Make MIPS MT implementation more cache friendly.

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@3981 c046a42c-6fe2-441c-8c8c-71466251a162

5 files changed, 59 insertions, 59 deletions

diff --git a/target-mips/cpu.h b/target-mips/cpu.h
index 8b6b8bdb0e..4e0a09d761 100644
--- a/target-mips/cpu.h
+++ b/target-mips/cpu.h
@@ -142,7 +142,7 @@ typedef struct mips_def_t mips_def_t;
 typedef struct CPUMIPSState CPUMIPSState;
 struct CPUMIPSState {
     /* General integer registers */
-    target_ulong gpr[32][MIPS_SHADOW_SET_MAX];
+    target_ulong gpr[MIPS_SHADOW_SET_MAX][32];
     /* Special registers */
     target_ulong PC[MIPS_TC_MAX];
 #if TARGET_LONG_BITS > HOST_LONG_BITS
@@ -150,9 +150,9 @@ struct CPUMIPSState {
     target_ulong t1;
     target_ulong t2;
 #endif
-    target_ulong HI[MIPS_DSP_ACC][MIPS_TC_MAX];
-    target_ulong LO[MIPS_DSP_ACC][MIPS_TC_MAX];
-    target_ulong ACX[MIPS_DSP_ACC][MIPS_TC_MAX];
+    target_ulong HI[MIPS_TC_MAX][MIPS_DSP_ACC];
+    target_ulong LO[MIPS_TC_MAX][MIPS_DSP_ACC];
+    target_ulong ACX[MIPS_TC_MAX][MIPS_DSP_ACC];
     target_ulong DSPControl[MIPS_TC_MAX];
 
     CPUMIPSMVPContext *mvp;
diff --git a/target-mips/op.c b/target-mips/op.c
index cece7f1f23..c5187fe9a3 100644
--- a/target-mips/op.c
+++ b/target-mips/op.c
@@ -255,25 +255,25 @@ void op_dup_T0 (void)
 
 void op_load_HI (void)
 {
-    T0 = env->HI[PARAM1][env->current_tc];
+    T0 = env->HI[env->current_tc][PARAM1];
     FORCE_RET();
 }
 
 void op_store_HI (void)
 {
-    env->HI[PARAM1][env->current_tc] = T0;
+    env->HI[env->current_tc][PARAM1] = T0;
     FORCE_RET();
 }
 
 void op_load_LO (void)
 {
-    T0 = env->LO[PARAM1][env->current_tc];
+    T0 = env->LO[env->current_tc][PARAM1];
     FORCE_RET();
 }
 
 void op_store_LO (void)
 {
-    env->LO[PARAM1][env->current_tc] = T0;
+    env->LO[env->current_tc][PARAM1] = T0;
     FORCE_RET();
 }
 
@@ -368,8 +368,8 @@ void op_div (void)
 void op_div (void)
 {
     if (T1 != 0) {
-        env->LO[0][env->current_tc] = (int32_t)((int64_t)(int32_t)T0 / (int32_t)T1);
-        env->HI[0][env->current_tc] = (int32_t)((int64_t)(int32_t)T0 % (int32_t)T1);
+        env->LO[env->current_tc][0] = (int32_t)((int64_t)(int32_t)T0 / (int32_t)T1);
+        env->HI[env->current_tc][0] = (int32_t)((int64_t)(int32_t)T0 % (int32_t)T1);
     }
     FORCE_RET();
 }
@@ -378,8 +378,8 @@ void op_div (void)
 void op_divu (void)
 {
     if (T1 != 0) {
-        env->LO[0][env->current_tc] = (int32_t)((uint32_t)T0 / (uint32_t)T1);
-        env->HI[0][env->current_tc] = (int32_t)((uint32_t)T0 % (uint32_t)T1);
+        env->LO[env->current_tc][0] = (int32_t)((uint32_t)T0 / (uint32_t)T1);
+        env->HI[env->current_tc][0] = (int32_t)((uint32_t)T0 % (uint32_t)T1);
     }
     FORCE_RET();
 }
@@ -447,8 +447,8 @@ void op_ddivu (void)
 void op_ddivu (void)
 {
     if (T1 != 0) {
-        env->LO[0][env->current_tc] = T0 / T1;
-        env->HI[0][env->current_tc] = T0 % T1;
+        env->LO[env->current_tc][0] = T0 / T1;
+        env->HI[env->current_tc][0] = T0 % T1;
     }
     FORCE_RET();
 }
@@ -869,26 +869,26 @@ void op_mulshiu (void)
 
 static always_inline uint64_t get_HILO (void)
 {
-    return ((uint64_t)env->HI[0][env->current_tc] << 32) |
-            ((uint64_t)(uint32_t)env->LO[0][env->current_tc]);
+    return ((uint64_t)env->HI[env->current_tc][0] << 32) |
+            ((uint64_t)(uint32_t)env->LO[env->current_tc][0]);
 }
 
 static always_inline void set_HILO (uint64_t HILO)
 {
-    env->LO[0][env->current_tc] = (int32_t)(HILO & 0xFFFFFFFF);
-    env->HI[0][env->current_tc] = (int32_t)(HILO >> 32);
+    env->LO[env->current_tc][0] = (int32_t)(HILO & 0xFFFFFFFF);
+    env->HI[env->current_tc][0] = (int32_t)(HILO >> 32);
 }
 
 static always_inline void set_HIT0_LO (uint64_t HILO)
 {
-    env->LO[0][env->current_tc] = (int32_t)(HILO & 0xFFFFFFFF);
-    T0 = env->HI[0][env->current_tc] = (int32_t)(HILO >> 32);
+    env->LO[env->current_tc][0] = (int32_t)(HILO & 0xFFFFFFFF);
+    T0 = env->HI[env->current_tc][0] = (int32_t)(HILO >> 32);
 }
 
 static always_inline void set_HI_LOT0 (uint64_t HILO)
 {
-    T0 = env->LO[0][env->current_tc] = (int32_t)(HILO & 0xFFFFFFFF);
-    env->HI[0][env->current_tc] = (int32_t)(HILO >> 32);
+    T0 = env->LO[env->current_tc][0] = (int32_t)(HILO & 0xFFFFFFFF);
+    env->HI[env->current_tc][0] = (int32_t)(HILO >> 32);
 }
 
 void op_mult (void)
@@ -1029,13 +1029,13 @@ void op_mulshiu (void)
 #if defined(TARGET_MIPS64)
 void op_dmult (void)
 {
-    CALL_FROM_TB4(muls64, &(env->LO[0][env->current_tc]), &(env->HI[0][env->current_tc]), T0, T1);
+    CALL_FROM_TB4(muls64, &(env->LO[env->current_tc][0]), &(env->HI[env->current_tc][0]), T0, T1);
     FORCE_RET();
 }
 
 void op_dmultu (void)
 {
-    CALL_FROM_TB4(mulu64, &(env->LO[0][env->current_tc]), &(env->HI[0][env->current_tc]), T0, T1);
+    CALL_FROM_TB4(mulu64, &(env->LO[env->current_tc][0]), &(env->HI[env->current_tc][0]), T0, T1);
     FORCE_RET();
 }
 #endif
@@ -1044,14 +1044,14 @@ void op_dmultu (void)
 void op_movn (void)
 {
     if (T1 != 0)
-        env->gpr[PARAM1][env->current_tc] = T0;
+        env->gpr[env->current_tc][PARAM1] = T0;
     FORCE_RET();
 }
 
 void op_movz (void)
 {
     if (T1 == 0)
-        env->gpr[PARAM1][env->current_tc] = T0;
+        env->gpr[env->current_tc][PARAM1] = T0;
     FORCE_RET();
 }
 
diff --git a/target-mips/op_helper.c b/target-mips/op_helper.c
index 07353567de..26420aa3cb 100644
--- a/target-mips/op_helper.c
+++ b/target-mips/op_helper.c
@@ -163,25 +163,25 @@ void do_dclz (void)
 #if TARGET_LONG_BITS > HOST_LONG_BITS
 static always_inline uint64_t get_HILO (void)
 {
-    return (env->HI[0][env->current_tc] << 32) | (uint32_t)env->LO[0][env->current_tc];
+    return (env->HI[env->current_tc][0] << 32) | (uint32_t)env->LO[env->current_tc][0];
 }
 
 static always_inline void set_HILO (uint64_t HILO)
 {
-    env->LO[0][env->current_tc] = (int32_t)HILO;
-    env->HI[0][env->current_tc] = (int32_t)(HILO >> 32);
+    env->LO[env->current_tc][0] = (int32_t)HILO;
+    env->HI[env->current_tc][0] = (int32_t)(HILO >> 32);
 }
 
 static always_inline void set_HIT0_LO (uint64_t HILO)
 {
-    env->LO[0][env->current_tc] = (int32_t)(HILO & 0xFFFFFFFF);
-    T0 = env->HI[0][env->current_tc] = (int32_t)(HILO >> 32);
+    env->LO[env->current_tc][0] = (int32_t)(HILO & 0xFFFFFFFF);
+    T0 = env->HI[env->current_tc][0] = (int32_t)(HILO >> 32);
 }
 
 static always_inline void set_HI_LOT0 (uint64_t HILO)
 {
-    T0 = env->LO[0][env->current_tc] = (int32_t)(HILO & 0xFFFFFFFF);
-    env->HI[0][env->current_tc] = (int32_t)(HILO >> 32);
+    T0 = env->LO[env->current_tc][0] = (int32_t)(HILO & 0xFFFFFFFF);
+    env->HI[env->current_tc][0] = (int32_t)(HILO >> 32);
 }
 
 void do_mult (void)
@@ -303,8 +303,8 @@ void do_div (void)
 {
     /* 64bit datatypes because we may see overflow/underflow. */
     if (T1 != 0) {
-        env->LO[0][env->current_tc] = (int32_t)((int64_t)(int32_t)T0 / (int32_t)T1);
-        env->HI[0][env->current_tc] = (int32_t)((int64_t)(int32_t)T0 % (int32_t)T1);
+        env->LO[env->current_tc][0] = (int32_t)((int64_t)(int32_t)T0 / (int32_t)T1);
+        env->HI[env->current_tc][0] = (int32_t)((int64_t)(int32_t)T0 % (int32_t)T1);
     }
 }
 #endif
@@ -316,12 +316,12 @@ void do_ddiv (void)
         int64_t arg0 = (int64_t)T0;
         int64_t arg1 = (int64_t)T1;
         if (arg0 == ((int64_t)-1 << 63) && arg1 == (int64_t)-1) {
-            env->LO[0][env->current_tc] = arg0;
-            env->HI[0][env->current_tc] = 0;
+            env->LO[env->current_tc][0] = arg0;
+            env->HI[env->current_tc][0] = 0;
         } else {
             lldiv_t res = lldiv(arg0, arg1);
-            env->LO[0][env->current_tc] = res.quot;
-            env->HI[0][env->current_tc] = res.rem;
+            env->LO[env->current_tc][0] = res.quot;
+            env->HI[env->current_tc][0] = res.rem;
         }
     }
 }
@@ -330,8 +330,8 @@ void do_ddiv (void)
 void do_ddivu (void)
 {
     if (T1 != 0) {
-        env->LO[0][env->current_tc] = T0 / T1;
-        env->HI[0][env->current_tc] = T0 % T1;
+        env->LO[env->current_tc][0] = T0 / T1;
+        env->HI[env->current_tc][0] = T0 % T1;
     }
 }
 #endif
@@ -627,21 +627,21 @@ void do_pmon (int function)
     function /= 2;
     switch (function) {
     case 2: /* TODO: char inbyte(int waitflag); */
-        if (env->gpr[4][env->current_tc] == 0)
-            env->gpr[2][env->current_tc] = -1;
+        if (env->gpr[env->current_tc][4] == 0)
+            env->gpr[env->current_tc][2] = -1;
         /* Fall through */
     case 11: /* TODO: char inbyte (void); */
-        env->gpr[2][env->current_tc] = -1;
+        env->gpr[env->current_tc][2] = -1;
         break;
     case 3:
     case 12:
-        printf("%c", (char)(env->gpr[4][env->current_tc] & 0xFF));
+        printf("%c", (char)(env->gpr[env->current_tc][4] & 0xFF));
         break;
     case 17:
         break;
     case 158:
         {
-            unsigned char *fmt = (void *)(unsigned long)env->gpr[4][env->current_tc];
+            unsigned char *fmt = (void *)(unsigned long)env->gpr[env->current_tc][4];
             printf("%s", fmt);
         }
         break;
diff --git a/target-mips/op_template.c b/target-mips/op_template.c
index 48b91e1be3..c683330024 100644
--- a/target-mips/op_template.c
+++ b/target-mips/op_template.c
@@ -21,31 +21,31 @@
 #if defined(REG)
 void glue(op_load_gpr_T0_gpr, REG) (void)
 {
-    T0 = env->gpr[REG][env->current_tc];
+    T0 = env->gpr[env->current_tc][REG];
     FORCE_RET();
 }
 
 void glue(op_store_T0_gpr_gpr, REG) (void)
 {
-    env->gpr[REG][env->current_tc] = T0;
+    env->gpr[env->current_tc][REG] = T0;
     FORCE_RET();
 }
 
 void glue(op_load_gpr_T1_gpr, REG) (void)
 {
-    T1 = env->gpr[REG][env->current_tc];
+    T1 = env->gpr[env->current_tc][REG];
     FORCE_RET();
 }
 
 void glue(op_store_T1_gpr_gpr, REG) (void)
 {
-    env->gpr[REG][env->current_tc] = T1;
+    env->gpr[env->current_tc][REG] = T1;
     FORCE_RET();
 }
 
 void glue(op_load_gpr_T2_gpr, REG) (void)
 {
-    T2 = env->gpr[REG][env->current_tc];
+    T2 = env->gpr[env->current_tc][REG];
     FORCE_RET();
 }
 
diff --git a/target-mips/translate.c b/target-mips/translate.c
index 1231c00cd3..65f0ba9041 100644
--- a/target-mips/translate.c
+++ b/target-mips/translate.c
@@ -6811,7 +6811,7 @@ void dump_fpu (CPUState *env)
 {
     if (loglevel) {
        fprintf(logfile, "pc=0x" TARGET_FMT_lx " HI=0x" TARGET_FMT_lx " LO=0x" TARGET_FMT_lx " ds %04x " TARGET_FMT_lx " %d\n",
-               env->PC[env->current_tc], env->HI[0][env->current_tc], env->LO[0][env->current_tc], env->hflags, env->btarget, env->bcond);
+               env->PC[env->current_tc], env->HI[env->current_tc][0], env->LO[env->current_tc][0], env->hflags, env->btarget, env->bcond);
        fpu_dump_state(env, logfile, fprintf, 0);
     }
 }
@@ -6830,16 +6830,16 @@ void cpu_mips_check_sign_extensions (CPUState *env, FILE *f,
 
     if (!SIGN_EXT_P(env->PC[env->current_tc]))
         cpu_fprintf(f, "BROKEN: pc=0x" TARGET_FMT_lx "\n", env->PC[env->current_tc]);
-    if (!SIGN_EXT_P(env->HI[0][env->current_tc]))
-        cpu_fprintf(f, "BROKEN: HI=0x" TARGET_FMT_lx "\n", env->HI[0][env->current_tc]);
-    if (!SIGN_EXT_P(env->LO[0][env->current_tc]))
-        cpu_fprintf(f, "BROKEN: LO=0x" TARGET_FMT_lx "\n", env->LO[0][env->current_tc]);
+    if (!SIGN_EXT_P(env->HI[env->current_tc][0]))
+        cpu_fprintf(f, "BROKEN: HI=0x" TARGET_FMT_lx "\n", env->HI[env->current_tc][0]);
+    if (!SIGN_EXT_P(env->LO[env->current_tc][0]))
+        cpu_fprintf(f, "BROKEN: LO=0x" TARGET_FMT_lx "\n", env->LO[env->current_tc][0]);
     if (!SIGN_EXT_P(env->btarget))
         cpu_fprintf(f, "BROKEN: btarget=0x" TARGET_FMT_lx "\n", env->btarget);
 
     for (i = 0; i < 32; i++) {
-        if (!SIGN_EXT_P(env->gpr[i][env->current_tc]))
-            cpu_fprintf(f, "BROKEN: %s=0x" TARGET_FMT_lx "\n", regnames[i], env->gpr[i][env->current_tc]);
+        if (!SIGN_EXT_P(env->gpr[env->current_tc][i]))
+            cpu_fprintf(f, "BROKEN: %s=0x" TARGET_FMT_lx "\n", regnames[i], env->gpr[env->current_tc][i]);
     }
 
     if (!SIGN_EXT_P(env->CP0_EPC))
@@ -6860,7 +6860,7 @@ void cpu_dump_state (CPUState *env, FILE *f,
     for (i = 0; i < 32; i++) {
         if ((i & 3) == 0)
             cpu_fprintf(f, "GPR%02d:", i);
-        cpu_fprintf(f, " %s " TARGET_FMT_lx, regnames[i], env->gpr[i][env->current_tc]);
+        cpu_fprintf(f, " %s " TARGET_FMT_lx, regnames[i], env->gpr[env->current_tc][i]);
         if ((i & 3) == 3)
             cpu_fprintf(f, "\n");
     }