; without even the implied warranty of + * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU + * Lesser General Public License for more details. + * + * You should have received a copy of the GNU Lesser General Public + * License along with this library; if not, see <http://www.gnu.org/licenses/>. + */ + +/* + * Modified by the GLib Team and others 1997-2000. See the AUTHORS + * file for a list of people on the GLib Team. See the ChangeLog + * files for a list of changes. These files are distributed with + * GLib at ftp://ftp.gtk.org/pub/gtk/. + */ + +/* + * MT safe + */ + +#include "qemu/osdep.h" +#include "qemu/qtree.h" + +/** + * SECTION:trees-binary + * @title: Balanced Binary Trees + * @short_description: a sorted collection of key/value pairs optimized + * for searching and traversing in order + * + * The #QTree structure and its associated functions provide a sorted + * collection of key/value pairs optimized for searching and traversing + * in order. This means that most of the operations (access, search, + * insertion, deletion, ...) on #QTree are O(log(n)) in average and O(n) + * in worst case for time complexity. But, note that maintaining a + * balanced sorted #QTree of n elements is done in time O(n log(n)). + * + * To create a new #QTree use q_tree_new(). + * + * To insert a key/value pair into a #QTree use q_tree_insert() + * (O(n log(n))). + * + * To remove a key/value pair use q_tree_remove() (O(n log(n))). + * + * To look up the value corresponding to a given key, use + * q_tree_lookup() and q_tree_lookup_extended(). + * + * To find out the number of nodes in a #QTree, use q_tree_nnodes(). To + * get the height of a #QTree, use q_tree_height(). + * + * To traverse a #QTree, calling a function for each node visited in + * the traversal, use q_tree_foreach(). + * + * To destroy a #QTree, use q_tree_destroy(). + **/ + +#define MAX_GTREE_HEIGHT 40 + +/** + * QTree: + * + * The QTree struct is an opaque data structure representing a + * [balanced binary tree][glib-Balanced-Binary-Trees]. It should be + * accessed only by using the following functions. + */ +struct _QTree { + QTreeNode *root; + GCompareDataFunc key_compare; + GDestroyNotify key_destroy_func; + GDestroyNotify value_destroy_func; + gpointer key_compare_data; + guint nnodes; + gint ref_count; +}; + +struct _QTreeNode { + gpointer key; /* key for this node */ + gpointer value; /* value stored at this node */ + QTreeNode *left; /* left subtree */ + QTreeNode *right; /* right subtree */ + gint8 balance; /* height (right) - height (left) */ + guint8 left_child; + guint8 right_child; +}; + + +static QTreeNode *q_tree_node_new(gpointer key, + gpointer value); +static QTreeNode *q_tree_insert_internal(QTree *tree, + gpointer key, + gpointer value, + gboolean replace); +static gboolean q_tree_remove_internal(QTree *tree, + gconstpointer key, + gboolean steal); +static QTreeNode *q_tree_node_balance(QTreeNode *node); +static QTreeNode *q_tree_find_node(QTree *tree, + gconstpointer key); +static QTreeNode *q_tree_node_search(QTreeNode *node, + GCompareFunc search_func, + gconstpointer data); +static QTreeNode *q_tree_node_rotate_left(QTreeNode *node); +static QTreeNode *q_tree_node_rotate_right(QTreeNode *node); +#ifdef Q_TREE_DEBUG +static void q_tree_node_check(QTreeNode *node); +#endif + +static QTreeNode* +q_tree_node_new(gpointer key, + gpointer value) +{ + QTreeNode *node = g_new(QTreeNode, 1); + + node->balance = 0; + node->left = NULL; + node->right = NULL; + node->left_child = FALSE; + node->right_child = FALSE; + node->key = key; + node->value = value; + + return node; +} + +/** + * q_tree_new: + * @key_compare_func: the function used to order the nodes in the #QTree. + * It should return values similar to the standard strcmp() function - + * 0 if the two arguments are equal, a negative value if the first argument + * comes before the second, or a positive value if the first argument comes + * after the second. + * + * Creates a new #QTree. + * + * Returns: a newly allocated #QTree + */ +QTree * +q_tree_new(GCompareFunc key_compare_func) +{ + g_return_val_if_fail(key_compare_func != NULL, NULL); + + return q_tree_new_full((GCompareDataFunc) key_compare_func, NULL, + NULL, NULL); +} + +/** + * q_tree_new_with_data: + * @key_compare_func: qsort()-style comparison function + * @key_compare_data: data to pass to comparison function + * + * Creates a new #QTree with a comparison function that accepts user data. + * See q_tree_new() for more details. + * + * Returns: a newly allocated #QTree + */ +QTree * +q_tree_new_with_data(GCompareDataFunc key_compare_func, + gpointer key_compare_data) +{ + g_return_val_if_fail(key_compare_func != NULL, NULL); + + return q_tree_new_full(key_compare_func, key_compare_data, + NULL, NULL); +} + +/** + * q_tree_new_full: + * @key_compare_func: qsort()-style comparison function + * @key_compare_data: data to pass to comparison function + * @key_destroy_func: a function to free the memory allocated for the key + * used when removing the entry from the #QTree or %NULL if you don't + * want to supply such a function + * @value_destroy_func: a function to free the memory allocated for the + * value used when removing the entry from the #QTree or %NULL if you + * don't want to supply such a function + * + * Creates a new #QTree like q_tree_new() and allows to specify functions + * to free the memory allocated for the key and value that get called when + * removing the entry from the #QTree. + * + * Returns: a newly allocated #QTree + */ +QTree * +q_tree_new_full(GCompareDataFunc key_compare_func, + gpointer key_compare_data, + GDestroyNotify key_destroy_func, + GDestroyNotify value_destroy_func) +{ + QTree *tree; + + g_return_val_if_fail(key_compare_func != NULL, NULL); + + tree = g_new(QTree, 1); + tree->root = NULL; + tree->key_compare = key_compare_func; + tree->key_destroy_func = key_destroy_func; + tree->value_destroy_func = value_destroy_func; + tree->key_compare_data = key_compare_data; + tree->nnodes = 0; + tree->ref_count = 1; + + return tree; +} + +/** + * q_tree_node_first: + * @tree: a #QTree + * + * Returns the first in-order node of the tree, or %NULL + * for an empty tree. + * + * Returns: (nullable) (transfer none): the first node in the tree + * + * Since: 2.68 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static QTreeNode * +q_tree_node_first(QTree *tree) +{ + QTreeNode *tmp; + + g_return_val_if_fail(tree != NULL, NULL); + + if (!tree->root) { + return NULL; + } + + tmp = tree->root; + + while (tmp->left_child) { + tmp = tmp->left; + } + + return tmp; +} + +/** + * q_tree_node_previous + * @node: a #QTree node + * + * Returns the previous in-order node of the tree, or %NULL + * if the passed node was already the first one. + * + * Returns: (nullable) (transfer none): the previous node in the tree + * + * Since: 2.68 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static QTreeNode * +q_tree_node_previous(QTreeNode *node) +{ + QTreeNode *tmp; + + g_return_val_if_fail(node != NULL, NULL); + + tmp = node->left; + + if (node->left_child) { + while (tmp->right_child) { + tmp = tmp->right; + } + } + + return tmp; +} + +/** + * q_tree_node_next + * @node: a #QTree node + * + * Returns the next in-order node of the tree, or %NULL + * if the passed node was already the last one. + * + * Returns: (nullable) (transfer none): the next node in the tree + * + * Since: 2.68 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static QTreeNode * +q_tree_node_next(QTreeNode *node) +{ + QTreeNode *tmp; + + g_return_val_if_fail(node != NULL, NULL); + + tmp = node->right; + + if (node->right_child) { + while (tmp->left_child) { + tmp = tmp->left; + } + } + + return tmp; +} + +/** + * q_tree_remove_all: + * @tree: a #QTree + * + * Removes all nodes from a #QTree and destroys their keys and values, + * then resets the #QTree’s root to %NULL. + * + * Since: 2.70 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static void +q_tree_remove_all(QTree *tree) +{ + QTreeNode *node; + QTreeNode *next; + + g_return_if_fail(tree != NULL); + + node = q_tree_node_first(tree); + + while (node) { + next = q_tree_node_next(node); + + if (tree->key_destroy_func) { + tree->key_destroy_func(node->key); + } + if (tree->value_destroy_func) { + tree->value_destroy_func(node->value); + } + g_free(node); + +#ifdef Q_TREE_DEBUG + g_assert(tree->nnodes > 0); + tree->nnodes--; +#endif + + node = next; + } + +#ifdef Q_TREE_DEBUG + g_assert(tree->nnodes == 0); +#endif + + tree->root = NULL; +#ifndef Q_TREE_DEBUG + tree->nnodes = 0; +#endif +} + +/** + * q_tree_ref: + * @tree: a #QTree + * + * Increments the reference count of @tree by one. + * + * It is safe to call this function from any thread. + * + * Returns: the passed in #QTree + * + * Since: 2.22 + */ +QTree * +q_tree_ref(QTree *tree) +{ + g_return_val_if_fail(tree != NULL, NULL); + + g_atomic_int_inc(&tree->ref_count); + + return tree; +} + +/** + * q_tree_unref: + * @tree: a #QTree + * + * Decrements the reference count of @tree by one. + * If the reference count drops to 0, all keys and values will + * be destroyed (if destroy functions were specified) and all + * memory allocated by @tree will be released. + * + * It is safe to call this function from any thread. + * + * Since: 2.22 + */ +void +q_tree_unref(QTree *tree) +{ + g_return_if_fail(tree != NULL); + + if (g_atomic_int_dec_and_test(&tree->ref_count)) { + q_tree_remove_all(tree); + g_free(tree); + } +} + +/** + * q_tree_destroy: + * @tree: a #QTree + * + * Removes all keys and values from the #QTree and decreases its + * reference count by one. If keys and/or values are dynamically + * allocated, you should either free them first or create the #QTree + * using q_tree_new_full(). In the latter case the destroy functions + * you supplied will be called on all keys and values before destroying + * the #QTree. + */ +void +q_tree_destroy(QTree *tree) +{ + g_return_if_fail(tree != NULL); + + q_tree_remove_all(tree); + q_tree_unref(tree); +} + +/** + * q_tree_insert_node: + * @tree: a #QTree + * @key: the key to insert + * @value: the value corresponding to the key + * + * Inserts a key/value pair into a #QTree. + * + * If the given key already exists in the #QTree its corresponding value + * is set to the new value. If you supplied a @value_destroy_func when + * creating the #QTree, the old value is freed using that function. If + * you supplied a @key_destroy_func when creating the #QTree, the passed + * key is freed using that function. + * + * The tree is automatically 'balanced' as new key/value pairs are added, + * so that the distance from the root to every leaf is as small as possible. + * The cost of maintaining a balanced tree while inserting new key/value + * result in a O(n log(n)) operation where most of the other operations + * are O(log(n)). + * + * Returns: (transfer none): the inserted (or set) node. + * + * Since: 2.68 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static QTreeNode * +q_tree_insert_node(QTree *tree, + gpointer key, + gpointer value) +{ + QTreeNode *node; + + g_return_val_if_fail(tree != NULL, NULL); + + node = q_tree_insert_internal(tree, key, value, FALSE); + +#ifdef Q_TREE_DEBUG + q_tree_node_check(tree->root); +#endif + + return node; +} + +/** + * q_tree_insert: + * @tree: a #QTree + * @key: the key to insert + * @value: the value corresponding to the key + * + * Inserts a key/value pair into a #QTree. + * + * Inserts a new key and value into a #QTree as q_tree_insert_node() does, + * only this function does not return the inserted or set node. + */ +void +q_tree_insert(QTree *tree, + gpointer key, + gpointer value) +{ + q_tree_insert_node(tree, key, value); +} + +/** + * q_tree_replace_node: + * @tree: a #QTree + * @key: the key to insert + * @value: the value corresponding to the key + * + * Inserts a new key and value into a #QTree similar to q_tree_insert_node(). + * The difference is that if the key already exists in the #QTree, it gets + * replaced by the new key. If you supplied a @value_destroy_func when + * creating the #QTree, the old value is freed using that function. If you + * supplied a @key_destroy_func when creating the #QTree, the old key is + * freed using that function. + * + * The tree is automatically 'balanced' as new key/value pairs are added, + * so that the distance from the root to every leaf is as small as possible. + * + * Returns: (transfer none): the inserted (or set) node. + * + * Since: 2.68 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static QTreeNode * +q_tree_replace_node(QTree *tree, + gpointer key, + gpointer value) +{ + QTreeNode *node; + + g_return_val_if_fail(tree != NULL, NULL); + + node = q_tree_insert_internal(tree, key, value, TRUE); + +#ifdef Q_TREE_DEBUG + q_tree_node_check(tree->root); +#endif + + return node; +} + +/** + * q_tree_replace: + * @tree: a #QTree + * @key: the key to insert + * @value: the value corresponding to the key + * + * Inserts a new key and value into a #QTree as q_tree_replace_node() does, + * only this function does not return the inserted or set node. + */ +void +q_tree_replace(QTree *tree, + gpointer key, + gpointer value) +{ + q_tree_replace_node(tree, key, value); +} + +/* internal insert routine */ +static QTreeNode * +q_tree_insert_internal(QTree *tree, + gpointer key, + gpointer value, + gboolean replace) +{ + QTreeNode *node, *retnode; + QTreeNode *path[MAX_GTREE_HEIGHT]; + int idx; + + g_return_val_if_fail(tree != NULL, NULL); + + if (!tree->root) { + tree->root = q_tree_node_new(key, value); + tree->nnodes++; + return tree->root; + } + + idx = 0; + path[idx++] = NULL; + node = tree->root; + + while (1) { + int cmp = tree->key_compare(key, node->key, tree->key_compare_data); + + if (cmp == 0) { + if (tree->value_destroy_func) { + tree->value_destroy_func(node->value); + } + + node->value = value; + + if (replace) { + if (tree->key_destroy_func) { + tree->key_destroy_func(node->key); + } + + node->key = key; + } else { + /* free the passed key */ + if (tree->key_destroy_func) { + tree->key_destroy_func(key); + } + } + + return node; + } else if (cmp < 0) { + if (node->left_child) { + path[idx++] = node; + node = node->left; + } else { + QTreeNode *child = q_tree_node_new(key, value); + + child->left = node->left; + child->right = node; + node->left = child; + node->left_child = TRUE; + node->balance -= 1; + + tree->nnodes++; + + retnode = child; + break; + } + } else { + if (node->right_child) { + path[idx++] = node; + node = node->right; + } else { + QTreeNode *child = q_tree_node_new(key, value); + + child->right = node->right; + child->left = node; + node->right = child; + node->right_child = TRUE; + node->balance += 1; + + tree->nnodes++; + + retnode = child; + break; + } + } + } + + /* + * Restore balance. This is the goodness of a non-recursive + * implementation, when we are done with balancing we 'break' + * the loop and we are done. + */ + while (1) { + QTreeNode *bparent = path[--idx]; + gboolean left_node = (bparent && node == bparent->left); + g_assert(!bparent || bparent->left == node || bparent->right == node); + + if (node->balance < -1 || node->balance > 1) { + node = q_tree_node_balance(node); + if (bparent == NULL) { + tree->root = node; + } else if (left_node) { + bparent->left = node; + } else { + bparent->right = node; + } + } + + if (node->balance == 0 || bparent == NULL) { + break; + } + + if (left_node) { + bparent->balance -= 1; + } else { + bparent->balance += 1; + } + + node = bparent; + } + + return retnode; +} + +/** + * q_tree_remove: + * @tree: a #QTree + * @key: the key to remove + * + * Removes a key/value pair from a #QTree. + * + * If the #QTree was created using q_tree_new_full(), the key and value + * are freed using the supplied destroy functions, otherwise you have to + * make sure that any dynamically allocated values are freed yourself. + * If the key does not exist in the #QTree, the function does nothing. + * + * The cost of maintaining a balanced tree while removing a key/value + * result in a O(n log(n)) operation where most of the other operations + * are O(log(n)). + * + * Returns: %TRUE if the key was found (prior to 2.8, this function + * returned nothing) + */ +gboolean +q_tree_remove(QTree *tree, + gconstpointer key) +{ + gboolean removed; + + g_return_val_if_fail(tree != NULL, FALSE); + + removed = q_tree_remove_internal(tree, key, FALSE); + +#ifdef Q_TREE_DEBUG + q_tree_node_check(tree->root); +#endif + + return removed; +} + +/** + * q_tree_steal: + * @tree: a #QTree + * @key: the key to remove + * + * Removes a key and its associated value from a #QTree without calling + * the key and value destroy functions. + * + * If the key does not exist in the #QTree, the function does nothing. + * + * Returns: %TRUE if the key was found (prior to 2.8, this function + * returned nothing) + */ +gboolean +q_tree_steal(QTree *tree, + gconstpointer key) +{ + gboolean removed; + + g_return_val_if_fail(tree != NULL, FALSE); + + removed = q_tree_remove_internal(tree, key, TRUE); + +#ifdef Q_TREE_DEBUG + q_tree_node_check(tree->root); +#endif + + return removed; +} + +/* internal remove routine */ +static gboolean +q_tree_remove_internal(QTree *tree, + gconstpointer key, + gboolean steal) +{ + QTreeNode *node, *parent, *balance; + QTreeNode *path[MAX_GTREE_HEIGHT]; + int idx; + gboolean left_node; + + g_return_val_if_fail(tree != NULL, FALSE); + + if (!tree->root) { + return FALSE; + } + + idx = 0; + path[idx++] = NULL; + node = tree->root; + + while (1) { + int cmp = tree->key_compare(key, node->key, tree->key_compare_data); + + if (cmp == 0) { + break; + } else if (cmp < 0) { + if (!node->left_child) { + return FALSE; + } + + path[idx++] = node; + node = node->left; + } else { + if (!node->right_child) { + return FALSE; + } + + path[idx++] = node; + node = node->right; + } + } + + /* + * The following code is almost equal to q_tree_remove_node, + * except that we do not have to call q_tree_node_parent. + */ + balance = parent = path[--idx]; + g_assert(!parent || parent->left == node || parent->right == node); + left_node = (parent && node == parent->left); + + if (!node->left_child) { + if (!node->right_child) { + if (!parent) { + tree->root = NULL; + } else if (left_node) { + parent->left_child = FALSE; + parent->left = node->left; + parent->balance += 1; + } else { + parent->right_child = FALSE; + parent->right = node->right; + parent->balance -= 1; + } + } else { + /* node has a right child */ + QTreeNode *tmp = q_tree_node_next(node); + tmp->left = node->left; + + if (!parent) { + tree->root = node->right; + } else if (left_node) { + parent->left = node->right; + parent->balance += 1; + } else { + parent->right = node->right; + parent->balance -= 1; + } + } + } else { + /* node has a left child */ + if (!node->right_child) { + QTreeNode *tmp = q_tree_node_previous(node); + tmp->right = node->right; + + if (parent == NULL) { + tree->root = node->left; + } else if (left_node) { + parent->left = node->left; + parent->balance += 1; + } else { + parent->right = node->left; + parent->balance -= 1; + } + } else { + /* node has a both children (pant, pant!) */ + QTreeNode *prev = node->left; + QTreeNode *next = node->right; + QTreeNode *nextp = node; + int old_idx = idx + 1; + idx++; + + /* path[idx] == parent */ + /* find the immediately next node (and its parent) */ + while (next->left_child) { + path[++idx] = nextp = next; + next = next->left; + } + + path[old_idx] = next; + balance = path[idx]; + + /* remove 'next' from the tree */ + if (nextp != node) { + if (next->right_child) { + nextp->left = next->right; + } else { + nextp->left_child = FALSE; + } + nextp->balance += 1; + + next->right_child = TRUE; + next->right = node->right; + } else { + node->balance -= 1; + } + + /* set the prev to point to the right place */ + while (prev->right_child) { + prev = prev->right; + } + prev->right = next; + + /* prepare 'next' to replace 'node' */ + next->left_child = TRUE; + next->left = node->left; + next->balance = node->balance; + + if (!parent) { + tree->root = next; + } else if (left_node) { + parent->left = next; + } else { + parent->right = next; + } + } + } + + /* restore balance */ + if (balance) { + while (1) { + QTreeNode *bparent = path[--idx]; + g_assert(!bparent || + bparent->left == balance || + bparent->right == balance); + left_node = (bparent && balance == bparent->left); + + if (balance->balance < -1 || balance->balance > 1) { + balance = q_tree_node_balance(balance); + if (!bparent) { + tree->root = balance; + } else if (left_node) { + bparent->left = balance; + } else { + bparent->right = balance; + } + } + + if (balance->balance != 0 || !bparent) { + break; + } + + if (left_node) { + bparent->balance += 1; + } else { + bparent->balance -= 1; + } + + balance = bparent; + } + } + + if (!steal) { + if (tree->key_destroy_func) { + tree->key_destroy_func(node->key); + } + if (tree->value_destroy_func) { + tree->value_destroy_func(node->value); + } + } + + g_free(node); + + tree->nnodes--; + + return TRUE; +} + +/** + * q_tree_lookup_node: + * @tree: a #QTree + * @key: the key to look up + * + * Gets the tree node corresponding to the given key. Since a #QTree is + * automatically balanced as key/value pairs are added, key lookup + * is O(log n) (where n is the number of key/value pairs in the tree). + * + * Returns: (nullable) (transfer none): the tree node corresponding to + * the key, or %NULL if the key was not found + * + * Since: 2.68 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static QTreeNode * +q_tree_lookup_node(QTree *tree, + gconstpointer key) +{ + g_return_val_if_fail(tree != NULL, NULL); + + return q_tree_find_node(tree, key); +} + +/** + * q_tree_lookup: + * @tree: a #QTree + * @key: the key to look up + * + * Gets the value corresponding to the given key. Since a #QTree is + * automatically balanced as key/value pairs are added, key lookup + * is O(log n) (where n is the number of key/value pairs in the tree). + * + * Returns: the value corresponding to the key, or %NULL + * if the key was not found + */ +gpointer +q_tree_lookup(QTree *tree, + gconstpointer key) +{ + QTreeNode *node; + + node = q_tree_lookup_node(tree, key); + + return node ? node->value : NULL; +} + +/** + * q_tree_lookup_extended: + * @tree: a #QTree + * @lookup_key: the key to look up + * @orig_key: (out) (optional) (nullable): returns the original key + * @value: (out) (optional) (nullable): returns the value associated with + * the key + * + * Looks up a key in the #QTree, returning the original key and the + * associated value. This is useful if you need to free the memory + * allocated for the original key, for example before calling + * q_tree_remove(). + * + * Returns: %TRUE if the key was found in the #QTree + */ +gboolean +q_tree_lookup_extended(QTree *tree, + gconstpointer lookup_key, + gpointer *orig_key, + gpointer *value) +{ + QTreeNode *node; + + g_return_val_if_fail(tree != NULL, FALSE); + + node = q_tree_find_node(tree, lookup_key); + + if (node) { + if (orig_key) { + *orig_key = node->key; + } + if (value) { + *value = node->value; + } + return TRUE; + } else { + return FALSE; + } +} + +/** + * q_tree_foreach: + * @tree: a #QTree + * @func: the function to call for each node visited. + * If this function returns %TRUE, the traversal is stopped. + * @user_data: user data to pass to the function + * + * Calls the given function for each of the key/value pairs in the #QTree. + * The function is passed the key and value of each pair, and the given + * @data parameter. The tree is traversed in sorted order. + * + * The tree may not be modified while iterating over it (you can't + * add/remove items). To remove all items matching a predicate, you need + * to add each item to a list in your #GTraverseFunc as you walk over + * the tree, then walk the list and remove each item. + */ +void +q_tree_foreach(QTree *tree, + GTraverseFunc func, + gpointer user_data) +{ + QTreeNode *node; + + g_return_if_fail(tree != NULL); + + if (!tree->root) { + return; + } + + node = q_tree_node_first(tree); + + while (node) { + if ((*func)(node->key, node->value, user_data)) { + break; + } + + node = q_tree_node_next(node); + } +} + +/** + * q_tree_search_node: + * @tree: a #QTree + * @search_func: a function used to search the #QTree + * @user_data: the data passed as the second argument to @search_func + * + * Searches a #QTree using @search_func. + * + * The @search_func is called with a pointer to the key of a key/value + * pair in the tree, and the passed in @user_data. If @search_func returns + * 0 for a key/value pair, then the corresponding node is returned as + * the result of q_tree_search(). If @search_func returns -1, searching + * will proceed among the key/value pairs that have a smaller key; if + * @search_func returns 1, searching will proceed among the key/value + * pairs that have a larger key. + * + * Returns: (nullable) (transfer none): the node corresponding to the + * found key, or %NULL if the key was not found + * + * Since: 2.68 in GLib. Internal in Qtree, i.e. not in the public API. + */ +static QTreeNode * +q_tree_search_node(QTree *tree, + GCompareFunc search_func, + gconstpointer user_data) +{ + g_return_val_if_fail(tree != NULL, NULL); + + if (!tree->root) { + return NULL; + } + + return q_tree_node_search(tree->root, search_func, user_data); +} + +/** + * q_tree_search: + * @tree: a #QTree + * @search_func: a function used to search the #QTree + * @user_data: the data passed as the second argument to @search_func + * + * Searches a #QTree using @search_func. + * + * The @search_func is called with a pointer to the key of a key/value + * pair in the tree, and the passed in @user_data. If @search_func returns + * 0 for a key/value pair, then the corresponding value is returned as + * the result of q_tree_search(). If @search_func returns -1, searching + * will proceed among the key/value pairs that have a smaller key; if + * @search_func returns 1, searching will proceed among the key/value + * pairs that have a larger key. + * + * Returns: the value corresponding to the found key, or %NULL + * if the key was not found + */ +gpointer +q_tree_search(QTree *tree, + GCompareFunc search_func, + gconstpointer user_data) +{ + QTreeNode *node; + + node = q_tree_search_node(tree, search_func, user_data); + + return node ? node->value : NULL; +} + +/** + * q_tree_height: + * @tree: a #QTree + * + * Gets the height of a #QTree. + * + * If the #QTree contains no nodes, the height is 0. + * If the #QTree contains only one root node the height is 1. + * If the root node has children the height is 2, etc. + * + * Returns: the height of @tree + */ +gint +q_tree_height(QTree *tree) +{ + QTreeNode *node; + gint height; + + g_return_val_if_fail(tree != NULL, 0); + + if (!tree->root) { + return 0; + } + + height = 0; + node = tree->root; + + while (1) { + height += 1 + MAX(node->balance, 0); + + if (!node->left_child) { + return height; + } + + node = node->left; + } +} + +/** + * q_tree_nnodes: + * @tree: a #QTree + * + * Gets the number of nodes in a #QTree. + * + * Returns: the number of nodes in @tree + */ +gint +q_tree_nnodes(QTree *tree) +{ + g_return_val_if_fail(tree != NULL, 0); + + return tree->nnodes; +} + +static QTreeNode * +q_tree_node_balance(QTreeNode *node) +{ + if (node->balance < -1) { + if (node->left->balance > 0) { + node->left = q_tree_node_rotate_left(node->left); + } + node = q_tree_node_rotate_right(node); + } else if (node->balance > 1) { + if (node->right->balance < 0) { + node->right = q_tree_node_rotate_right(node->right); + } + node = q_tree_node_rotate_left(node); + } + + return node; +} + +static QTreeNode * +q_tree_find_node(QTree *tree, + gconstpointer key) +{ + QTreeNode *node; + gint cmp; + + node = tree->root; + if (!node) { + return NULL; + } + + while (1) { + cmp = tree->key_compare(key, node->key, tree->key_compare_data); + if (cmp == 0) { + return node; + } else if (cmp < 0) { + if (!node->left_child) { + return NULL; + } + + node = node->left; + } else { + if (!node->right_child) { + return NULL; + } + + node = node->right; + } + } +} + +static QTreeNode * +q_tree_node_search(QTreeNode *node, + GCompareFunc search_func, + gconstpointer data) +{ + gint dir; + + if (!node) { + return NULL; + } + + while (1) { + dir = (*search_func)(node->key, data); + if (dir == 0) { + return node; + } else if (dir < 0) { + if (!node->left_child) { + return NULL; + } + + node = node->left; + } else { + if (!node->right_child) { + return NULL; + } + + node = node->right; + } + } +} + +static QTreeNode * +q_tree_node_rotate_left(QTreeNode *node) +{ + QTreeNode *right; + gint a_bal; + gint b_bal; + + right = node->right; + + if (right->left_child) { + node->right = right->left; + } else { + node->right_child = FALSE; + right->left_child = TRUE; + } + right->left = node; + + a_bal = node->balance; + b_bal = right->balance; + + if (b_bal <= 0) { + if (a_bal >= 1) { + right->balance = b_bal - 1; + } else { + right->balance = a_bal + b_bal - 2; + } + node->balance = a_bal - 1; + } else { + if (a_bal <= b_bal) { + right->balance = a_bal - 2; + } else { + right->balance = b_bal - 1; + } + node->balance = a_bal - b_bal - 1; + } + + return right; +} + +static QTreeNode * +q_tree_node_rotate_right(QTreeNode *node) +{ + QTreeNode *left; + gint a_bal; + gint b_bal; + + left = node->left; + + if (left->right_child) { + node->left = left->right; + } else { + node->left_child = FALSE; + left->right_child = TRUE; + } + left->right = node; + + a_bal = node->balance; + b_bal = left->balance; + + if (b_bal <= 0) { + if (b_bal > a_bal) { + left->balance = b_bal + 1; + } else { + left->balance = a_bal + 2; + } + node->balance = a_bal - b_bal + 1; + } else { + if (a_bal <= -1) { + left->balance = b_bal + 1; + } else { + left->balance = a_bal + b_bal + 2; + } + node->balance = a_bal + 1; + } + + return left; +} + +#ifdef Q_TREE_DEBUG +static gint +q_tree_node_height(QTreeNode *node) +{ + gint left_height; + gint right_height; + + if (node) { + left_height = 0; + right_height = 0; + + if (node->left_child) { + left_height = q_tree_node_height(node->left); + } + + if (node->right_child) { + right_height = q_tree_node_height(node->right); + } + + return MAX(left_height, right_height) + 1; + } + + return 0; +} + +static void q_tree_node_check(QTreeNode *node) +{ + gint left_height; + gint right_height; + gint balance; + QTreeNode *tmp; + + if (node) { + if (node->left_child) { + tmp = q_tree_node_previous(node); + g_assert(tmp->right == node); + } + + if (node->right_child) { + tmp = q_tree_node_next(node); + g_assert(tmp->left == node); + } + + left_height = 0; + right_height = 0; + + if (node->left_child) { + left_height = q_tree_node_height(node->left); + } + if (node->right_child) { + right_height = q_tree_node_height(node->right); + } + + balance = right_height - left_height; + g_assert(balance == node->balance); + + if (node->left_child) { + q_tree_node_check(node->left); + } + if (node->right_child) { + q_tree_node_check(node->right); + } + } +} +#endif diff --git a/tests/bench/meson.build b/tests/bench/meson.build index XXXXXXX..XXXXXXX 100644 --- a/tests/bench/meson.build +++ b/tests/bench/meson.build @@ -XXX,XX +XXX,XX @@ xbzrle_bench = executable('xbzrle-bench', dependencies: [qemuutil,migration]) endif +qtree_bench = executable('qtree-bench', + sources: 'qtree-bench.c', + dependencies: [qemuutil]) + executable('atomic_add-bench', sources: files('atomic_add-bench.c'), dependencies: [qemuutil], diff --git a/tests/unit/meson.build b/tests/unit/meson.build index XXXXXXX..XXXXXXX 100644 --- a/tests/unit/meson.build +++ b/tests/unit/meson.build @@ -XXX,XX +XXX,XX @@ tests = { 'test-rcu-slist': [], 'test-qdist': [], 'test-qht': [], + 'test-qtree': [], 'test-bitops': [], 'test-bitcnt': [], 'test-qgraph': ['../qtest/libqos/qgraph.c'], diff --git a/util/meson.build b/util/meson.build index XXXXXXX..XXXXXXX 100644 --- a/util/meson.build +++ b/util/meson.build @@ -XXX,XX +XXX,XX @@ util_ss.add(when: 'CONFIG_WIN32', if_true: files('oslib-win32.c')) util_ss.add(when: 'CONFIG_WIN32', if_true: files('qemu-thread-win32.c')) util_ss.add(when: 'CONFIG_WIN32', if_true: winmm) util_ss.add(when: 'CONFIG_WIN32', if_true: pathcch) +util_ss.add(when: 'HAVE_GLIB_WITH_SLICE_ALLOCATOR', if_true: files('qtree.c')) util_ss.add(files('envlist.c', 'path.c', 'module.c')) util_ss.add(files('host-utils.c')) util_ss.add(files('bitmap.c', 'bitops.c')) -- 2.34.1 From: Emilio Cota <cota@braap.org> qemu-user can hang in a multi-threaded fork. One common reason is that when creating a TB, between fork and exec we manipulate a GTree whose memory allocator (GSlice) is not fork-safe. Although POSIX does not mandate it, the system's allocator (e.g. tcmalloc, libc malloc) is probably fork-safe. Fix some of these hangs by using QTree, which uses the system's allocator regardless of the Glib version that we used at configuration time. Tested with the test program in the original bug report, i.e.: ``` void garble() { int pid = fork(); if (pid == 0) { exit(0); } else { int wstatus; waitpid(pid, &wstatus, 0); } } void supragarble(unsigned depth) { if (depth == 0) return ; std::thread a(supragarble, depth-1); std::thread b(supragarble, depth-1); garble(); a.join(); b.join(); } int main() { supragarble(10); } ``` Resolves: https://gitlab.com/qemu-project/qemu/-/issues/285 Reported-by: Valentin David <me@valentindavid.com> Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Emilio Cota <cota@braap.org> Message-Id: <20230205163758.416992-3-cota@braap.org> [rth: Add QEMU_DISABLE_CFI for all callback using functions.] Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- accel/tcg/tb-maint.c | 17 +++++++++-------- tcg/region.c | 19 ++++++++++--------- util/qtree.c | 8 ++++---- 3 files changed, 23 insertions(+), 21 deletions(-) diff --git a/accel/tcg/tb-maint.c b/accel/tcg/tb-maint.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/tb-maint.c +++ b/accel/tcg/tb-maint.c @@ -XXX,XX +XXX,XX @@ #include "qemu/osdep.h" #include "qemu/interval-tree.h" +#include "qemu/qtree.h" #include "exec/cputlb.h" #include "exec/log.h" #include "exec/exec-all.h" @@ -XXX,XX +XXX,XX @@ struct page_entry { * See also: page_collection_lock(). */ struct page_collection { - GTree *tree; + QTree *tree; struct page_entry *max; }; @@ -XXX,XX +XXX,XX @@ static bool page_trylock_add(struct page_collection *set, tb_page_addr_t addr) struct page_entry *pe; PageDesc *pd; - pe = g_tree_lookup(set->tree, &index); + pe = q_tree_lookup(set->tree, &index); if (pe) { return false; } @@ -XXX,XX +XXX,XX @@ static bool page_trylock_add(struct page_collection *set, tb_page_addr_t addr) } pe = page_entry_new(pd, index); - g_tree_insert(set->tree, &pe->index, pe); + q_tree_insert(set->tree, &pe->index, pe); /* * If this is either (1) the first insertion or (2) a page whose index @@ -XXX,XX +XXX,XX @@ static struct page_collection *page_collection_lock(tb_page_addr_t start, end >>= TARGET_PAGE_BITS; g_assert(start <= end); - set->tree = g_tree_new_full(tb_page_addr_cmp, NULL, NULL, + set->tree = q_tree_new_full(tb_page_addr_cmp, NULL, NULL, page_entry_destroy); set->max = NULL; assert_no_pages_locked(); retry: - g_tree_foreach(set->tree, page_entry_lock, NULL); + q_tree_foreach(set->tree, page_entry_lock, NULL); for (index = start; index <= end; index++) { TranslationBlock *tb; @@ -XXX,XX +XXX,XX @@ static struct page_collection *page_collection_lock(tb_page_addr_t start, continue; } if (page_trylock_add(set, index << TARGET_PAGE_BITS)) { - g_tree_foreach(set->tree, page_entry_unlock, NULL); + q_tree_foreach(set->tree, page_entry_unlock, NULL); goto retry; } assert_page_locked(pd); @@ -XXX,XX +XXX,XX @@ static struct page_collection *page_collection_lock(tb_page_addr_t start, (tb_page_addr1(tb) != -1 && page_trylock_add(set, tb_page_addr1(tb)))) { /* drop all locks, and reacquire in order */ - g_tree_foreach(set->tree, page_entry_unlock, NULL); + q_tree_foreach(set->tree, page_entry_unlock, NULL); goto retry; } } @@ -XXX,XX +XXX,XX @@ static struct page_collection *page_collection_lock(tb_page_addr_t start, static void page_collection_unlock(struct page_collection *set) { /* entries are unlocked and freed via page_entry_destroy */ - g_tree_destroy(set->tree); + q_tree_destroy(set->tree); g_free(set); } diff --git a/tcg/region.c b/tcg/region.c index XXXXXXX..XXXXXXX 100644 --- a/tcg/region.c +++ b/tcg/region.c @@ -XXX,XX +XXX,XX @@ #include "qemu/mprotect.h" #include "qemu/memalign.h" #include "qemu/cacheinfo.h" +#include "qemu/qtree.h" #include "qapi/error.h" #include "exec/exec-all.h" #include "tcg/tcg.h" @@ -XXX,XX +XXX,XX @@ struct tcg_region_tree { QemuMutex lock; - GTree *tree; + QTree *tree; /* padding to avoid false sharing is computed at run-time */ }; @@ -XXX,XX +XXX,XX @@ static void tcg_region_trees_init(void) struct tcg_region_tree *rt = region_trees + i * tree_size; qemu_mutex_init(&rt->lock); - rt->tree = g_tree_new_full(tb_tc_cmp, NULL, NULL, tb_destroy); + rt->tree = q_tree_new_full(tb_tc_cmp, NULL, NULL, tb_destroy); } } @@ -XXX,XX +XXX,XX @@ void tcg_tb_insert(TranslationBlock *tb) g_assert(rt != NULL); qemu_mutex_lock(&rt->lock); - g_tree_insert(rt->tree, &tb->tc, tb); + q_tree_insert(rt->tree, &tb->tc, tb); qemu_mutex_unlock(&rt->lock); } @@ -XXX,XX +XXX,XX @@ void tcg_tb_remove(TranslationBlock *tb) g_assert(rt != NULL); qemu_mutex_lock(&rt->lock); - g_tree_remove(rt->tree, &tb->tc); + q_tree_remove(rt->tree, &tb->tc); qemu_mutex_unlock(&rt->lock); } @@ -XXX,XX +XXX,XX @@ TranslationBlock *tcg_tb_lookup(uintptr_t tc_ptr) } qemu_mutex_lock(&rt->lock); - tb = g_tree_lookup(rt->tree, &s); + tb = q_tree_lookup(rt->tree, &s); qemu_mutex_unlock(&rt->lock); return tb; } @@ -XXX,XX +XXX,XX @@ void tcg_tb_foreach(GTraverseFunc func, gpointer user_data) for (i = 0; i < region.n; i++) { struct tcg_region_tree *rt = region_trees + i * tree_size; - g_tree_foreach(rt->tree, func, user_data); + q_tree_foreach(rt->tree, func, user_data); } tcg_region_tree_unlock_all(); } @@ -XXX,XX +XXX,XX @@ size_t tcg_nb_tbs(void) for (i = 0; i < region.n; i++) { struct tcg_region_tree *rt = region_trees + i * tree_size; - nb_tbs += g_tree_nnodes(rt->tree); + nb_tbs += q_tree_nnodes(rt->tree); } tcg_region_tree_unlock_all(); return nb_tbs; @@ -XXX,XX +XXX,XX @@ static void tcg_region_tree_reset_all(void) struct tcg_region_tree *rt = region_trees + i * tree_size; /* Increment the refcount first so that destroy acts as a reset */ - g_tree_ref(rt->tree); - g_tree_destroy(rt->tree); + q_tree_ref(rt->tree); + q_tree_destroy(rt->tree); } tcg_region_tree_unlock_all(); } diff --git a/util/qtree.c b/util/qtree.c index XXXXXXX..XXXXXXX 100644 --- a/util/qtree.c +++ b/util/qtree.c @@ -XXX,XX +XXX,XX @@ q_tree_node_next(QTreeNode *node) * * Since: 2.70 in GLib. Internal in Qtree, i.e. not in the public API. */ -static void +static void QEMU_DISABLE_CFI q_tree_remove_all(QTree *tree) { QTreeNode *node; @@ -XXX,XX +XXX,XX @@ q_tree_replace(QTree *tree, } /* internal insert routine */ -static QTreeNode * +static QTreeNode * QEMU_DISABLE_CFI q_tree_insert_internal(QTree *tree, gpointer key, gpointer value, @@ -XXX,XX +XXX,XX @@ q_tree_steal(QTree *tree, } /* internal remove routine */ -static gboolean +static gboolean QEMU_DISABLE_CFI q_tree_remove_internal(QTree *tree, gconstpointer key, gboolean steal) @@ -XXX,XX +XXX,XX @@ q_tree_node_balance(QTreeNode *node) return node; } -static QTreeNode * +static QTreeNode * QEMU_DISABLE_CFI q_tree_find_node(QTree *tree, gconstpointer key) { -- 2.34.1 We have been enforcing host page alignment for the non-R fallback of MAX_RESERVED_VA, but failing to enforce for -R. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- linux-user/main.c | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/linux-user/main.c b/linux-user/main.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/main.c +++ b/linux-user/main.c @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp) */ max_reserved_va = MAX_RESERVED_VA(cpu); if (reserved_va != 0) { + if (reserved_va % qemu_host_page_size) { + char *s = size_to_str(qemu_host_page_size); + fprintf(stderr, "Reserved virtual address not aligned mod %s\n", s); + g_free(s); + exit(EXIT_FAILURE); + } if (max_reserved_va && reserved_va > max_reserved_va) { fprintf(stderr, "Reserved virtual address too big\n"); exit(EXIT_FAILURE); -- 2.34.1 Pass the address of the last byte to be changed, rather than the first address past the last byte. This avoids overflow when the last page of the address space is involved. Resolves: https://gitlab.com/qemu-project/qemu/-/issues/1528 Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/exec/cpu-all.h | 2 +- accel/tcg/user-exec.c | 16 +++++++--------- bsd-user/mmap.c | 6 +++--- linux-user/elfload.c | 11 ++++++----- linux-user/mmap.c | 16 ++++++++-------- linux-user/syscall.c | 4 ++-- 6 files changed, 27 insertions(+), 28 deletions(-) diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h index XXXXXXX..XXXXXXX 100644 --- a/include/exec/cpu-all.h +++ b/include/exec/cpu-all.h @@ -XXX,XX +XXX,XX @@ typedef int (*walk_memory_regions_fn)(void *, target_ulong, int walk_memory_regions(void *, walk_memory_regions_fn); int page_get_flags(target_ulong address); -void page_set_flags(target_ulong start, target_ulong end, int flags); +void page_set_flags(target_ulong start, target_ulong last, int flags); void page_reset_target_data(target_ulong start, target_ulong end); int page_check_range(target_ulong start, target_ulong len, int flags); diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/user-exec.c +++ b/accel/tcg/user-exec.c @@ -XXX,XX +XXX,XX @@ static bool pageflags_set_clear(target_ulong start, target_ulong last, * The flag PAGE_WRITE_ORG is positioned automatically depending * on PAGE_WRITE. The mmap_lock should already be held. */ -void page_set_flags(target_ulong start, target_ulong end, int flags) +void page_set_flags(target_ulong start, target_ulong last, int flags) { - target_ulong last; bool reset = false; bool inval_tb = false; /* This function should never be called with addresses outside the guest address space. If this assert fires, it probably indicates a missing call to h2g_valid. */ - assert(start < end); - assert(end - 1 <= GUEST_ADDR_MAX); + assert(start <= last); + assert(last <= GUEST_ADDR_MAX); /* Only set PAGE_ANON with new mappings. */ assert(!(flags & PAGE_ANON) || (flags & PAGE_RESET)); assert_memory_lock(); - start = start & TARGET_PAGE_MASK; - end = TARGET_PAGE_ALIGN(end); - last = end - 1; + start &= TARGET_PAGE_MASK; + last |= ~TARGET_PAGE_MASK; if (!(flags & PAGE_VALID)) { flags = 0; @@ -XXX,XX +XXX,XX @@ void page_set_flags(target_ulong start, target_ulong end, int flags) } if (!flags || reset) { - page_reset_target_data(start, end); + page_reset_target_data(start, last + 1); inval_tb |= pageflags_unset(start, last); } if (flags) { @@ -XXX,XX +XXX,XX @@ void page_set_flags(target_ulong start, target_ulong end, int flags) ~(reset ? 0 : PAGE_STICKY)); } if (inval_tb) { - tb_invalidate_phys_range(start, end); + tb_invalidate_phys_range(start, last + 1); } } diff --git a/bsd-user/mmap.c b/bsd-user/mmap.c index XXXXXXX..XXXXXXX 100644 --- a/bsd-user/mmap.c +++ b/bsd-user/mmap.c @@ -XXX,XX +XXX,XX @@ int target_mprotect(abi_ulong start, abi_ulong len, int prot) if (ret != 0) goto error; } - page_set_flags(start, start + len, prot | PAGE_VALID); + page_set_flags(start, start + len - 1, prot | PAGE_VALID); mmap_unlock(); return 0; error: @@ -XXX,XX +XXX,XX @@ abi_long target_mmap(abi_ulong start, abi_ulong len, int prot, } } the_end1: - page_set_flags(start, start + len, prot | PAGE_VALID); + page_set_flags(start, start + len - 1, prot | PAGE_VALID); the_end: #ifdef DEBUG_MMAP printf("ret=0x" TARGET_ABI_FMT_lx "\n", start); @@ -XXX,XX +XXX,XX @@ int target_munmap(abi_ulong start, abi_ulong len) } if (ret == 0) { - page_set_flags(start, start + len, 0); + page_set_flags(start, start + len - 1, 0); } mmap_unlock(); return ret; diff --git a/linux-user/elfload.c b/linux-user/elfload.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/elfload.c +++ b/linux-user/elfload.c @@ -XXX,XX +XXX,XX @@ static bool init_guest_commpage(void) exit(EXIT_FAILURE); } page_set_flags(TARGET_VSYSCALL_PAGE, - TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE, + TARGET_VSYSCALL_PAGE | ~TARGET_PAGE_MASK, PAGE_EXEC | PAGE_VALID); return true; } @@ -XXX,XX +XXX,XX @@ static bool init_guest_commpage(void) exit(EXIT_FAILURE); } - page_set_flags(commpage, commpage + qemu_host_page_size, + page_set_flags(commpage, commpage | ~qemu_host_page_mask, PAGE_READ | PAGE_EXEC | PAGE_VALID); return true; } @@ -XXX,XX +XXX,XX @@ static bool init_guest_commpage(void) exit(EXIT_FAILURE); } - page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE, + page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, PAGE_READ | PAGE_EXEC | PAGE_VALID); return true; } @@ -XXX,XX +XXX,XX @@ static bool init_guest_commpage(void) * and implement syscalls. Here, simply mark the page executable. * Special case the entry points during translation (see do_page_zero). */ - page_set_flags(LO_COMMPAGE, LO_COMMPAGE + TARGET_PAGE_SIZE, + page_set_flags(LO_COMMPAGE, LO_COMMPAGE | ~TARGET_PAGE_MASK, PAGE_EXEC | PAGE_VALID); return true; } @@ -XXX,XX +XXX,XX @@ static void zero_bss(abi_ulong elf_bss, abi_ulong last_bss, int prot) /* Ensure that the bss page(s) are valid */ if ((page_get_flags(last_bss-1) & prot) != prot) { - page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss, prot | PAGE_VALID); + page_set_flags(elf_bss & TARGET_PAGE_MASK, last_bss - 1, + prot | PAGE_VALID); } if (host_start < host_map_start) { diff --git a/linux-user/mmap.c b/linux-user/mmap.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/mmap.c +++ b/linux-user/mmap.c @@ -XXX,XX +XXX,XX @@ int target_mprotect(abi_ulong start, abi_ulong len, int target_prot) } } - page_set_flags(start, start + len, page_flags); + page_set_flags(start, start + len - 1, page_flags); ret = 0; error: @@ -XXX,XX +XXX,XX @@ abi_long target_mmap(abi_ulong start, abi_ulong len, int target_prot, } page_flags |= PAGE_RESET; if (passthrough_start == passthrough_end) { - page_set_flags(start, start + len, page_flags); + page_set_flags(start, start + len - 1, page_flags); } else { if (start < passthrough_start) { - page_set_flags(start, passthrough_start, page_flags); + page_set_flags(start, passthrough_start - 1, page_flags); } - page_set_flags(passthrough_start, passthrough_end, + page_set_flags(passthrough_start, passthrough_end - 1, page_flags | PAGE_PASSTHROUGH); if (passthrough_end < start + len) { - page_set_flags(passthrough_end, start + len, page_flags); + page_set_flags(passthrough_end, start + len - 1, page_flags); } } the_end: @@ -XXX,XX +XXX,XX @@ int target_munmap(abi_ulong start, abi_ulong len) } if (ret == 0) { - page_set_flags(start, start + len, 0); + page_set_flags(start, start + len - 1, 0); } mmap_unlock(); return ret; @@ -XXX,XX +XXX,XX @@ abi_long target_mremap(abi_ulong old_addr, abi_ulong old_size, } else { new_addr = h2g(host_addr); prot = page_get_flags(old_addr); - page_set_flags(old_addr, old_addr + old_size, 0); - page_set_flags(new_addr, new_addr + new_size, + page_set_flags(old_addr, old_addr + old_size - 1, 0); + page_set_flags(new_addr, new_addr + new_size - 1, prot | PAGE_VALID | PAGE_RESET); } mmap_unlock(); diff --git a/linux-user/syscall.c b/linux-user/syscall.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/syscall.c +++ b/linux-user/syscall.c @@ -XXX,XX +XXX,XX @@ static inline abi_ulong do_shmat(CPUArchState *cpu_env, } raddr=h2g((unsigned long)host_raddr); - page_set_flags(raddr, raddr + shm_info.shm_segsz, + page_set_flags(raddr, raddr + shm_info.shm_segsz - 1, PAGE_VALID | PAGE_RESET | PAGE_READ | (shmflg & SHM_RDONLY ? 0 : PAGE_WRITE)); @@ -XXX,XX +XXX,XX @@ static inline abi_long do_shmdt(abi_ulong shmaddr) for (i = 0; i < N_SHM_REGIONS; ++i) { if (shm_regions[i].in_use && shm_regions[i].start == shmaddr) { shm_regions[i].in_use = false; - page_set_flags(shmaddr, shmaddr + shm_regions[i].size, 0); + page_set_flags(shmaddr, shmaddr + shm_regions[i].size - 1, 0); break; } } -- 2.34.1 Pass the address of the last byte to be changed, rather than the first address past the last byte. This avoids overflow when the last page of the address space is involved. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/exec/cpu-all.h | 2 +- accel/tcg/user-exec.c | 11 +++++------ linux-user/mmap.c | 2 +- 3 files changed, 7 insertions(+), 8 deletions(-) diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h index XXXXXXX..XXXXXXX 100644 --- a/include/exec/cpu-all.h +++ b/include/exec/cpu-all.h @@ -XXX,XX +XXX,XX @@ int walk_memory_regions(void *, walk_memory_regions_fn); int page_get_flags(target_ulong address); void page_set_flags(target_ulong start, target_ulong last, int flags); -void page_reset_target_data(target_ulong start, target_ulong end); +void page_reset_target_data(target_ulong start, target_ulong last); int page_check_range(target_ulong start, target_ulong len, int flags); /** diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/user-exec.c +++ b/accel/tcg/user-exec.c @@ -XXX,XX +XXX,XX @@ void page_set_flags(target_ulong start, target_ulong last, int flags) } if (!flags || reset) { - page_reset_target_data(start, last + 1); + page_reset_target_data(start, last); inval_tb |= pageflags_unset(start, last); } if (flags) { @@ -XXX,XX +XXX,XX @@ typedef struct TargetPageDataNode { static IntervalTreeRoot targetdata_root; -void page_reset_target_data(target_ulong start, target_ulong end) +void page_reset_target_data(target_ulong start, target_ulong last) { IntervalTreeNode *n, *next; - target_ulong last; assert_memory_lock(); - start = start & TARGET_PAGE_MASK; - last = TARGET_PAGE_ALIGN(end) - 1; + start &= TARGET_PAGE_MASK; + last |= ~TARGET_PAGE_MASK; for (n = interval_tree_iter_first(&targetdata_root, start, last), next = n ? interval_tree_iter_next(n, start, last) : NULL; @@ -XXX,XX +XXX,XX @@ void *page_get_target_data(target_ulong address) return t->data[(page - region) >> TARGET_PAGE_BITS]; } #else -void page_reset_target_data(target_ulong start, target_ulong end) { } +void page_reset_target_data(target_ulong start, target_ulong last) { } #endif /* TARGET_PAGE_DATA_SIZE */ /* The softmmu versions of these helpers are in cputlb.c. */ diff --git a/linux-user/mmap.c b/linux-user/mmap.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/mmap.c +++ b/linux-user/mmap.c @@ -XXX,XX +XXX,XX @@ abi_long target_madvise(abi_ulong start, abi_ulong len_in, int advice) if (can_passthrough_madvise(start, end)) { ret = get_errno(madvise(g2h_untagged(start), len, advice)); if ((advice == MADV_DONTNEED) && (ret == 0)) { - page_reset_target_data(start, start + len); + page_reset_target_data(start, start + len - 1); } } } -- 2.34.1 Pass the address of the last byte to be changed, rather than the first address past the last byte. This avoids overflow when the last page of the address space is involved. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- accel/tcg/tb-maint.c | 28 ++++++++++++++++------------ 1 file changed, 16 insertions(+), 12 deletions(-) diff --git a/accel/tcg/tb-maint.c b/accel/tcg/tb-maint.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/tb-maint.c +++ b/accel/tcg/tb-maint.c @@ -XXX,XX +XXX,XX @@ static void tb_remove(TranslationBlock *tb) } /* TODO: For now, still shared with translate-all.c for system mode. */ -#define PAGE_FOR_EACH_TB(start, end, pagedesc, T, N) \ - for (T = foreach_tb_first(start, end), \ - N = foreach_tb_next(T, start, end); \ +#define PAGE_FOR_EACH_TB(start, last, pagedesc, T, N) \ + for (T = foreach_tb_first(start, last), \ + N = foreach_tb_next(T, start, last); \ T != NULL; \ - T = N, N = foreach_tb_next(N, start, end)) + T = N, N = foreach_tb_next(N, start, last)) typedef TranslationBlock *PageForEachNext; static PageForEachNext foreach_tb_first(tb_page_addr_t start, - tb_page_addr_t end) + tb_page_addr_t last) { - IntervalTreeNode *n = interval_tree_iter_first(&tb_root, start, end - 1); + IntervalTreeNode *n = interval_tree_iter_first(&tb_root, start, last); return n ? container_of(n, TranslationBlock, itree) : NULL; } static PageForEachNext foreach_tb_next(PageForEachNext tb, tb_page_addr_t start, - tb_page_addr_t end) + tb_page_addr_t last) { IntervalTreeNode *n; if (tb) { - n = interval_tree_iter_next(&tb->itree, start, end - 1); + n = interval_tree_iter_next(&tb->itree, start, last); if (n) { return container_of(n, TranslationBlock, itree); } @@ -XXX,XX +XXX,XX @@ struct page_collection { }; typedef int PageForEachNext; -#define PAGE_FOR_EACH_TB(start, end, pagedesc, tb, n) \ +#define PAGE_FOR_EACH_TB(start, last, pagedesc, tb, n) \ TB_FOR_EACH_TAGGED((pagedesc)->first_tb, tb, n, page_next) #ifdef CONFIG_DEBUG_TCG @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end) { TranslationBlock *tb; PageForEachNext n; + tb_page_addr_t last = end - 1; assert_memory_lock(); - PAGE_FOR_EACH_TB(start, end, unused, tb, n) { + PAGE_FOR_EACH_TB(start, last, unused, tb, n) { tb_phys_invalidate__locked(tb); } } @@ -XXX,XX +XXX,XX @@ bool tb_invalidate_phys_page_unwind(tb_page_addr_t addr, uintptr_t pc) bool current_tb_modified; TranslationBlock *tb; PageForEachNext n; + tb_page_addr_t last; /* * Without precise smc semantics, or when outside of a TB, @@ -XXX,XX +XXX,XX @@ bool tb_invalidate_phys_page_unwind(tb_page_addr_t addr, uintptr_t pc) assert_memory_lock(); current_tb = tcg_tb_lookup(pc); + last = addr | ~TARGET_PAGE_MASK; addr &= TARGET_PAGE_MASK; current_tb_modified = false; - PAGE_FOR_EACH_TB(addr, addr + TARGET_PAGE_SIZE, unused, tb, n) { + PAGE_FOR_EACH_TB(addr, last, unused, tb, n) { if (current_tb == tb && (tb_cflags(current_tb) & CF_COUNT_MASK) != 1) { /* @@ -XXX,XX +XXX,XX @@ tb_invalidate_phys_page_range__locked(struct page_collection *pages, bool current_tb_modified = false; TranslationBlock *current_tb = retaddr ? tcg_tb_lookup(retaddr) : NULL; #endif /* TARGET_HAS_PRECISE_SMC */ + tb_page_addr_t last G_GNUC_UNUSED = end - 1; /* * We remove all the TBs in the range [start, end[. * XXX: see if in some cases it could be faster to invalidate all the code */ - PAGE_FOR_EACH_TB(start, end, p, tb, n) { + PAGE_FOR_EACH_TB(start, last, p, tb, n) { /* NOTE: this is subtle as a TB may span two physical pages */ if (n == 0) { /* NOTE: tb_end may be after the end of the page, but -- 2.34.1 Pass the address of the last byte to be changed, rather than the first address past the last byte. This avoids overflow when the last page of the address space is involved. Fixes a bug in the loop comparision where "<= end" would lock one more page than required. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- accel/tcg/tb-maint.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/accel/tcg/tb-maint.c b/accel/tcg/tb-maint.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/tb-maint.c +++ b/accel/tcg/tb-maint.c @@ -XXX,XX +XXX,XX @@ static gint tb_page_addr_cmp(gconstpointer ap, gconstpointer bp, gpointer udata) } /* - * Lock a range of pages ([@start,@end[) as well as the pages of all + * Lock a range of pages ([@start,@last]) as well as the pages of all * intersecting TBs. * Locking order: acquire locks in ascending order of page index. */ static struct page_collection *page_collection_lock(tb_page_addr_t start, - tb_page_addr_t end) + tb_page_addr_t last) { struct page_collection *set = g_malloc(sizeof(*set)); tb_page_addr_t index; PageDesc *pd; start >>= TARGET_PAGE_BITS; - end >>= TARGET_PAGE_BITS; - g_assert(start <= end); + last >>= TARGET_PAGE_BITS; + g_assert(start <= last); set->tree = q_tree_new_full(tb_page_addr_cmp, NULL, NULL, page_entry_destroy); @@ -XXX,XX +XXX,XX @@ static struct page_collection *page_collection_lock(tb_page_addr_t start, retry: q_tree_foreach(set->tree, page_entry_lock, NULL); - for (index = start; index <= end; index++) { + for (index = start; index <= last; index++) { TranslationBlock *tb; PageForEachNext n; @@ -XXX,XX +XXX,XX @@ tb_invalidate_phys_page_range__locked(struct page_collection *pages, void tb_invalidate_phys_page(tb_page_addr_t addr) { struct page_collection *pages; - tb_page_addr_t start, end; + tb_page_addr_t start, last; PageDesc *p; p = page_find(addr >> TARGET_PAGE_BITS); @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_page(tb_page_addr_t addr) } start = addr & TARGET_PAGE_MASK; - end = start + TARGET_PAGE_SIZE; - pages = page_collection_lock(start, end); - tb_invalidate_phys_page_range__locked(pages, p, start, end, 0); + last = addr | ~TARGET_PAGE_MASK; + pages = page_collection_lock(start, last); + tb_invalidate_phys_page_range__locked(pages, p, start, last + 1, 0); page_collection_unlock(pages); } @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end) struct page_collection *pages; tb_page_addr_t next; - pages = page_collection_lock(start, end); + pages = page_collection_lock(start, end - 1); for (next = (start & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE; start < end; start = next, next += TARGET_PAGE_SIZE) { @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_range_fast(ram_addr_t ram_addr, { struct page_collection *pages; - pages = page_collection_lock(ram_addr, ram_addr + size); + pages = page_collection_lock(ram_addr, ram_addr + size - 1); tb_invalidate_phys_page_fast__locked(pages, ram_addr, size, retaddr); page_collection_unlock(pages); } -- 2.34.1 Pass the address of the last byte to be changed, rather than the first address past the last byte. This avoids overflow when the last page of the address space is involved. Properly truncate tb_last to the end of the page; the comment about tb_end being past the end of the page being ok is not correct, considering overflow. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- accel/tcg/tb-maint.c | 26 ++++++++++++-------------- 1 file changed, 12 insertions(+), 14 deletions(-) diff --git a/accel/tcg/tb-maint.c b/accel/tcg/tb-maint.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/tb-maint.c +++ b/accel/tcg/tb-maint.c @@ -XXX,XX +XXX,XX @@ bool tb_invalidate_phys_page_unwind(tb_page_addr_t addr, uintptr_t pc) static void tb_invalidate_phys_page_range__locked(struct page_collection *pages, PageDesc *p, tb_page_addr_t start, - tb_page_addr_t end, + tb_page_addr_t last, uintptr_t retaddr) { TranslationBlock *tb; - tb_page_addr_t tb_start, tb_end; PageForEachNext n; #ifdef TARGET_HAS_PRECISE_SMC bool current_tb_modified = false; TranslationBlock *current_tb = retaddr ? tcg_tb_lookup(retaddr) : NULL; #endif /* TARGET_HAS_PRECISE_SMC */ - tb_page_addr_t last G_GNUC_UNUSED = end - 1; /* - * We remove all the TBs in the range [start, end[. + * We remove all the TBs in the range [start, last]. * XXX: see if in some cases it could be faster to invalidate all the code */ PAGE_FOR_EACH_TB(start, last, p, tb, n) { + tb_page_addr_t tb_start, tb_last; + /* NOTE: this is subtle as a TB may span two physical pages */ + tb_start = tb_page_addr0(tb); + tb_last = tb_start + tb->size - 1; if (n == 0) { - /* NOTE: tb_end may be after the end of the page, but - it is not a problem */ - tb_start = tb_page_addr0(tb); - tb_end = tb_start + tb->size; + tb_last = MIN(tb_last, tb_start | ~TARGET_PAGE_MASK); } else { tb_start = tb_page_addr1(tb); - tb_end = tb_start + ((tb_page_addr0(tb) + tb->size) - & ~TARGET_PAGE_MASK); + tb_last = tb_start + (tb_last & ~TARGET_PAGE_MASK); } - if (!(tb_end <= start || tb_start >= end)) { + if (!(tb_last < start || tb_start > last)) { #ifdef TARGET_HAS_PRECISE_SMC if (current_tb == tb && (tb_cflags(current_tb) & CF_COUNT_MASK) != 1) { @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_page(tb_page_addr_t addr) start = addr & TARGET_PAGE_MASK; last = addr | ~TARGET_PAGE_MASK; pages = page_collection_lock(start, last); - tb_invalidate_phys_page_range__locked(pages, p, start, last + 1, 0); + tb_invalidate_phys_page_range__locked(pages, p, start, last, 0); page_collection_unlock(pages); } @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end) continue; } assert_page_locked(pd); - tb_invalidate_phys_page_range__locked(pages, pd, start, bound, 0); + tb_invalidate_phys_page_range__locked(pages, pd, start, bound - 1, 0); } page_collection_unlock(pages); } @@ -XXX,XX +XXX,XX @@ static void tb_invalidate_phys_page_fast__locked(struct page_collection *pages, } assert_page_locked(p); - tb_invalidate_phys_page_range__locked(pages, p, start, start + len, ra); + tb_invalidate_phys_page_range__locked(pages, p, start, start + len - 1, ra); } /* -- 2.34.1 Pass the address of the last byte to be changed, rather than the first address past the last byte. This avoids overflow when the last page of the address space is involved. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/exec/exec-all.h | 2 +- accel/tcg/tb-maint.c | 31 ++++++++++++++++--------------- accel/tcg/translate-all.c | 2 +- accel/tcg/user-exec.c | 2 +- softmmu/physmem.c | 2 +- 5 files changed, 20 insertions(+), 19 deletions(-) diff --git a/include/exec/exec-all.h b/include/exec/exec-all.h index XXXXXXX..XXXXXXX 100644 --- a/include/exec/exec-all.h +++ b/include/exec/exec-all.h @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_addr(target_ulong addr); void tb_invalidate_phys_addr(AddressSpace *as, hwaddr addr, MemTxAttrs attrs); #endif void tb_phys_invalidate(TranslationBlock *tb, tb_page_addr_t page_addr); -void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end); +void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t last); void tb_set_jmp_target(TranslationBlock *tb, int n, uintptr_t addr); /* GETPC is the true target of the return instruction that we'll execute. */ diff --git a/accel/tcg/tb-maint.c b/accel/tcg/tb-maint.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/tb-maint.c +++ b/accel/tcg/tb-maint.c @@ -XXX,XX +XXX,XX @@ TranslationBlock *tb_link_page(TranslationBlock *tb, tb_page_addr_t phys_pc, * Called with mmap_lock held for user-mode emulation. * NOTE: this function must not be called while a TB is running. */ -void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end) +void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t last) { TranslationBlock *tb; PageForEachNext n; - tb_page_addr_t last = end - 1; assert_memory_lock(); @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end) */ void tb_invalidate_phys_page(tb_page_addr_t addr) { - tb_page_addr_t start, end; + tb_page_addr_t start, last; start = addr & TARGET_PAGE_MASK; - end = start + TARGET_PAGE_SIZE; - tb_invalidate_phys_range(start, end); + last = addr | ~TARGET_PAGE_MASK; + tb_invalidate_phys_range(start, last); } /* @@ -XXX,XX +XXX,XX @@ void tb_invalidate_phys_page(tb_page_addr_t addr) /* * Invalidate all TBs which intersect with the target physical address range - * [start;end[. NOTE: start and end may refer to *different* physical pages. + * [start;last]. NOTE: start and end may refer to *different* physical pages. * 'is_cpu_write_access' should be true if called from a real cpu write * access: the virtual CPU will exit the current TB if code is modified inside * this TB. */ -void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t end) +void tb_invalidate_phys_range(tb_page_addr_t start, tb_page_addr_t last) { struct page_collection *pages; - tb_page_addr_t next; + tb_page_addr_t index, index_last; - pages = page_collection_lock(start, end - 1); - for (next = (start & TARGET_PAGE_MASK) + TARGET_PAGE_SIZE; - start < end; - start = next, next += TARGET_PAGE_SIZE) { - PageDesc *pd = page_find(start >> TARGET_PAGE_BITS); - tb_page_addr_t bound = MIN(next, end); + pages = page_collection_lock(start, last); + + index_last = last >> TARGET_PAGE_BITS; + for (index = start >> TARGET_PAGE_BITS; index <= index_last; index++) { + PageDesc *pd = page_find(index); + tb_page_addr_t bound; if (pd == NULL) { continue; } assert_page_locked(pd); - tb_invalidate_phys_page_range__locked(pages, pd, start, bound - 1, 0); + bound = (index << TARGET_PAGE_BITS) | ~TARGET_PAGE_MASK; + bound = MIN(bound, last); + tb_invalidate_phys_page_range__locked(pages, pd, start, bound, 0); } page_collection_unlock(pages); } diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/translate-all.c +++ b/accel/tcg/translate-all.c @@ -XXX,XX +XXX,XX @@ void tb_check_watchpoint(CPUState *cpu, uintptr_t retaddr) cpu_get_tb_cpu_state(env, &pc, &cs_base, &flags); addr = get_page_addr_code(env, pc); if (addr != -1) { - tb_invalidate_phys_range(addr, addr + 1); + tb_invalidate_phys_range(addr, addr); } } } diff --git a/accel/tcg/user-exec.c b/accel/tcg/user-exec.c index XXXXXXX..XXXXXXX 100644 --- a/accel/tcg/user-exec.c +++ b/accel/tcg/user-exec.c @@ -XXX,XX +XXX,XX @@ void page_set_flags(target_ulong start, target_ulong last, int flags) ~(reset ? 0 : PAGE_STICKY)); } if (inval_tb) { - tb_invalidate_phys_range(start, last + 1); + tb_invalidate_phys_range(start, last); } } diff --git a/softmmu/physmem.c b/softmmu/physmem.c index XXXXXXX..XXXXXXX 100644 --- a/softmmu/physmem.c +++ b/softmmu/physmem.c @@ -XXX,XX +XXX,XX @@ static void invalidate_and_set_dirty(MemoryRegion *mr, hwaddr addr, } if (dirty_log_mask & (1 << DIRTY_MEMORY_CODE)) { assert(tcg_enabled()); - tb_invalidate_phys_range(addr, addr + length); + tb_invalidate_phys_range(addr, addr + length - 1); dirty_log_mask &= ~(1 << DIRTY_MEMORY_CODE); } cpu_physical_memory_set_dirty_range(addr, length, dirty_log_mask); -- 2.34.1 Pass the address of the last byte of the image, rather than the first address past the last byte. This avoids overflow when the last page of the address space is involved. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- linux-user/elfload.c | 24 ++++++++++++------------ linux-user/flatload.c | 2 +- 2 files changed, 13 insertions(+), 13 deletions(-) diff --git a/linux-user/elfload.c b/linux-user/elfload.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/elfload.c +++ b/linux-user/elfload.c @@ -XXX,XX +XXX,XX @@ static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, if (guest_hiaddr > reserved_va) { error_report("%s: requires more than reserved virtual " "address space (0x%" PRIx64 " > 0x%lx)", - image_name, (uint64_t)guest_hiaddr, reserved_va); + image_name, (uint64_t)guest_hiaddr + 1, reserved_va); exit(EXIT_FAILURE); } } else { @@ -XXX,XX +XXX,XX @@ static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, if ((guest_hiaddr - guest_base) > ~(uintptr_t)0) { error_report("%s: requires more virtual address space " "than the host can provide (0x%" PRIx64 ")", - image_name, (uint64_t)guest_hiaddr - guest_base); + image_name, (uint64_t)guest_hiaddr + 1 - guest_base); exit(EXIT_FAILURE); } #endif @@ -XXX,XX +XXX,XX @@ static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, if (reserved_va) { guest_loaddr = (guest_base >= mmap_min_addr ? 0 : mmap_min_addr - guest_base); - guest_hiaddr = reserved_va; + guest_hiaddr = reserved_va - 1; } /* Reserve the address space for the binary, or reserved_va. */ test = g2h_untagged(guest_loaddr); - addr = mmap(test, guest_hiaddr - guest_loaddr, PROT_NONE, flags, -1, 0); + addr = mmap(test, guest_hiaddr - guest_loaddr + 1, PROT_NONE, flags, -1, 0); if (test != addr) { pgb_fail_in_use(image_name); } qemu_log_mask(CPU_LOG_PAGE, - "%s: base @ %p for " TARGET_ABI_FMT_ld " bytes\n", - __func__, addr, guest_hiaddr - guest_loaddr); + "%s: base @ %p for %" PRIu64 " bytes\n", + __func__, addr, (uint64_t)guest_hiaddr - guest_loaddr + 1); } /** @@ -XXX,XX +XXX,XX @@ static void pgb_static(const char *image_name, abi_ulong orig_loaddr, if (hiaddr != orig_hiaddr) { error_report("%s: requires virtual address space that the " "host cannot provide (0x%" PRIx64 ")", - image_name, (uint64_t)orig_hiaddr); + image_name, (uint64_t)orig_hiaddr + 1); exit(EXIT_FAILURE); } @@ -XXX,XX +XXX,XX @@ static void pgb_static(const char *image_name, abi_ulong orig_loaddr, * arithmetic wraps around. */ if (sizeof(uintptr_t) == 8 || loaddr >= 0x80000000u) { - hiaddr = (uintptr_t) 4 << 30; + hiaddr = UINT32_MAX; } else { offset = -(HI_COMMPAGE & -align); } @@ -XXX,XX +XXX,XX @@ static void pgb_static(const char *image_name, abi_ulong orig_loaddr, loaddr = MIN(loaddr, LO_COMMPAGE & -align); } - addr = pgb_find_hole(loaddr, hiaddr - loaddr, align, offset); + addr = pgb_find_hole(loaddr, hiaddr - loaddr + 1, align, offset); if (addr == -1) { /* * If HI_COMMPAGE, there *might* be a non-consecutive allocation @@ -XXX,XX +XXX,XX @@ static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, if (guest_hiaddr > reserved_va) { error_report("%s: requires more than reserved virtual " "address space (0x%" PRIx64 " > 0x%lx)", - image_name, (uint64_t)guest_hiaddr, reserved_va); + image_name, (uint64_t)guest_hiaddr + 1, reserved_va); exit(EXIT_FAILURE); } @@ -XXX,XX +XXX,XX @@ static void load_elf_image(const char *image_name, int image_fd, if (a < loaddr) { loaddr = a; } - a = eppnt->p_vaddr + eppnt->p_memsz; + a = eppnt->p_vaddr + eppnt->p_memsz - 1; if (a > hiaddr) { hiaddr = a; } @@ -XXX,XX +XXX,XX @@ static void load_elf_image(const char *image_name, int image_fd, * In both cases, we will overwrite pages in this range with mappings * from the executable. */ - load_addr = target_mmap(loaddr, hiaddr - loaddr, PROT_NONE, + load_addr = target_mmap(loaddr, (size_t)hiaddr - loaddr + 1, PROT_NONE, MAP_PRIVATE | MAP_ANON | MAP_NORESERVE | (ehdr->e_type == ET_EXEC ? MAP_FIXED : 0), -1, 0); diff --git a/linux-user/flatload.c b/linux-user/flatload.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/flatload.c +++ b/linux-user/flatload.c @@ -XXX,XX +XXX,XX @@ static int load_flat_file(struct linux_binprm * bprm, * Allocate the address space. */ probe_guest_base(bprm->filename, 0, - text_len + data_len + extra + indx_len); + text_len + data_len + extra + indx_len - 1); /* * there are a couple of cases here, the separate code/data -- 2.34.1 Change the semantics to be the last byte of the guest va, rather than the following byte. This avoids some overflow conditions. Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org> Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- include/exec/cpu-all.h | 11 ++++++++++- linux-user/arm/target_cpu.h | 2 +- bsd-user/main.c | 10 +++------- bsd-user/mmap.c | 4 ++-- linux-user/elfload.c | 14 +++++++------- linux-user/main.c | 27 +++++++++++++-------------- linux-user/mmap.c | 4 ++-- 7 files changed, 38 insertions(+), 34 deletions(-) diff --git a/include/exec/cpu-all.h b/include/exec/cpu-all.h index XXXXXXX..XXXXXXX 100644 --- a/include/exec/cpu-all.h +++ b/include/exec/cpu-all.h @@ -XXX,XX +XXX,XX @@ static inline void tswap64s(uint64_t *s) */ extern uintptr_t guest_base; extern bool have_guest_base; + +/* + * If non-zero, the guest virtual address space is a contiguous subset + * of the host virtual address space, i.e. '-R reserved_va' is in effect + * either from the command-line or by default. The value is the last + * byte of the guest address space e.g. UINT32_MAX. + * + * If zero, the host and guest virtual address spaces are intermingled. + */ extern unsigned long reserved_va; /* @@ -XXX,XX +XXX,XX @@ extern unsigned long reserved_va; #define GUEST_ADDR_MAX_ \ ((MIN_CONST(TARGET_VIRT_ADDR_SPACE_BITS, TARGET_ABI_BITS) <= 32) ? \ UINT32_MAX : ~0ul) -#define GUEST_ADDR_MAX (reserved_va ? reserved_va - 1 : GUEST_ADDR_MAX_) +#define GUEST_ADDR_MAX (reserved_va ? : GUEST_ADDR_MAX_) #else diff --git a/linux-user/arm/target_cpu.h b/linux-user/arm/target_cpu.h index XXXXXXX..XXXXXXX 100644 --- a/linux-user/arm/target_cpu.h +++ b/linux-user/arm/target_cpu.h @@ -XXX,XX +XXX,XX @@ static inline unsigned long arm_max_reserved_va(CPUState *cs) * the high addresses. Restrict linux-user to the * cached write-back RAM in the system map. */ - return 0x80000000ul; + return 0x7ffffffful; } else { /* * We need to be able to map the commpage. diff --git a/bsd-user/main.c b/bsd-user/main.c index XXXXXXX..XXXXXXX 100644 --- a/bsd-user/main.c +++ b/bsd-user/main.c @@ -XXX,XX +XXX,XX @@ bool have_guest_base; # if HOST_LONG_BITS > TARGET_VIRT_ADDR_SPACE_BITS # if TARGET_VIRT_ADDR_SPACE_BITS == 32 && \ (TARGET_LONG_BITS == 32 || defined(TARGET_ABI32)) -/* - * There are a number of places where we assign reserved_va to a variable - * of type abi_ulong and expect it to fit. Avoid the last page. - */ -# define MAX_RESERVED_VA (0xfffffffful & TARGET_PAGE_MASK) +# define MAX_RESERVED_VA 0xfffffffful # else -# define MAX_RESERVED_VA (1ul << TARGET_VIRT_ADDR_SPACE_BITS) +# define MAX_RESERVED_VA ((1ul << TARGET_VIRT_ADDR_SPACE_BITS) - 1) # endif # else # define MAX_RESERVED_VA 0 @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv) envlist_free(envlist); if (reserved_va) { - mmap_next_start = reserved_va; + mmap_next_start = reserved_va + 1; } { diff --git a/bsd-user/mmap.c b/bsd-user/mmap.c index XXXXXXX..XXXXXXX 100644 --- a/bsd-user/mmap.c +++ b/bsd-user/mmap.c @@ -XXX,XX +XXX,XX @@ static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size, size = HOST_PAGE_ALIGN(size) + alignment; end_addr = start + size; if (end_addr > reserved_va) { - end_addr = reserved_va; + end_addr = reserved_va + 1; } addr = end_addr - qemu_host_page_size; @@ -XXX,XX +XXX,XX @@ static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size, if (looped) { return (abi_ulong)-1; } - end_addr = reserved_va; + end_addr = reserved_va + 1; addr = end_addr - qemu_host_page_size; looped = 1; continue; diff --git a/linux-user/elfload.c b/linux-user/elfload.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/elfload.c +++ b/linux-user/elfload.c @@ -XXX,XX +XXX,XX @@ static bool init_guest_commpage(void) * has specified -R reserved_va, which would trigger an assert(). */ if (reserved_va != 0 && - TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE >= reserved_va) { + TARGET_VSYSCALL_PAGE + TARGET_PAGE_SIZE - 1 > reserved_va) { error_report("Cannot allocate vsyscall page"); exit(EXIT_FAILURE); } @@ -XXX,XX +XXX,XX @@ static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, if (guest_hiaddr > reserved_va) { error_report("%s: requires more than reserved virtual " "address space (0x%" PRIx64 " > 0x%lx)", - image_name, (uint64_t)guest_hiaddr + 1, reserved_va); + image_name, (uint64_t)guest_hiaddr, reserved_va); exit(EXIT_FAILURE); } } else { @@ -XXX,XX +XXX,XX @@ static void pgb_have_guest_base(const char *image_name, abi_ulong guest_loaddr, if (reserved_va) { guest_loaddr = (guest_base >= mmap_min_addr ? 0 : mmap_min_addr - guest_base); - guest_hiaddr = reserved_va - 1; + guest_hiaddr = reserved_va; } /* Reserve the address space for the binary, or reserved_va. */ @@ -XXX,XX +XXX,XX @@ static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, if (guest_hiaddr > reserved_va) { error_report("%s: requires more than reserved virtual " "address space (0x%" PRIx64 " > 0x%lx)", - image_name, (uint64_t)guest_hiaddr + 1, reserved_va); + image_name, (uint64_t)guest_hiaddr, reserved_va); exit(EXIT_FAILURE); } @@ -XXX,XX +XXX,XX @@ static void pgb_reserved_va(const char *image_name, abi_ulong guest_loaddr, /* Reserve the memory on the host. */ assert(guest_base != 0); test = g2h_untagged(0); - addr = mmap(test, reserved_va, PROT_NONE, flags, -1, 0); + addr = mmap(test, reserved_va + 1, PROT_NONE, flags, -1, 0); if (addr == MAP_FAILED || addr != test) { error_report("Unable to reserve 0x%lx bytes of virtual address " "space at %p (%s) for use as guest address space (check your " "virtual memory ulimit setting, min_mmap_addr or reserve less " - "using -R option)", reserved_va, test, strerror(errno)); + "using -R option)", reserved_va + 1, test, strerror(errno)); exit(EXIT_FAILURE); } qemu_log_mask(CPU_LOG_PAGE, "%s: base @ %p for %lu bytes\n", - __func__, addr, reserved_va); + __func__, addr, reserved_va + 1); } void probe_guest_base(const char *image_name, abi_ulong guest_loaddr, diff --git a/linux-user/main.c b/linux-user/main.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/main.c +++ b/linux-user/main.c @@ -XXX,XX +XXX,XX @@ static const char *last_log_filename; # if HOST_LONG_BITS > TARGET_VIRT_ADDR_SPACE_BITS # if TARGET_VIRT_ADDR_SPACE_BITS == 32 && \ (TARGET_LONG_BITS == 32 || defined(TARGET_ABI32)) -/* There are a number of places where we assign reserved_va to a variable - of type abi_ulong and expect it to fit. Avoid the last page. */ -# define MAX_RESERVED_VA(CPU) (0xfffffffful & TARGET_PAGE_MASK) +# define MAX_RESERVED_VA(CPU) 0xfffffffful # else -# define MAX_RESERVED_VA(CPU) (1ul << TARGET_VIRT_ADDR_SPACE_BITS) +# define MAX_RESERVED_VA(CPU) ((1ul << TARGET_VIRT_ADDR_SPACE_BITS) - 1) # endif # else # define MAX_RESERVED_VA(CPU) 0 @@ -XXX,XX +XXX,XX @@ static void handle_arg_reserved_va(const char *arg) { char *p; int shift = 0; - reserved_va = strtoul(arg, &p, 0); + unsigned long val; + + val = strtoul(arg, &p, 0); switch (*p) { case 'k': case 'K': @@ -XXX,XX +XXX,XX @@ static void handle_arg_reserved_va(const char *arg) break; } if (shift) { - unsigned long unshifted = reserved_va; + unsigned long unshifted = val; p++; - reserved_va <<= shift; - if (reserved_va >> shift != unshifted) { + val <<= shift; + if (val >> shift != unshifted) { fprintf(stderr, "Reserved virtual address too big\n"); exit(EXIT_FAILURE); } @@ -XXX,XX +XXX,XX @@ static void handle_arg_reserved_va(const char *arg) fprintf(stderr, "Unrecognised -R size suffix '%s'\n", p); exit(EXIT_FAILURE); } + /* The representation is size - 1, with 0 remaining "default". */ + reserved_va = val ? val - 1 : 0; } static void handle_arg_singlestep(const char *arg) @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp) */ max_reserved_va = MAX_RESERVED_VA(cpu); if (reserved_va != 0) { - if (reserved_va % qemu_host_page_size) { + if ((reserved_va + 1) % qemu_host_page_size) { char *s = size_to_str(qemu_host_page_size); fprintf(stderr, "Reserved virtual address not aligned mod %s\n", s); g_free(s); @@ -XXX,XX +XXX,XX @@ int main(int argc, char **argv, char **envp) exit(EXIT_FAILURE); } } else if (HOST_LONG_BITS == 64 && TARGET_VIRT_ADDR_SPACE_BITS <= 32) { - /* - * reserved_va must be aligned with the host page size - * as it is used with mmap() - */ - reserved_va = max_reserved_va & qemu_host_page_mask; + /* MAX_RESERVED_VA + 1 is a large power of 2, so is aligned. */ + reserved_va = max_reserved_va; } { diff --git a/linux-user/mmap.c b/linux-user/mmap.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/mmap.c +++ b/linux-user/mmap.c @@ -XXX,XX +XXX,XX @@ static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size, end_addr = start + size; if (start > reserved_va - size) { /* Start at the top of the address space. */ - end_addr = ((reserved_va - size) & -align) + size; + end_addr = ((reserved_va + 1 - size) & -align) + size; looped = true; } @@ -XXX,XX +XXX,XX @@ static abi_ulong mmap_find_vma_reserved(abi_ulong start, abi_ulong size, return (abi_ulong)-1; } /* Re-start at the top of the address space. */ - addr = end_addr = ((reserved_va - size) & -align) + size; + addr = end_addr = ((reserved_va + 1 - size) & -align) + size; looped = true; } else { prot = page_get_flags(addr); -- 2.34.1 User setting of -R reserved_va can lead to an assertion failure in page_set_flags. Sanity check the value of reserved_va and print an error message instead. Do not allocate a commpage at all for m-profile cpus. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- linux-user/elfload.c | 37 +++++++++++++++++++++++++++---------- 1 file changed, 27 insertions(+), 10 deletions(-) diff --git a/linux-user/elfload.c b/linux-user/elfload.c index XXXXXXX..XXXXXXX 100644 --- a/linux-user/elfload.c +++ b/linux-user/elfload.c @@ -XXX,XX +XXX,XX @@ enum { static bool init_guest_commpage(void) { - abi_ptr commpage = HI_COMMPAGE & -qemu_host_page_size; - void *want = g2h_untagged(commpage); - void *addr = mmap(want, qemu_host_page_size, PROT_READ | PROT_WRITE, - MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); + ARMCPU *cpu = ARM_CPU(thread_cpu); + abi_ptr want = HI_COMMPAGE & TARGET_PAGE_MASK; + abi_ptr addr; - if (addr == MAP_FAILED) { + /* + * M-profile allocates maximum of 2GB address space, so can never + * allocate the commpage. Skip it. + */ + if (arm_feature(&cpu->env, ARM_FEATURE_M)) { + return true; + } + + /* + * If reserved_va does not cover the commpage, we get an assert + * in page_set_flags. Produce an intelligent error instead. + */ + if (reserved_va != 0 && want + TARGET_PAGE_SIZE - 1 > reserved_va) { + error_report("Allocating guest commpage: -R 0x%" PRIx64 " too small", + (uint64_t)reserved_va + 1); + exit(EXIT_FAILURE); + } + + addr = target_mmap(want, TARGET_PAGE_SIZE, PROT_READ | PROT_WRITE, + MAP_ANONYMOUS | MAP_PRIVATE | MAP_FIXED, -1, 0); + + if (addr == -1) { perror("Allocating guest commpage"); exit(EXIT_FAILURE); } @@ -XXX,XX +XXX,XX @@ static bool init_guest_commpage(void) } /* Set kernel helper versions; rest of page is 0. */ - __put_user(5, (uint32_t *)g2h_untagged(0xffff0ffcu)); + put_user_u32(5, 0xffff0ffcu); - if (mprotect(addr, qemu_host_page_size, PROT_READ)) { + if (target_mprotect(addr, qemu_host_page_size, PROT_READ | PROT_EXEC)) { perror("Protecting guest commpage"); exit(EXIT_FAILURE); } - - page_set_flags(commpage, commpage | ~qemu_host_page_mask, - PAGE_READ | PAGE_EXEC | PAGE_VALID); return true; } -- 2.34.1