[Qemu-devel] [PATCH v2 16/27] target/sh4: Load/store Dr as 64-bit quantities

[Richard Henderson <rth@twiddle.net>]

This enforces proper alignment and makes the register update
more natural.  Note that there is a more serious bug fix for
fmov {DX}Rn,@(R0,Rn) to use a store instead of a load.

Signed-off-by: Richard Henderson <rth@twiddle.net>
---
 target/sh4/translate.c | 74 ++++++++++++++++++++++++--------------------------
 1 file changed, 35 insertions(+), 39 deletions(-)

diff --git a/target/sh4/translate.c b/target/sh4/translate.c
index 616e615..fcdabe8 100644
--- a/target/sh4/translate.c
+++ b/target/sh4/translate.c
@@ -1004,12 +1004,10 @@ static void _decode_opc(DisasContext * ctx)
     case 0xf00a: /* fmov {F,D,X}Rm,@Rn - FPSCR: Nothing */
 	CHECK_FPU_ENABLED
         if (ctx->tbflags & FPSCR_SZ) {
-	    TCGv addr_hi = tcg_temp_new();
-	    int fr = XHACK(B7_4);
-	    tcg_gen_addi_i32(addr_hi, REG(B11_8), 4);
-            tcg_gen_qemu_st_i32(FREG(fr), REG(B11_8), ctx->memidx, MO_TEUL);
-            tcg_gen_qemu_st_i32(FREG(fr + 1), addr_hi, ctx->memidx, MO_TEUL);
-	    tcg_temp_free(addr_hi);
+            TCGv_i64 fp = tcg_temp_new_i64();
+            gen_load_fpr64(ctx, fp, XHACK(B7_4));
+            tcg_gen_qemu_st_i64(fp, REG(B11_8), ctx->memidx, MO_TEQ);
+            tcg_temp_free_i64(fp);
 	} else {
             tcg_gen_qemu_st_i32(FREG(B7_4), REG(B11_8), ctx->memidx, MO_TEUL);
 	}
@@ -1017,12 +1015,10 @@ static void _decode_opc(DisasContext * ctx)
     case 0xf008: /* fmov @Rm,{F,D,X}Rn - FPSCR: Nothing */
 	CHECK_FPU_ENABLED
         if (ctx->tbflags & FPSCR_SZ) {
-	    TCGv addr_hi = tcg_temp_new();
-	    int fr = XHACK(B11_8);
-	    tcg_gen_addi_i32(addr_hi, REG(B7_4), 4);
-            tcg_gen_qemu_ld_i32(FREG(fr), REG(B7_4), ctx->memidx, MO_TEUL);
-            tcg_gen_qemu_ld_i32(FREG(fr + 1), addr_hi, ctx->memidx, MO_TEUL);
-	    tcg_temp_free(addr_hi);
+            TCGv_i64 fp = tcg_temp_new_i64();
+            tcg_gen_qemu_ld_i64(fp, REG(B7_4), ctx->memidx, MO_TEQ);
+            gen_store_fpr64(ctx, fp, XHACK(B11_8));
+            tcg_temp_free_i64(fp);
 	} else {
             tcg_gen_qemu_ld_i32(FREG(B11_8), REG(B7_4), ctx->memidx, MO_TEUL);
 	}
@@ -1030,13 +1026,11 @@ static void _decode_opc(DisasContext * ctx)
     case 0xf009: /* fmov @Rm+,{F,D,X}Rn - FPSCR: Nothing */
 	CHECK_FPU_ENABLED
         if (ctx->tbflags & FPSCR_SZ) {
-	    TCGv addr_hi = tcg_temp_new();
-	    int fr = XHACK(B11_8);
-	    tcg_gen_addi_i32(addr_hi, REG(B7_4), 4);
-            tcg_gen_qemu_ld_i32(FREG(fr), REG(B7_4), ctx->memidx, MO_TEUL);
-            tcg_gen_qemu_ld_i32(FREG(fr + 1), addr_hi, ctx->memidx, MO_TEUL);
-	    tcg_gen_addi_i32(REG(B7_4), REG(B7_4), 8);
-	    tcg_temp_free(addr_hi);
+            TCGv_i64 fp = tcg_temp_new_i64();
+            tcg_gen_qemu_ld_i64(fp, REG(B7_4), ctx->memidx, MO_TEQ);
+            gen_store_fpr64(ctx, fp, XHACK(B11_8));
+            tcg_temp_free_i64(fp);
+            tcg_gen_addi_i32(REG(B7_4), REG(B7_4), 8);
 	} else {
             tcg_gen_qemu_ld_i32(FREG(B11_8), REG(B7_4), ctx->memidx, MO_TEUL);
 	    tcg_gen_addi_i32(REG(B7_4), REG(B7_4), 4);
@@ -1044,18 +1038,20 @@ static void _decode_opc(DisasContext * ctx)
 	return;
     case 0xf00b: /* fmov {F,D,X}Rm,@-Rn - FPSCR: Nothing */
 	CHECK_FPU_ENABLED
-        TCGv addr = tcg_temp_new_i32();
-        tcg_gen_subi_i32(addr, REG(B11_8), 4);
-        if (ctx->tbflags & FPSCR_SZ) {
-	    int fr = XHACK(B7_4);
-            tcg_gen_qemu_st_i32(FREG(fr + 1), addr, ctx->memidx, MO_TEUL);
-	    tcg_gen_subi_i32(addr, addr, 4);
-            tcg_gen_qemu_st_i32(FREG(fr), addr, ctx->memidx, MO_TEUL);
-	} else {
-            tcg_gen_qemu_st_i32(FREG(B7_4), addr, ctx->memidx, MO_TEUL);
-	}
-        tcg_gen_mov_i32(REG(B11_8), addr);
-        tcg_temp_free(addr);
+        {
+            TCGv addr = tcg_temp_new_i32();
+            if (ctx->tbflags & FPSCR_SZ) {
+                TCGv_i64 fp = tcg_temp_new_i64();
+                gen_load_fpr64(ctx, fp, XHACK(B7_4));
+                tcg_gen_qemu_st_i64(fp, addr, ctx->memidx, MO_TEQ);
+                tcg_temp_free_i64(fp);
+            } else {
+                tcg_gen_subi_i32(addr, REG(B11_8), 4);
+                tcg_gen_qemu_st_i32(FREG(B7_4), addr, ctx->memidx, MO_TEUL);
+            }
+            tcg_gen_mov_i32(REG(B11_8), addr);
+            tcg_temp_free(addr);
+        }
 	return;
     case 0xf006: /* fmov @(R0,Rm),{F,D,X}Rm - FPSCR: Nothing */
 	CHECK_FPU_ENABLED
@@ -1063,10 +1059,10 @@ static void _decode_opc(DisasContext * ctx)
 	    TCGv addr = tcg_temp_new_i32();
 	    tcg_gen_add_i32(addr, REG(B7_4), REG(0));
             if (ctx->tbflags & FPSCR_SZ) {
-		int fr = XHACK(B11_8);
-                tcg_gen_qemu_ld_i32(FREG(fr), addr, ctx->memidx, MO_TEUL);
-		tcg_gen_addi_i32(addr, addr, 4);
-                tcg_gen_qemu_ld_i32(FREG(fr + 1), addr, ctx->memidx, MO_TEUL);
+                TCGv_i64 fp = tcg_temp_new_i64();
+                tcg_gen_qemu_ld_i64(fp, addr, ctx->memidx, MO_TEQ);
+                gen_store_fpr64(ctx, fp, XHACK(B11_8));
+                tcg_temp_free_i64(fp);
 	    } else {
                 tcg_gen_qemu_ld_i32(FREG(B11_8), addr, ctx->memidx, MO_TEUL);
 	    }
@@ -1079,10 +1075,10 @@ static void _decode_opc(DisasContext * ctx)
 	    TCGv addr = tcg_temp_new();
 	    tcg_gen_add_i32(addr, REG(B11_8), REG(0));
             if (ctx->tbflags & FPSCR_SZ) {
-		int fr = XHACK(B7_4);
-                tcg_gen_qemu_ld_i32(FREG(fr), addr, ctx->memidx, MO_TEUL);
-		tcg_gen_addi_i32(addr, addr, 4);
-                tcg_gen_qemu_ld_i32(FREG(fr + 1), addr, ctx->memidx, MO_TEUL);
+                TCGv_i64 fp = tcg_temp_new_i64();
+                gen_load_fpr64(ctx, fp, XHACK(B7_4));
+                tcg_gen_qemu_st_i64(fp, addr, ctx->memidx, MO_TEQ);
+                tcg_temp_free_i64(fp);
 	    } else {
                 tcg_gen_qemu_st_i32(FREG(B7_4), addr, ctx->memidx, MO_TEUL);
 	    }
-- 
2.9.4