[Qemu-devel] [PULL 6/8] target-sh4: split out Q and M from of SR and optimize div1

[Aurelien Jarno]

Splitting Q and M out of SR, it's possible to optimize div1 by using
TCG code instead of an helper.

Signed-off-by: Aurelien Jarno <address@hidden>
---
 target-sh4/cpu.h       |  12 +++--
 target-sh4/helper.h    |   1 -
 target-sh4/op_helper.c | 118 -------------------------------------------------
 target-sh4/translate.c |  70 ++++++++++++++++++++++++-----
 4 files changed, 69 insertions(+), 132 deletions(-)

diff --git a/target-sh4/cpu.h b/target-sh4/cpu.h
index a308c53..4a027a6 100644
--- a/target-sh4/cpu.h
+++ b/target-sh4/cpu.h
@@ -139,6 +139,8 @@ typedef struct CPUSH4State {
     uint32_t gregs[24];                /* general registers */
     float32 fregs[32];         /* floating point registers */
     uint32_t sr;                /* status register (with T split out) */
+    uint32_t sr_m;              /* M bit of status register */
+    uint32_t sr_q;              /* Q bit of status register */
     uint32_t sr_t;              /* T bit of status register */
     uint32_t ssr;              /* saved status register */
     uint32_t spc;              /* saved program counter */
@@ -334,13 +336,17 @@ static inline int cpu_ptel_pr (uint32_t ptel)
 
 static inline target_ulong cpu_read_sr(CPUSH4State *env)
 {
-    return env->sr | (env->sr_t << SR_T);
+    return env->sr | (env->sr_m << SR_M) |
+                     (env->sr_q << SR_Q) |
+                     (env->sr_t << SR_T);
 }
 
 static inline void cpu_write_sr(CPUSH4State *env, target_ulong sr)
 {
-    env->sr_t = sr & (1u << SR_T);
-    env->sr = sr & ~(1u << SR_T);
+    env->sr_m = (sr >> SR_M) & 1;
+    env->sr_q = (sr >> SR_Q) & 1;
+    env->sr_t = (sr >> SR_T) & 1;
+    env->sr = sr & ~((1u << SR_M) | (1u << SR_Q) | (1u << SR_T));
 }
 
 static inline void cpu_get_tb_cpu_state(CPUSH4State *env, target_ulong *pc,
diff --git a/target-sh4/helper.h b/target-sh4/helper.h
index 3b5c436..c9bc407 100644
--- a/target-sh4/helper.h
+++ b/target-sh4/helper.h
@@ -11,7 +11,6 @@ DEF_HELPER_3(movcal, void, env, i32, i32)
 DEF_HELPER_1(discard_movcal_backup, void, env)
 DEF_HELPER_2(ocbi, void, env, i32)
 
-DEF_HELPER_3(div1, i32, env, i32, i32)
 DEF_HELPER_3(macl, void, env, i32, i32)
 DEF_HELPER_3(macw, void, env, i32, i32)
 
diff --git a/target-sh4/op_helper.c b/target-sh4/op_helper.c
index 524d7f6..cbc11ae 100644
--- a/target-sh4/op_helper.c
+++ b/target-sh4/op_helper.c
@@ -156,124 +156,6 @@ void helper_ocbi(CPUSH4State *env, uint32_t address)
     }
 }
 
-#define T (env->sr_t)
-#define Q (env->sr & (1u << SR_Q) ? 1 : 0)
-#define M (env->sr & (1u << SR_M) ? 1 : 0)
-#define SETT (env->sr_t = 1)
-#define CLRT (env->sr_t = 0)
-#define SETQ (env->sr |= (1u << SR_Q))
-#define CLRQ (env->sr &= ~(1u << SR_Q))
-#define SETM (env->sr |= (1u << SR_M))
-#define CLRM (env->sr &= ~(1u << SR_M))
-
-uint32_t helper_div1(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
-{
-    uint32_t tmp0, tmp2;
-    uint8_t old_q, tmp1 = 0xff;
-
-    //printf("div1 arg0=0x%08x arg1=0x%08x M=%d Q=%d T=%d\n", arg0, arg1, M, 
Q, T);
-    old_q = Q;
-    if ((0x80000000 & arg1) != 0)
-       SETQ;
-    else
-       CLRQ;
-    tmp2 = arg0;
-    arg1 <<= 1;
-    arg1 |= T;
-    switch (old_q) {
-    case 0:
-       switch (M) {
-       case 0:
-           tmp0 = arg1;
-           arg1 -= tmp2;
-           tmp1 = arg1 > tmp0;
-           switch (Q) {
-           case 0:
-               if (tmp1)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           case 1:
-               if (tmp1 == 0)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           }
-           break;
-       case 1:
-           tmp0 = arg1;
-           arg1 += tmp2;
-           tmp1 = arg1 < tmp0;
-           switch (Q) {
-           case 0:
-               if (tmp1 == 0)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           case 1:
-               if (tmp1)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           }
-           break;
-       }
-       break;
-    case 1:
-       switch (M) {
-       case 0:
-           tmp0 = arg1;
-           arg1 += tmp2;
-           tmp1 = arg1 < tmp0;
-           switch (Q) {
-           case 0:
-               if (tmp1)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           case 1:
-               if (tmp1 == 0)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           }
-           break;
-       case 1:
-           tmp0 = arg1;
-           arg1 -= tmp2;
-           tmp1 = arg1 > tmp0;
-           switch (Q) {
-           case 0:
-               if (tmp1 == 0)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           case 1:
-               if (tmp1)
-                   SETQ;
-               else
-                   CLRQ;
-               break;
-           }
-           break;
-       }
-       break;
-    }
-    if (Q == M)
-       SETT;
-    else
-       CLRT;
-    //printf("Output: arg1=0x%08x M=%d Q=%d T=%d\n", arg1, M, Q, T);
-    return arg1;
-}
-
 void helper_macl(CPUSH4State *env, uint32_t arg0, uint32_t arg1)
 {
     int64_t res;
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index 250632a..daea268 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -62,7 +62,8 @@ enum {
 /* global register indexes */
 static TCGv_ptr cpu_env;
 static TCGv cpu_gregs[24];
-static TCGv cpu_pc, cpu_sr, cpu_sr_t, cpu_ssr, cpu_spc, cpu_gbr;
+static TCGv cpu_sr, cpu_sr_m, cpu_sr_q, cpu_sr_t;
+static TCGv cpu_pc, cpu_ssr, cpu_spc, cpu_gbr;
 static TCGv cpu_vbr, cpu_sgr, cpu_dbr, cpu_mach, cpu_macl;
 static TCGv cpu_pr, cpu_fpscr, cpu_fpul, cpu_ldst;
 static TCGv cpu_fregs[32];
@@ -110,6 +111,10 @@ void sh4_translate_init(void)
                                     offsetof(CPUSH4State, pc), "PC");
     cpu_sr = tcg_global_mem_new_i32(TCG_AREG0,
                                     offsetof(CPUSH4State, sr), "SR");
+    cpu_sr_m = tcg_global_mem_new_i32(TCG_AREG0,
+                                    offsetof(CPUSH4State, sr_m), "SR_M");
+    cpu_sr_q = tcg_global_mem_new_i32(TCG_AREG0,
+                                    offsetof(CPUSH4State, sr_q), "SR_Q");
     cpu_sr_t = tcg_global_mem_new_i32(TCG_AREG0,
                                     offsetof(CPUSH4State, sr_t), "SR_T");
     cpu_ssr = tcg_global_mem_new_i32(TCG_AREG0,
@@ -178,13 +183,26 @@ void superh_cpu_dump_state(CPUState *cs, FILE *f,
 }
 static void gen_read_sr(TCGv dst)
 {
-    tcg_gen_or_i32(dst, cpu_sr, cpu_sr_t);
+    TCGv t0 = tcg_temp_new();
+    tcg_gen_shli_i32(t0, cpu_sr_q, SR_Q);
+    tcg_gen_or_i32(dst, dst, t0);
+    tcg_gen_shli_i32(t0, cpu_sr_m, SR_M);
+    tcg_gen_or_i32(dst, dst, t0);
+    tcg_gen_shli_i32(t0, cpu_sr_t, SR_T);
+    tcg_gen_or_i32(dst, cpu_sr, t0);
+    tcg_temp_free_i32(t0);
 }
 
 static void gen_write_sr(TCGv src)
 {
-    tcg_gen_andi_i32(cpu_sr, src, ~(1u << SR_T));
-    tcg_gen_andi_i32(cpu_sr_t, src, (1u << SR_T));
+    tcg_gen_andi_i32(cpu_sr, src,
+                     ~((1u << SR_Q) | (1u << SR_M) | (1u << SR_T)));
+    tcg_gen_shri_i32(cpu_sr_q, src, SR_Q);
+    tcg_gen_andi_i32(cpu_sr_q, cpu_sr_q, 1);
+    tcg_gen_shri_i32(cpu_sr_m, src, SR_M);
+    tcg_gen_andi_i32(cpu_sr_m, cpu_sr_m, 1);
+    tcg_gen_shri_i32(cpu_sr_t, src, SR_T);
+    tcg_gen_andi_i32(cpu_sr_t, cpu_sr_t, 1);
 }
 
 static void gen_goto_tb(DisasContext * ctx, int n, target_ulong dest)
@@ -391,7 +409,8 @@ static void _decode_opc(DisasContext * ctx)
 
     switch (ctx->opcode) {
     case 0x0019:               /* div0u */
-        tcg_gen_andi_i32(cpu_sr, cpu_sr, ~((1u << SR_M) | (1u << SR_Q)));
+        tcg_gen_movi_i32(cpu_sr_m, 0);
+        tcg_gen_movi_i32(cpu_sr_q, 0);
         tcg_gen_movi_i32(cpu_sr_t, 0);
        return;
     case 0x000b:               /* rts */
@@ -709,13 +728,44 @@ static void _decode_opc(DisasContext * ctx)
        }
        return;
     case 0x2007:               /* div0s Rm,Rn */
-        gen_copy_bit_i32(cpu_sr, SR_Q, REG(B11_8), 31);     /* SR_Q */
-        gen_copy_bit_i32(cpu_sr, SR_M, REG(B7_4), 31);      /* SR_M */
-        tcg_gen_xor_i32(cpu_sr_t, REG(B7_4), REG(B11_8));
-        tcg_gen_shri_i32(cpu_sr_t, cpu_sr_t, 31);           /* SR_T */
+        tcg_gen_shri_i32(cpu_sr_q, REG(B11_8), 31);         /* SR_Q */
+        tcg_gen_shri_i32(cpu_sr_m, REG(B7_4), 31);          /* SR_M */
+        tcg_gen_xor_i32(cpu_sr_t, cpu_sr_q, cpu_sr_m);      /* SR_T */
        return;
     case 0x3004:               /* div1 Rm,Rn */
-        gen_helper_div1(REG(B11_8), cpu_env, REG(B7_4), REG(B11_8));
+        {
+            TCGv t0 = tcg_temp_new();
+            TCGv t1 = tcg_temp_new();
+            TCGv t2 = tcg_temp_new();
+            TCGv zero = tcg_const_i32(0);
+
+            /* shift left arg1, saving the bit being pushed out and inserting
+               T on the right */
+            tcg_gen_shri_i32(t0, REG(B11_8), 31);
+            tcg_gen_shli_i32(REG(B11_8), REG(B11_8), 1);
+            tcg_gen_or_i32(REG(B11_8), REG(B11_8), cpu_sr_t);
+
+            /* Add or subtract arg0 from arg1 depending if Q == M. To avoid
+               using 64-bit temps, we compute arg0's high part from q ^ m, so
+               that it is 0x00000000 when adding the value or 0xffffffff when
+               subtracting it. */
+            tcg_gen_xor_i32(t1, cpu_sr_q, cpu_sr_m);
+            tcg_gen_subi_i32(t1, t1, 1);
+            tcg_gen_neg_i32(t2, REG(B7_4));
+            tcg_gen_movcond_i32(TCG_COND_EQ, t2, t1, zero, REG(B7_4), t2);
+            tcg_gen_add2_i32(REG(B11_8), t1, REG(B11_8), zero, t2, t1);
+
+            /* compute T and Q depending on carry */
+            tcg_gen_andi_i32(t1, t1, 1);
+            tcg_gen_xor_i32(t1, t1, t0);
+            tcg_gen_xori_i32(cpu_sr_t, t1, 1);
+            tcg_gen_xor_i32(cpu_sr_q, cpu_sr_m, t1);
+
+            tcg_temp_free(zero);
+            tcg_temp_free(t2);
+            tcg_temp_free(t1);
+            tcg_temp_free(t0);
+        }
        return;
     case 0x300d:               /* dmuls.l Rm,Rn */
         tcg_gen_muls2_i32(cpu_macl, cpu_mach, REG(B7_4), REG(B11_8));
-- 
2.1.4