[Qemu-devel] [PATCH 4/4] target-sparc: Don't compute full flags value so often

[Richard Henderson <rth@twiddle.net>]

Avoid speculatively computing flags before every potentially trapping
operation and instead do the flags computation when a trap actually
occurs.  This gives approximately 30% speedup in emulation.

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 linux-user/main.c           |  5 +++++
 target-sparc/int32_helper.c |  5 +++++
 target-sparc/int64_helper.c |  5 +++++
 target-sparc/translate.c    | 30 +++++++++++-------------------
 4 files changed, 26 insertions(+), 19 deletions(-)

diff --git a/linux-user/main.c b/linux-user/main.c
index 9f3476b..f4bbe69 100644
--- a/linux-user/main.c
+++ b/linux-user/main.c
@@ -1114,6 +1114,11 @@ void cpu_loop (CPUSPARCState *env)
     while (1) {
         trapnr = cpu_sparc_exec (env);
 
+        /* Compute PSR before exposing state.  */
+        if (env->cc_op != CC_OP_FLAGS) {
+            cpu_get_psr(env);
+        }
+
         switch (trapnr) {
 #ifndef TARGET_SPARC64
         case 0x88:
diff --git a/target-sparc/int32_helper.c b/target-sparc/int32_helper.c
index 9ac5aac..507c355 100644
--- a/target-sparc/int32_helper.c
+++ b/target-sparc/int32_helper.c
@@ -62,6 +62,11 @@ void do_interrupt(CPUSPARCState *env)
 {
     int cwp, intno = env->exception_index;
 
+    /* Compute PSR before exposing state.  */
+    if (env->cc_op != CC_OP_FLAGS) {
+        cpu_get_psr(env);
+    }
+
 #ifdef DEBUG_PCALL
     if (qemu_loglevel_mask(CPU_LOG_INT)) {
         static int count;
diff --git a/target-sparc/int64_helper.c b/target-sparc/int64_helper.c
index 5d0bc6c..df37aa1 100644
--- a/target-sparc/int64_helper.c
+++ b/target-sparc/int64_helper.c
@@ -64,6 +64,11 @@ void do_interrupt(CPUSPARCState *env)
     int intno = env->exception_index;
     trap_state *tsptr;
 
+    /* Compute PSR before exposing state.  */
+    if (env->cc_op != CC_OP_FLAGS) {
+        cpu_get_psr(env);
+    }
+
 #ifdef DEBUG_PCALL
     if (qemu_loglevel_mask(CPU_LOG_INT)) {
         static int count;
diff --git a/target-sparc/translate.c b/target-sparc/translate.c
index 65e6f23..6cef96b 100644
--- a/target-sparc/translate.c
+++ b/target-sparc/translate.c
@@ -1005,14 +1005,17 @@ static inline void save_npc(DisasContext *dc)
     }
 }
 
-static inline void save_state(DisasContext *dc)
+static inline void update_psr(DisasContext *dc)
 {
-    tcg_gen_movi_tl(cpu_pc, dc->pc);
-    /* flush pending conditional evaluations before exposing cpu state */
     if (dc->cc_op != CC_OP_FLAGS) {
         dc->cc_op = CC_OP_FLAGS;
         gen_helper_compute_psr(cpu_env);
     }
+}
+
+static inline void save_state(DisasContext *dc)
+{
+    tcg_gen_movi_tl(cpu_pc, dc->pc);
     save_npc(dc);
 }
 
@@ -2704,7 +2707,7 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn)
                     break;
 #ifdef TARGET_SPARC64
                 case 0x2: /* V9 rdccr */
-                    gen_helper_compute_psr(cpu_env);
+                    update_psr(dc);
                     gen_helper_rdccr(cpu_dst, cpu_env);
                     gen_movl_TN_reg(rd, cpu_dst);
                     break;
@@ -2783,10 +2786,10 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn)
 #if !defined(CONFIG_USER_ONLY)
             } else if (xop == 0x29) { /* rdpsr / UA2005 rdhpr */
 #ifndef TARGET_SPARC64
-                if (!supervisor(dc))
+                if (!supervisor(dc)) {
                     goto priv_insn;
-                gen_helper_compute_psr(cpu_env);
-                dc->cc_op = CC_OP_FLAGS;
+                }
+                update_psr(dc);
                 gen_helper_rdpsr(cpu_dst, cpu_env);
 #else
                 CHECK_IU_FEATURE(dc, HYPV);
@@ -3612,7 +3615,7 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn)
                         dc->cc_op = CC_OP_TSUBTV;
                         break;
                     case 0x24: /* mulscc */
-                        gen_helper_compute_psr(cpu_env);
+                        update_psr(dc);
                         gen_op_mulscc(cpu_dst, cpu_src1, cpu_src2);
                         gen_movl_TN_reg(rd, cpu_dst);
                         tcg_gen_movi_i32(cpu_cc_op, CC_OP_ADD);
@@ -4651,12 +4654,6 @@ static void disas_sparc_insn(DisasContext * dc, unsigned int insn)
         {
             unsigned int xop = GET_FIELD(insn, 7, 12);
 
-            /* flush pending conditional evaluations before exposing
-               cpu state */
-            if (dc->cc_op != CC_OP_FLAGS) {
-                dc->cc_op = CC_OP_FLAGS;
-                gen_helper_compute_psr(cpu_env);
-            }
             cpu_src1 = get_src1(insn, cpu_src1);
             if (xop == 0x3c || xop == 0x3e) { // V9 casa/casxa
                 rs2 = GET_FIELD(insn, 27, 31);
@@ -5507,9 +5504,4 @@ void restore_state_to_opc(CPUSPARCState *env, TranslationBlock *tb, int pc_pos)
     } else {
         env->npc = npc;
     }
-
-    /* flush pending conditional evaluations before exposing cpu state */
-    if (CC_OP != CC_OP_FLAGS) {
-        helper_compute_psr(env);
-    }
 }
-- 
1.7.11.4