[PATCH v3 23/24] tcg/tci: Split out constraint sets to tcg-target-con-set.h

Richard Henderson posted 24 patches 3 years, 9 months ago
Maintainers: Huacai Chen <chenhuacai@kernel.org>, Stefan Weil <sw@weilnetz.de>, Palmer Dabbelt <palmer@dabbelt.com>, "Philippe Mathieu-Daudé" <f4bug@amsat.org>, Thomas Huth <thuth@redhat.com>, Aleksandar Rikalo <aleksandar.rikalo@syrmia.com>, Cornelia Huck <cohuck@redhat.com>, Richard Henderson <richard.henderson@linaro.org>, Andrzej Zaborowski <balrogg@gmail.com>, Alistair Francis <Alistair.Francis@wdc.com>, Aurelien Jarno <aurelien@aurel32.net>, Jiaxun Yang <jiaxun.yang@flygoat.com>
[PATCH v3 23/24] tcg/tci: Split out constraint sets to tcg-target-con-set.h
Posted by Richard Henderson 3 years, 9 months ago
This requires finishing the conversion to tcg_target_op_def.

Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 tcg/tci/tcg-target-con-set.h |  25 ++++
 tcg/tci/tcg-target.h         |   2 +
 tcg/tci/tcg-target.c.inc     | 279 +++++++++++++++++------------------
 3 files changed, 161 insertions(+), 145 deletions(-)
 create mode 100644 tcg/tci/tcg-target-con-set.h

diff --git a/tcg/tci/tcg-target-con-set.h b/tcg/tci/tcg-target-con-set.h
new file mode 100644
index 0000000000..38e82f7535
--- /dev/null
+++ b/tcg/tci/tcg-target-con-set.h
@@ -0,0 +1,25 @@
+/* SPDX-License-Identifier: MIT */
+/*
+ * TCI target-specific constraint sets.
+ * Copyright (c) 2021 Linaro
+ */
+
+/*
+ * C_On_Im(...) defines a constraint set with <n> outputs and <m> inputs.
+ * Each operand should be a sequence of constraint letters as defined by
+ * tcg-target-con-str.h; the constraint combination is inclusive or.
+ */
+C_O0_I2(r, r)
+C_O0_I2(r, ri)
+C_O0_I3(r, r, r)
+C_O0_I4(r, r, ri, ri)
+C_O0_I4(r, r, r, r)
+C_O1_I1(r, r)
+C_O1_I2(r, 0, r)
+C_O1_I2(r, ri, ri)
+C_O1_I2(r, r, r)
+C_O1_I2(r, r, ri)
+C_O1_I4(r, r, r, ri, ri)
+C_O2_I1(r, r, r)
+C_O2_I2(r, r, r, r)
+C_O2_I4(r, r, r, r, r, r)
diff --git a/tcg/tci/tcg-target.h b/tcg/tci/tcg-target.h
index bb784e018e..1efd8c4fb0 100644
--- a/tcg/tci/tcg-target.h
+++ b/tcg/tci/tcg-target.h
@@ -207,4 +207,6 @@ static inline void tb_target_set_jmp_target(uintptr_t tc_ptr, uintptr_t jmp_rx,
     /* no need to flush icache explicitly */
 }
 
+#define TCG_TARGET_CON_SET_H
+
 #endif /* TCG_TARGET_H */
diff --git a/tcg/tci/tcg-target.c.inc b/tcg/tci/tcg-target.c.inc
index 493bbf1e39..62bedaca28 100644
--- a/tcg/tci/tcg-target.c.inc
+++ b/tcg/tci/tcg-target.c.inc
@@ -37,154 +37,143 @@
 /* Bitfield n...m (in 32 bit value). */
 #define BITS(n, m) (((0xffffffffU << (31 - n)) >> (31 - n + m)) << m)
 
-/* Macros used in tcg_target_op_defs. */
-#define R       "r"
-#define RI      "ri"
-#if TCG_TARGET_REG_BITS == 32
-# define R64    "r", "r"
-#else
-# define R64    "r"
-#endif
-#if TARGET_LONG_BITS > TCG_TARGET_REG_BITS
-# define L      "r", "r"
-# define S      "r", "r"
-#else
-# define L      "r"
-# define S      "r"
-#endif
-
-/* TODO: documentation. */
-static const TCGTargetOpDef tcg_target_op_defs[] = {
-    { INDEX_op_exit_tb, { NULL } },
-    { INDEX_op_goto_tb, { NULL } },
-    { INDEX_op_br, { NULL } },
-
-    { INDEX_op_ld8u_i32, { R, R } },
-    { INDEX_op_ld8s_i32, { R, R } },
-    { INDEX_op_ld16u_i32, { R, R } },
-    { INDEX_op_ld16s_i32, { R, R } },
-    { INDEX_op_ld_i32, { R, R } },
-    { INDEX_op_st8_i32, { R, R } },
-    { INDEX_op_st16_i32, { R, R } },
-    { INDEX_op_st_i32, { R, R } },
-
-    { INDEX_op_add_i32, { R, RI, RI } },
-    { INDEX_op_sub_i32, { R, RI, RI } },
-    { INDEX_op_mul_i32, { R, RI, RI } },
-    { INDEX_op_div_i32, { R, R, R } },
-    { INDEX_op_divu_i32, { R, R, R } },
-    { INDEX_op_rem_i32, { R, R, R } },
-    { INDEX_op_remu_i32, { R, R, R } },
-    /* TODO: Does R, RI, RI result in faster code than R, R, RI?
-       If both operands are constants, we can optimize. */
-    { INDEX_op_and_i32, { R, RI, RI } },
-    { INDEX_op_andc_i32, { R, RI, RI } },
-    { INDEX_op_eqv_i32, { R, RI, RI } },
-    { INDEX_op_nand_i32, { R, RI, RI } },
-    { INDEX_op_nor_i32, { R, RI, RI } },
-    { INDEX_op_or_i32, { R, RI, RI } },
-    { INDEX_op_orc_i32, { R, RI, RI } },
-    { INDEX_op_xor_i32, { R, RI, RI } },
-    { INDEX_op_shl_i32, { R, RI, RI } },
-    { INDEX_op_shr_i32, { R, RI, RI } },
-    { INDEX_op_sar_i32, { R, RI, RI } },
-    { INDEX_op_rotl_i32, { R, RI, RI } },
-    { INDEX_op_rotr_i32, { R, RI, RI } },
-    { INDEX_op_deposit_i32, { R, "0", R } },
-
-    { INDEX_op_brcond_i32, { R, RI } },
-
-    { INDEX_op_setcond_i32, { R, R, RI } },
-    { INDEX_op_setcond_i64, { R, R, RI } },
-
-    /* TODO: Support R, R, R, R, RI, RI? Will it be faster? */
-    { INDEX_op_add2_i32, { R, R, R, R, R, R } },
-    { INDEX_op_sub2_i32, { R, R, R, R, R, R } },
-    { INDEX_op_brcond2_i32, { R, R, RI, RI } },
-    { INDEX_op_mulu2_i32, { R, R, R, R } },
-    { INDEX_op_setcond2_i32, { R, R, R, RI, RI } },
-
-    { INDEX_op_not_i32, { R, R } },
-    { INDEX_op_neg_i32, { R, R } },
-
-    { INDEX_op_ld8u_i64, { R, R } },
-    { INDEX_op_ld8s_i64, { R, R } },
-    { INDEX_op_ld16u_i64, { R, R } },
-    { INDEX_op_ld16s_i64, { R, R } },
-    { INDEX_op_ld32u_i64, { R, R } },
-    { INDEX_op_ld32s_i64, { R, R } },
-    { INDEX_op_ld_i64, { R, R } },
-
-    { INDEX_op_st8_i64, { R, R } },
-    { INDEX_op_st16_i64, { R, R } },
-    { INDEX_op_st32_i64, { R, R } },
-    { INDEX_op_st_i64, { R, R } },
-
-    { INDEX_op_add_i64, { R, RI, RI } },
-    { INDEX_op_sub_i64, { R, RI, RI } },
-    { INDEX_op_mul_i64, { R, RI, RI } },
-    { INDEX_op_div_i64, { R, R, R } },
-    { INDEX_op_divu_i64, { R, R, R } },
-    { INDEX_op_rem_i64, { R, R, R } },
-    { INDEX_op_remu_i64, { R, R, R } },
-    { INDEX_op_and_i64, { R, RI, RI } },
-    { INDEX_op_andc_i64, { R, RI, RI } },
-    { INDEX_op_eqv_i64, { R, RI, RI } },
-    { INDEX_op_nand_i64, { R, RI, RI } },
-    { INDEX_op_nor_i64, { R, RI, RI } },
-    { INDEX_op_or_i64, { R, RI, RI } },
-    { INDEX_op_orc_i64, { R, RI, RI } },
-    { INDEX_op_xor_i64, { R, RI, RI } },
-    { INDEX_op_shl_i64, { R, RI, RI } },
-    { INDEX_op_shr_i64, { R, RI, RI } },
-    { INDEX_op_sar_i64, { R, RI, RI } },
-    { INDEX_op_rotl_i64, { R, RI, RI } },
-    { INDEX_op_rotr_i64, { R, RI, RI } },
-    { INDEX_op_deposit_i64, { R, "0", R } },
-    { INDEX_op_brcond_i64, { R, RI } },
-
-    { INDEX_op_ext8s_i64, { R, R } },
-    { INDEX_op_ext16s_i64, { R, R } },
-    { INDEX_op_ext32s_i64, { R, R } },
-    { INDEX_op_ext8u_i64, { R, R } },
-    { INDEX_op_ext16u_i64, { R, R } },
-    { INDEX_op_ext32u_i64, { R, R } },
-    { INDEX_op_ext_i32_i64, { R, R } },
-    { INDEX_op_extu_i32_i64, { R, R } },
-    { INDEX_op_bswap16_i64, { R, R } },
-    { INDEX_op_bswap32_i64, { R, R } },
-    { INDEX_op_bswap64_i64, { R, R } },
-    { INDEX_op_not_i64, { R, R } },
-    { INDEX_op_neg_i64, { R, R } },
-
-    { INDEX_op_qemu_ld_i32, { R, L } },
-    { INDEX_op_qemu_ld_i64, { R64, L } },
-
-    { INDEX_op_qemu_st_i32, { R, S } },
-    { INDEX_op_qemu_st_i64, { R64, S } },
-
-    { INDEX_op_ext8s_i32, { R, R } },
-    { INDEX_op_ext16s_i32, { R, R } },
-    { INDEX_op_ext8u_i32, { R, R } },
-    { INDEX_op_ext16u_i32, { R, R } },
-
-    { INDEX_op_bswap16_i32, { R, R } },
-    { INDEX_op_bswap32_i32, { R, R } },
-
-    { INDEX_op_mb, { } },
-    { -1 },
-};
-
-static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
+static TCGConstraintSetIndex tcg_target_op_def(TCGOpcode op)
 {
-    int i, n = ARRAY_SIZE(tcg_target_op_defs);
+    switch (op) {
+    case INDEX_op_ld8u_i32:
+    case INDEX_op_ld8s_i32:
+    case INDEX_op_ld16u_i32:
+    case INDEX_op_ld16s_i32:
+    case INDEX_op_ld_i32:
+    case INDEX_op_ld8u_i64:
+    case INDEX_op_ld8s_i64:
+    case INDEX_op_ld16u_i64:
+    case INDEX_op_ld16s_i64:
+    case INDEX_op_ld32u_i64:
+    case INDEX_op_ld32s_i64:
+    case INDEX_op_ld_i64:
+    case INDEX_op_not_i32:
+    case INDEX_op_not_i64:
+    case INDEX_op_neg_i32:
+    case INDEX_op_neg_i64:
+    case INDEX_op_ext8s_i32:
+    case INDEX_op_ext8s_i64:
+    case INDEX_op_ext16s_i32:
+    case INDEX_op_ext16s_i64:
+    case INDEX_op_ext8u_i32:
+    case INDEX_op_ext8u_i64:
+    case INDEX_op_ext16u_i32:
+    case INDEX_op_ext16u_i64:
+    case INDEX_op_ext32s_i64:
+    case INDEX_op_ext32u_i64:
+    case INDEX_op_ext_i32_i64:
+    case INDEX_op_extu_i32_i64:
+    case INDEX_op_bswap16_i32:
+    case INDEX_op_bswap16_i64:
+    case INDEX_op_bswap32_i32:
+    case INDEX_op_bswap32_i64:
+    case INDEX_op_bswap64_i64:
+        return C_O1_I1(r, r);
 
-    for (i = 0; i < n; ++i) {
-        if (tcg_target_op_defs[i].op == op) {
-            return &tcg_target_op_defs[i];
-        }
+    case INDEX_op_st8_i32:
+    case INDEX_op_st16_i32:
+    case INDEX_op_st_i32:
+    case INDEX_op_st8_i64:
+    case INDEX_op_st16_i64:
+    case INDEX_op_st32_i64:
+    case INDEX_op_st_i64:
+        return C_O0_I2(r, r);
+
+    case INDEX_op_div_i32:
+    case INDEX_op_div_i64:
+    case INDEX_op_divu_i32:
+    case INDEX_op_divu_i64:
+    case INDEX_op_rem_i32:
+    case INDEX_op_rem_i64:
+    case INDEX_op_remu_i32:
+    case INDEX_op_remu_i64:
+        return C_O1_I2(r, r, r);
+
+    case INDEX_op_add_i32:
+    case INDEX_op_add_i64:
+    case INDEX_op_sub_i32:
+    case INDEX_op_sub_i64:
+    case INDEX_op_mul_i32:
+    case INDEX_op_mul_i64:
+    case INDEX_op_and_i32:
+    case INDEX_op_and_i64:
+    case INDEX_op_andc_i32:
+    case INDEX_op_andc_i64:
+    case INDEX_op_eqv_i32:
+    case INDEX_op_eqv_i64:
+    case INDEX_op_nand_i32:
+    case INDEX_op_nand_i64:
+    case INDEX_op_nor_i32:
+    case INDEX_op_nor_i64:
+    case INDEX_op_or_i32:
+    case INDEX_op_or_i64:
+    case INDEX_op_orc_i32:
+    case INDEX_op_orc_i64:
+    case INDEX_op_xor_i32:
+    case INDEX_op_xor_i64:
+    case INDEX_op_shl_i32:
+    case INDEX_op_shl_i64:
+    case INDEX_op_shr_i32:
+    case INDEX_op_shr_i64:
+    case INDEX_op_sar_i32:
+    case INDEX_op_sar_i64:
+    case INDEX_op_rotl_i32:
+    case INDEX_op_rotl_i64:
+    case INDEX_op_rotr_i32:
+    case INDEX_op_rotr_i64:
+        /* TODO: Does R, RI, RI result in faster code than R, R, RI? */
+        return C_O1_I2(r, ri, ri);
+
+    case INDEX_op_deposit_i32:
+    case INDEX_op_deposit_i64:
+        return C_O1_I2(r, 0, r);
+
+    case INDEX_op_brcond_i32:
+    case INDEX_op_brcond_i64:
+        return C_O0_I2(r, ri);
+
+    case INDEX_op_setcond_i32:
+    case INDEX_op_setcond_i64:
+        return C_O1_I2(r, r, ri);
+
+#if TCG_TARGET_REG_BITS == 32
+    /* TODO: Support R, R, R, R, RI, RI? Will it be faster? */
+    case INDEX_op_add2_i32:
+    case INDEX_op_sub2_i32:
+        return C_O2_I4(r, r, r, r, r, r);
+    case INDEX_op_brcond2_i32:
+        return C_O0_I4(r, r, ri, ri);
+    case INDEX_op_mulu2_i32:
+        return C_O2_I2(r, r, r, r);
+    case INDEX_op_setcond2_i32
+        return C_O1_I4(r, r, r, ri, ri);
+#endif
+
+    case INDEX_op_qemu_ld_i32:
+        return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
+                ? C_O1_I1(r, r)
+                : C_O1_I2(r, r, r));
+    case INDEX_op_qemu_ld_i64:
+        return (TCG_TARGET_REG_BITS == 64 ? C_O1_I1(r, r)
+                : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O2_I1(r, r, r)
+                : C_O2_I2(r, r, r, r));
+    case INDEX_op_qemu_st_i32:
+        return (TARGET_LONG_BITS <= TCG_TARGET_REG_BITS
+                ? C_O0_I2(r, r)
+                : C_O0_I3(r, r, r));
+    case INDEX_op_qemu_st_i64:
+        return (TCG_TARGET_REG_BITS == 64 ? C_O0_I2(r, r)
+                : TARGET_LONG_BITS <= TCG_TARGET_REG_BITS ? C_O0_I3(r, r, r)
+                : C_O0_I4(r, r, r, r));
+
+    default:
+        g_assert_not_reached();
     }
-    return NULL;
 }
 
 static const int tcg_target_reg_alloc_order[] = {
-- 
2.25.1


Re: [PATCH v3 23/24] tcg/tci: Split out constraint sets to tcg-target-con-set.h
Posted by Peter Maydell 3 years, 9 months ago
On Fri, 29 Jan 2021 at 20:26, Richard Henderson
<richard.henderson@linaro.org> wrote:
>
> This requires finishing the conversion to tcg_target_op_def.
>
> Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
> ---
>  tcg/tci/tcg-target-con-set.h |  25 ++++
>  tcg/tci/tcg-target.h         |   2 +
>  tcg/tci/tcg-target.c.inc     | 279 +++++++++++++++++------------------
>  3 files changed, 161 insertions(+), 145 deletions(-)
>  create mode 100644 tcg/tci/tcg-target-con-set.h
>
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>

thanks
-- PMM