target/i386: fix memory operand size for CVTPS2PD

CVTPS2PD only loads a half-register for memory, unlike the other
operations under 0x0F 0x5A.  "Unpack" the group into separate
emission functions instead of using gen_unary_fp_sse.

Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
(cherry picked from commit abd41884c530aa025ada253bf1a5bd0c2b808219)
Signed-off-by: Michael Tokarev <mjt@tls.msk.ru>

2 files changed, 37 insertions, 7 deletions

diff --git a/target/i386/tcg/decode-new.c.inc b/target/i386/tcg/decode-new.c.inc
index a894e874bc..528e2fdfbb 100644
--- a/target/i386/tcg/decode-new.c.inc
+++ b/target/i386/tcg/decode-new.c.inc
@@ -805,10 +805,20 @@ static void decode_sse_unary(DisasContext *s, CPUX86State *env, X86OpEntry *entr
     case 0x51: entry->gen = gen_VSQRT; break;
     case 0x52: entry->gen = gen_VRSQRT; break;
     case 0x53: entry->gen = gen_VRCP; break;
-    case 0x5A: entry->gen = gen_VCVTfp2fp; break;
     }
 }
 
+static void decode_0F5A(DisasContext *s, CPUX86State *env, X86OpEntry *entry, uint8_t *b)
+{
+    static const X86OpEntry opcodes_0F5A[4] = {
+        X86_OP_ENTRY2(VCVTPS2PD,  V,x,       W,xh, vex2),      /* VCVTPS2PD */
+        X86_OP_ENTRY2(VCVTPD2PS,  V,x,       W,x,  vex2),      /* VCVTPD2PS */
+        X86_OP_ENTRY3(VCVTSS2SD,  V,x,  H,x, W,x,  vex2_rep3), /* VCVTSS2SD */
+        X86_OP_ENTRY3(VCVTSD2SS,  V,x,  H,x, W,x,  vex2_rep3), /* VCVTSD2SS */
+    };
+    *entry = *decode_by_prefix(s, opcodes_0F5A);
+}
+
 static void decode_0F5B(DisasContext *s, CPUX86State *env, X86OpEntry *entry, uint8_t *b)
 {
     static const X86OpEntry opcodes_0F5B[4] = {
@@ -891,7 +901,7 @@ static const X86OpEntry opcodes_0F[256] = {
 
     [0x58] = X86_OP_ENTRY3(VADD,       V,x, H,x, W,x, vex2_rep3 p_00_66_f3_f2),
     [0x59] = X86_OP_ENTRY3(VMUL,       V,x, H,x, W,x, vex2_rep3 p_00_66_f3_f2),
-    [0x5a] = X86_OP_GROUP3(sse_unary,  V,x, H,x, W,x, vex2_rep3 p_00_66_f3_f2), /* CVTPS2PD */
+    [0x5a] = X86_OP_GROUP0(0F5A),
     [0x5b] = X86_OP_GROUP0(0F5B),
     [0x5c] = X86_OP_ENTRY3(VSUB,       V,x, H,x, W,x, vex2_rep3 p_00_66_f3_f2),
     [0x5d] = X86_OP_ENTRY3(VMIN,       V,x, H,x, W,x, vex2_rep3 p_00_66_f3_f2),
diff --git a/target/i386/tcg/emit.c.inc b/target/i386/tcg/emit.c.inc
index 5d31fce65d..d6a9de8b3d 100644
--- a/target/i386/tcg/emit.c.inc
+++ b/target/i386/tcg/emit.c.inc
@@ -1917,12 +1917,22 @@ static void gen_VCOMI(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
     set_cc_op(s, CC_OP_EFLAGS);
 }
 
-static void gen_VCVTfp2fp(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
+static void gen_VCVTPD2PS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 {
-    gen_unary_fp_sse(s, env, decode,
-                     gen_helper_cvtpd2ps_xmm, gen_helper_cvtps2pd_xmm,
-                     gen_helper_cvtpd2ps_ymm, gen_helper_cvtps2pd_ymm,
-                     gen_helper_cvtsd2ss, gen_helper_cvtss2sd);
+    if (s->vex_l) {
+        gen_helper_cvtpd2ps_ymm(cpu_env, OP_PTR0, OP_PTR2);
+    } else {
+        gen_helper_cvtpd2ps_xmm(cpu_env, OP_PTR0, OP_PTR2);
+    }
+}
+
+static void gen_VCVTPS2PD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
+{
+    if (s->vex_l) {
+        gen_helper_cvtps2pd_ymm(cpu_env, OP_PTR0, OP_PTR2);
+    } else {
+        gen_helper_cvtps2pd_xmm(cpu_env, OP_PTR0, OP_PTR2);
+    }
 }
 
 static void gen_VCVTPS2PH(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
@@ -1939,6 +1949,16 @@ static void gen_VCVTPS2PH(DisasContext *s, CPUX86State *env, X86DecodedInsn *dec
     }
 }
 
+static void gen_VCVTSD2SS(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
+{
+    gen_helper_cvtsd2ss(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
+}
+
+static void gen_VCVTSS2SD(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
+{
+    gen_helper_cvtss2sd(cpu_env, OP_PTR0, OP_PTR1, OP_PTR2);
+}
+
 static void gen_VCVTSI2Sx(DisasContext *s, CPUX86State *env, X86DecodedInsn *decode)
 {
     int vec_len = vector_len(s, decode);