[PULL 58/94] target/sparc: Split out fp ldst functions with asi precomputed

Richard Henderson posted 94 patches 1 year, 1 month ago
Maintainers: Richard Henderson <richard.henderson@linaro.org>, Paolo Bonzini <pbonzini@redhat.com>, Laurent Vivier <laurent@vivier.eu>, Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>, Artyom Tarasenko <atar4qemu@gmail.com>, WANG Xuerui <git@xen0n.name>, "Philippe Mathieu-Daudé" <philmd@linaro.org>, Aurelien Jarno <aurelien@aurel32.net>, Huacai Chen <chenhuacai@kernel.org>, Jiaxun Yang <jiaxun.yang@flygoat.com>, Aleksandar Rikalo <aleksandar.rikalo@syrmia.com>, Palmer Dabbelt <palmer@dabbelt.com>, Alistair Francis <Alistair.Francis@wdc.com>, Stefan Weil <sw@weilnetz.de>
There is a newer version of this series
[PULL 58/94] target/sparc: Split out fp ldst functions with asi precomputed
Posted by Richard Henderson 1 year, 1 month ago
Take the operation size from the MemOp instead of a
separate parameter.

Tested-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Acked-by: Mark Cave-Ayland <mark.cave-ayland@ilande.co.uk>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
---
 target/sparc/translate.c | 138 +++++++++++++++++++++++----------------
 1 file changed, 80 insertions(+), 58 deletions(-)

diff --git a/target/sparc/translate.c b/target/sparc/translate.c
index 363d8b7fc8..55d6f83736 100644
--- a/target/sparc/translate.c
+++ b/target/sparc/translate.c
@@ -2320,35 +2320,41 @@ static void gen_ldstub_asi(DisasContext *dc, DisasASI *da, TCGv dst, TCGv addr)
     }
 }
 
-static void __attribute__((unused))
-gen_ldf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
+static void gen_ldf_asi0(DisasContext *dc, DisasASI *da, MemOp orig_size,
+                         TCGv addr, int rd)
 {
-    DisasASI da = get_asi(dc, insn, (size == 4 ? MO_TEUL : MO_TEUQ));
+    MemOp memop = da->memop;
+    MemOp size = memop & MO_SIZE;
     TCGv_i32 d32;
     TCGv_i64 d64;
 
-    switch (da.type) {
+    /* TODO: Use 128-bit load/store below. */
+    if (size == MO_128) {
+        memop = (memop & ~MO_SIZE) | MO_64;
+    }
+
+    switch (da->type) {
     case GET_ASI_EXCP:
         break;
 
     case GET_ASI_DIRECT:
-        gen_address_mask(dc, addr);
+        memop |= MO_ALIGN_4;
         switch (size) {
-        case 4:
+        case MO_32:
             d32 = gen_dest_fpr_F(dc);
-            tcg_gen_qemu_ld_i32(d32, addr, da.mem_idx, da.memop | MO_ALIGN);
+            tcg_gen_qemu_ld_i32(d32, addr, da->mem_idx, memop);
             gen_store_fpr_F(dc, rd, d32);
             break;
-        case 8:
-            tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2], addr, da.mem_idx,
-                                da.memop | MO_ALIGN_4);
+
+        case MO_64:
+            tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2], addr, da->mem_idx, memop);
             break;
-        case 16:
+
+        case MO_128:
             d64 = tcg_temp_new_i64();
-            tcg_gen_qemu_ld_i64(d64, addr, da.mem_idx, da.memop | MO_ALIGN_4);
+            tcg_gen_qemu_ld_i64(d64, addr, da->mem_idx, memop);
             tcg_gen_addi_tl(addr, addr, 8);
-            tcg_gen_qemu_ld_i64(cpu_fpr[rd/2+1], addr, da.mem_idx,
-                                da.memop | MO_ALIGN_4);
+            tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2 + 1], addr, da->mem_idx, memop);
             tcg_gen_mov_i64(cpu_fpr[rd / 2], d64);
             break;
         default:
@@ -2358,24 +2364,19 @@ gen_ldf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
 
     case GET_ASI_BLOCK:
         /* Valid for lddfa on aligned registers only.  */
-        if (size == 8 && (rd & 7) == 0) {
-            MemOp memop;
+        if (orig_size == MO_64 && (rd & 7) == 0) {
             TCGv eight;
             int i;
 
-            gen_address_mask(dc, addr);
-
             /* The first operation checks required alignment.  */
-            memop = da.memop | MO_ALIGN_64;
             eight = tcg_constant_tl(8);
             for (i = 0; ; ++i) {
-                tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2 + i], addr,
-                                    da.mem_idx, memop);
+                tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2 + i], addr, da->mem_idx,
+                                    memop | (i == 0 ? MO_ALIGN_64 : 0));
                 if (i == 7) {
                     break;
                 }
                 tcg_gen_add_tl(addr, addr, eight);
-                memop = da.memop;
             }
         } else {
             gen_exception(dc, TT_ILL_INSN);
@@ -2384,10 +2385,9 @@ gen_ldf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
 
     case GET_ASI_SHORT:
         /* Valid for lddfa only.  */
-        if (size == 8) {
-            gen_address_mask(dc, addr);
-            tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2], addr, da.mem_idx,
-                                da.memop | MO_ALIGN);
+        if (orig_size == MO_64) {
+            tcg_gen_qemu_ld_i64(cpu_fpr[rd / 2], addr, da->mem_idx,
+                                memop | MO_ALIGN);
         } else {
             gen_exception(dc, TT_ILL_INSN);
         }
@@ -2395,8 +2395,8 @@ gen_ldf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
 
     default:
         {
-            TCGv_i32 r_asi = tcg_constant_i32(da.asi);
-            TCGv_i32 r_mop = tcg_constant_i32(da.memop | MO_ALIGN);
+            TCGv_i32 r_asi = tcg_constant_i32(da->asi);
+            TCGv_i32 r_mop = tcg_constant_i32(memop | MO_ALIGN);
 
             save_state(dc);
             /* According to the table in the UA2011 manual, the only
@@ -2404,21 +2404,23 @@ gen_ldf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
                the NO_FAULT asis.  We still need a helper for these,
                but we can just use the integer asi helper for them.  */
             switch (size) {
-            case 4:
+            case MO_32:
                 d64 = tcg_temp_new_i64();
                 gen_helper_ld_asi(d64, tcg_env, addr, r_asi, r_mop);
                 d32 = gen_dest_fpr_F(dc);
                 tcg_gen_extrl_i64_i32(d32, d64);
                 gen_store_fpr_F(dc, rd, d32);
                 break;
-            case 8:
-                gen_helper_ld_asi(cpu_fpr[rd / 2], tcg_env, addr, r_asi, r_mop);
+            case MO_64:
+                gen_helper_ld_asi(cpu_fpr[rd / 2], tcg_env, addr,
+                                  r_asi, r_mop);
                 break;
-            case 16:
+            case MO_128:
                 d64 = tcg_temp_new_i64();
                 gen_helper_ld_asi(d64, tcg_env, addr, r_asi, r_mop);
                 tcg_gen_addi_tl(addr, addr, 8);
-                gen_helper_ld_asi(cpu_fpr[rd/2+1], tcg_env, addr, r_asi, r_mop);
+                gen_helper_ld_asi(cpu_fpr[rd / 2 + 1], tcg_env, addr,
+                                  r_asi, r_mop);
                 tcg_gen_mov_i64(cpu_fpr[rd / 2], d64);
                 break;
             default:
@@ -2430,36 +2432,52 @@ gen_ldf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
 }
 
 static void __attribute__((unused))
-gen_stf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
+gen_ldf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
 {
-    DisasASI da = get_asi(dc, insn, (size == 4 ? MO_TEUL : MO_TEUQ));
+    MemOp sz = ctz32(size);
+    DisasASI da = get_asi(dc, insn, MO_TE | sz);
+
+    gen_address_mask(dc, addr);
+    gen_ldf_asi0(dc, &da, sz, addr, rd);
+}
+
+static void gen_stf_asi0(DisasContext *dc, DisasASI *da, MemOp orig_size,
+                         TCGv addr, int rd)
+{
+    MemOp memop = da->memop;
+    MemOp size = memop & MO_SIZE;
     TCGv_i32 d32;
 
-    switch (da.type) {
+    /* TODO: Use 128-bit load/store below. */
+    if (size == MO_128) {
+        memop = (memop & ~MO_SIZE) | MO_64;
+    }
+
+    switch (da->type) {
     case GET_ASI_EXCP:
         break;
 
     case GET_ASI_DIRECT:
-        gen_address_mask(dc, addr);
+        memop |= MO_ALIGN_4;
         switch (size) {
-        case 4:
+        case MO_32:
             d32 = gen_load_fpr_F(dc, rd);
-            tcg_gen_qemu_st_i32(d32, addr, da.mem_idx, da.memop | MO_ALIGN);
+            tcg_gen_qemu_st_i32(d32, addr, da->mem_idx, memop | MO_ALIGN);
             break;
-        case 8:
-            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx,
-                                da.memop | MO_ALIGN_4);
+        case MO_64:
+            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da->mem_idx,
+                                memop | MO_ALIGN_4);
             break;
-        case 16:
+        case MO_128:
             /* Only 4-byte alignment required.  However, it is legal for the
                cpu to signal the alignment fault, and the OS trap handler is
                required to fix it up.  Requiring 16-byte alignment here avoids
                having to probe the second page before performing the first
                write.  */
-            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx,
-                                da.memop | MO_ALIGN_16);
+            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da->mem_idx,
+                                memop | MO_ALIGN_16);
             tcg_gen_addi_tl(addr, addr, 8);
-            tcg_gen_qemu_st_i64(cpu_fpr[rd/2+1], addr, da.mem_idx, da.memop);
+            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2 + 1], addr, da->mem_idx, memop);
             break;
         default:
             g_assert_not_reached();
@@ -2468,24 +2486,19 @@ gen_stf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
 
     case GET_ASI_BLOCK:
         /* Valid for stdfa on aligned registers only.  */
-        if (size == 8 && (rd & 7) == 0) {
-            MemOp memop;
+        if (orig_size == MO_64 && (rd & 7) == 0) {
             TCGv eight;
             int i;
 
-            gen_address_mask(dc, addr);
-
             /* The first operation checks required alignment.  */
-            memop = da.memop | MO_ALIGN_64;
             eight = tcg_constant_tl(8);
             for (i = 0; ; ++i) {
-                tcg_gen_qemu_st_i64(cpu_fpr[rd / 2 + i], addr,
-                                    da.mem_idx, memop);
+                tcg_gen_qemu_st_i64(cpu_fpr[rd / 2 + i], addr, da->mem_idx,
+                                    memop | (i == 0 ? MO_ALIGN_64 : 0));
                 if (i == 7) {
                     break;
                 }
                 tcg_gen_add_tl(addr, addr, eight);
-                memop = da.memop;
             }
         } else {
             gen_exception(dc, TT_ILL_INSN);
@@ -2494,10 +2507,9 @@ gen_stf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
 
     case GET_ASI_SHORT:
         /* Valid for stdfa only.  */
-        if (size == 8) {
-            gen_address_mask(dc, addr);
-            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da.mem_idx,
-                                da.memop | MO_ALIGN);
+        if (orig_size == MO_64) {
+            tcg_gen_qemu_st_i64(cpu_fpr[rd / 2], addr, da->mem_idx,
+                                memop | MO_ALIGN);
         } else {
             gen_exception(dc, TT_ILL_INSN);
         }
@@ -2512,6 +2524,16 @@ gen_stf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
     }
 }
 
+static void __attribute__((unused))
+gen_stf_asi(DisasContext *dc, TCGv addr, int insn, int size, int rd)
+{
+    MemOp sz = ctz32(size);
+    DisasASI da = get_asi(dc, insn, MO_TE | sz);
+
+    gen_address_mask(dc, addr);
+    gen_stf_asi0(dc, &da, sz, addr, rd);
+}
+
 static void gen_ldda_asi(DisasContext *dc, DisasASI *da, TCGv addr, int rd)
 {
     TCGv hi = gen_dest_gpr(dc, rd);
-- 
2.34.1