Add concat_i32_i64 op.

git-svn-id: svn://svn.savannah.nongnu.org/qemu/trunk@5280 c046a42c-6fe2-441c-8c8c-71466251a162

6 files changed, 44 insertions, 51 deletions

diff --git a/target-arm/translate.c b/target-arm/translate.c
index 836b356dc8..5d71349a7f 100644
--- a/target-arm/translate.c
+++ b/target-arm/translate.c
@@ -1447,10 +1447,7 @@ static void gen_iwmmxt_movl_T0_T1_wRn(int rn)
 
 static void gen_iwmmxt_movl_wRn_T0_T1(int rn)
 {
-    tcg_gen_extu_i32_i64(cpu_V0, cpu_T[0]);
-    tcg_gen_extu_i32_i64(cpu_V1, cpu_T[0]);
-    tcg_gen_shli_i64(cpu_V1, cpu_V1, 32);
-    tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
+    tcg_gen_concat_i32_i64(cpu_V0, cpu_T[0], cpu_T[1]);
     iwmmxt_store_reg(cpu_V0, rn);
 }
 
@@ -4663,14 +4660,11 @@ static int disas_neon_data_insn(CPUState * env, DisasContext *s, uint32_t insn)
                     } else {
                         tmp = neon_load_reg(rm + pass, 0);
                         gen_neon_shift_narrow(size, tmp, tmp2, q, u);
-                        tcg_gen_extu_i32_i64(cpu_V0, tmp);
+                        tmp3 = neon_load_reg(rm + pass, 1);
+                        gen_neon_shift_narrow(size, tmp3, tmp2, q, u);
+                        tcg_gen_concat_i32_i64(cpu_V0, tmp, tmp3);
                         dead_tmp(tmp);
-                        tmp = neon_load_reg(rm + pass, 1);
-                        gen_neon_shift_narrow(size, tmp, tmp2, q, u);
-                        tcg_gen_extu_i32_i64(cpu_V1, tmp);
-                        dead_tmp(tmp);
-                        tcg_gen_shli_i64(cpu_V1, cpu_V1, 32);
-                        tcg_gen_or_i64(cpu_V0, cpu_V0, cpu_V1);
+                        dead_tmp(tmp3);
                     }
                     tmp = new_tmp();
                     if (op == 8 && !u) {
@@ -5600,7 +5594,7 @@ static void gen_addq_lo(DisasContext *s, TCGv val, int rlow)
     TCGv tmp;
     TCGv tmp2;
 
-    /* Load 64-bit value rd:rn.  */
+    /* Load value and extend to 64 bits.  */
     tmp = tcg_temp_new(TCG_TYPE_I64);
     tmp2 = load_reg(s, rlow);
     tcg_gen_extu_i32_i64(tmp, tmp2);
@@ -5612,19 +5606,16 @@ static void gen_addq_lo(DisasContext *s, TCGv val, int rlow)
 static void gen_addq(DisasContext *s, TCGv val, int rlow, int rhigh)
 {
     TCGv tmp;
-    TCGv tmp2;
+    TCGv tmpl;
+    TCGv tmph;
 
     /* Load 64-bit value rd:rn.  */
+    tmpl = load_reg(s, rlow);
+    tmph = load_reg(s, rhigh);
     tmp = tcg_temp_new(TCG_TYPE_I64);
-    tmp2 = load_reg(s, rhigh);
-    tcg_gen_extu_i32_i64(tmp, tmp2);
-    dead_tmp(tmp2);
-    tcg_gen_shli_i64(tmp, tmp, 32);
-    tcg_gen_add_i64(val, val, tmp);
-
-    tmp2 = load_reg(s, rlow);
-    tcg_gen_extu_i32_i64(tmp, tmp2);
-    dead_tmp(tmp2);
+    tcg_gen_concat_i32_i64(tmp, tmpl, tmph);
+    dead_tmp(tmpl);
+    dead_tmp(tmph);
     tcg_gen_add_i64(val, val, tmp);
 }
 
diff --git a/target-mips/translate.c b/target-mips/translate.c
index b553b5f173..3ac0f60f2f 100644
--- a/target-mips/translate.c
+++ b/target-mips/translate.c
@@ -693,13 +693,7 @@ static inline void gen_load_fpr64 (DisasContext *ctx, TCGv t, int reg)
     if (ctx->hflags & MIPS_HFLAG_F64)
         tcg_gen_mov_i64(t, fpu_fpr64[reg]);
     else {
-        TCGv r_tmp2 = tcg_temp_new(TCG_TYPE_I64);
-
-        tcg_gen_extu_i32_i64(t, fpu_fpr32[reg | 1]);
-        tcg_gen_shli_i64(t, t, 32);
-        tcg_gen_extu_i32_i64(r_tmp2, fpu_fpr32[reg & ~1]);
-        tcg_gen_or_i64(t, t, r_tmp2);
-        tcg_temp_free(r_tmp2);
+        tcg_gen_concat_i32_i64(t, fpu_fpr32[reg & ~1], fpu_fpr32[reg | 1]);
     }
 }
 
@@ -6546,22 +6540,17 @@ static void gen_farith (DisasContext *ctx, uint32_t op1,
     case FOP(38, 16):
         check_cp1_64bitmode(ctx);
         {
-            TCGv fp64_0 = tcg_temp_new(TCG_TYPE_I64);
-            TCGv fp64_1 = tcg_temp_new(TCG_TYPE_I64);
+            TCGv fp64 = tcg_temp_new(TCG_TYPE_I64);
             TCGv fp32_0 = tcg_temp_new(TCG_TYPE_I32);
             TCGv fp32_1 = tcg_temp_new(TCG_TYPE_I32);
 
             gen_load_fpr32(fp32_0, fs);
             gen_load_fpr32(fp32_1, ft);
-            tcg_gen_extu_i32_i64(fp64_0, fp32_0);
-            tcg_gen_extu_i32_i64(fp64_1, fp32_1);
-            tcg_temp_free(fp32_0);
+            tcg_gen_concat_i32_i64(fp64, fp32_0, fp32_1);
             tcg_temp_free(fp32_1);
-            tcg_gen_shli_i64(fp64_1, fp64_1, 32);
-            tcg_gen_or_i64(fp64_0, fp64_0, fp64_1);
-            tcg_temp_free(fp64_1);
-            gen_store_fpr64(ctx, fp64_0, fd);
-            tcg_temp_free(fp64_0);
+            tcg_temp_free(fp32_0);
+            gen_store_fpr64(ctx, fp64, fd);
+            tcg_temp_free(fp64);
         }
         opn = "cvt.ps.s";
         break;
diff --git a/target-ppc/translate.c b/target-ppc/translate.c
index 2d646b9afc..275f7807c9 100644
--- a/target-ppc/translate.c
+++ b/target-ppc/translate.c
@@ -5300,12 +5300,7 @@ static always_inline void gen_load_gpr64(TCGv t, int reg) {
 #if defined(TARGET_PPC64)
     tcg_gen_mov_i64(t, cpu_gpr[reg]);
 #else
-    tcg_gen_extu_i32_i64(t, cpu_gprh[reg]);
-    tcg_gen_shli_i64(t, t, 32);
-    TCGv tmp = tcg_temp_local_new(TCG_TYPE_I64);
-    tcg_gen_extu_i32_i64(tmp, cpu_gpr[reg]);
-    tcg_gen_or_i64(t, t, tmp);
-    tcg_temp_free(tmp);
+    tcg_gen_concat_i32_i64(t, cpu_gpr[reg], cpu_gprh[reg]);
 #endif
 }
 
diff --git a/target-sh4/translate.c b/target-sh4/translate.c
index b9dcbc2f03..4614c8674b 100644
--- a/target-sh4/translate.c
+++ b/target-sh4/translate.c
@@ -400,15 +400,12 @@ static inline void gen_load_fpr32(TCGv t, int reg)
 static inline void gen_load_fpr64(TCGv t, int reg)
 {
     TCGv tmp1 = tcg_temp_new(TCG_TYPE_I32);
-    TCGv tmp2 = tcg_temp_new(TCG_TYPE_I64);
+    TCGv tmp2 = tcg_temp_new(TCG_TYPE_I32);
 
     tcg_gen_ld_i32(tmp1, cpu_env, offsetof(CPUState, fregs[reg]));
-    tcg_gen_extu_i32_i64(t, tmp1);
-    tcg_gen_shli_i64(t, t, 32);
-    tcg_gen_ld_i32(tmp1, cpu_env, offsetof(CPUState, fregs[reg + 1]));
-    tcg_gen_extu_i32_i64(tmp2, tmp1);
+    tcg_gen_ld_i32(tmp2, cpu_env, offsetof(CPUState, fregs[reg + 1]));
+    tcg_gen_concat_i32_i64(t, tmp2, tmp1);
     tcg_temp_free(tmp1);
-    tcg_gen_or_i64(t, t, tmp2);
     tcg_temp_free(tmp2);
 }
 
diff --git a/tcg/README b/tcg/README
index b03432e23a..b4d07b6f42 100644
--- a/tcg/README
+++ b/tcg/README
@@ -265,6 +265,10 @@ Convert t1 (32 bit) to t0 (64 bit) and does zero extension
 * trunc_i64_i32 t0, t1
 Truncate t1 (64 bit) to t0 (32 bit)
 
+* concat_i32_i64 t0, t1, t2
+Construct t0 (64-bit) taking the low half from t1 (32 bit) and the high half
+from t2 (32 bit).
+
 ********* Load/Store
 
 * ld_i32/i64 t0, t1, offset
diff --git a/tcg/tcg-op.h b/tcg/tcg-op.h
index bc6be8537e..7b39fe865a 100644
--- a/tcg/tcg-op.h
+++ b/tcg/tcg-op.h
@@ -1395,6 +1395,23 @@ static inline void tcg_gen_discard_i64(TCGv arg)
 }
 #endif
 
+static inline void tcg_gen_concat_i32_i64(TCGv dest, TCGv low, TCGv high)
+{
+#if TCG_TARGET_REG_BITS == 32
+    tcg_gen_mov_i32(dest, low);
+    tcg_gen_mov_i32(TCGV_HIGH(dest), high);
+#else
+    TCGv tmp = tcg_temp_new (TCG_TYPE_I64);
+    /* This extension is only needed for type correctness.
+       We may be able to do better given target specific information.  */
+    tcg_gen_extu_i32_i64(tmp, high);
+    tcg_gen_shli_i64(tmp, tmp, 32);
+    tcg_gen_extu_i32_i64(dest, low);
+    tcg_gen_or_i64(dest, dest, tmp);
+    tcg_temp_free(tmp);
+#endif
+}
+
 /***************************************/
 /* QEMU specific operations. Their type depend on the QEMU CPU
    type. */