[X86][SSE] Sign extension for target vector sizes less than 128 bits (pt2)

RKSimon · RKSimon · commit b504263e4a7a · 2015-07-03T08:01:36.000Z
Add support for v2i8/v2i16 to v2f64 by using a sign extension to v2i32 before conversion to v2f64. Differential Revision: http://reviews.llvm.org/D10589 llvm-svn: 241325
diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -24668,16 +24668,18 @@ static SDValue PerformSINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG,
 
   // Now move on to more general possibilities.
   SDValue Op0 = N->getOperand(0);
-  EVT InVT = Op0->getValueType(0);
+  EVT VT = N->getValueType(0);
+  EVT InVT = Op0.getValueType();
+  EVT InSVT = InVT.getScalarType();
 
   // SINT_TO_FP(vXi8) -> SINT_TO_FP(SEXT(vXi8 to vXi32))
   // SINT_TO_FP(vXi16) -> SINT_TO_FP(SEXT(vXi16 to vXi32))
-  if (InVT == MVT::v8i8 || InVT == MVT::v4i8 ||
-      InVT == MVT::v8i16 || InVT == MVT::v4i16) {
+  if (InVT.isVector() && (InSVT == MVT::i8 || InSVT == MVT::i16)) {
     SDLoc dl(N);
-    MVT DstVT = MVT::getVectorVT(MVT::i32, InVT.getVectorNumElements());
+    EVT DstVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32,
+                                 InVT.getVectorNumElements());
     SDValue P = DAG.getNode(ISD::SIGN_EXTEND, dl, DstVT, Op0);
-    return DAG.getNode(ISD::SINT_TO_FP, dl, N->getValueType(0), P);
+    return DAG.getNode(ISD::SINT_TO_FP, dl, VT, P);
   }
 
   // Transform (SINT_TO_FP (i64 ...)) into an x87 operation if we have
@@ -24687,10 +24689,10 @@ static SDValue PerformSINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG,
     EVT LdVT = Ld->getValueType(0);
 
     // This transformation is not supported if the result type is f16
-    if (N->getValueType(0) == MVT::f16)
+    if (VT == MVT::f16)
       return SDValue();
 
-    if (!Ld->isVolatile() && !N->getValueType(0).isVector() &&
+    if (!Ld->isVolatile() && !VT.isVector() &&
         ISD::isNON_EXTLoad(Op0.getNode()) && Op0.hasOneUse() &&
         !Subtarget->is64Bit() && LdVT == MVT::i64) {
       SDValue FILDChain = Subtarget->getTargetLowering()->BuildFILD(
diff --git a/llvm/test/CodeGen/X86/vec_int_to_fp.ll b/llvm/test/CodeGen/X86/vec_int_to_fp.ll
@@ -50,31 +50,15 @@ define <2 x double> @sitofp_2vf64_i32(<4 x i32> %a) {
 define <2 x double> @sitofp_2vf64_i16(<8 x i16> %a) {
 ; SSE2-LABEL: sitofp_2vf64_i16:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,1,0,3]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm0 = xmm0[0,1,2,3,5,5,6,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movd %xmm1, %rax
-; SSE2-NEXT:    movswq %ax, %rax
-; SSE2-NEXT:    movd %xmm0, %rcx
-; SSE2-NEXT:    movswq %cx, %rcx
-; SSE2-NEXT:    xorps %xmm0, %xmm0
-; SSE2-NEXT:    cvtsi2sdq %rcx, %xmm0
-; SSE2-NEXT:    xorps %xmm1, %xmm1
-; SSE2-NEXT:    cvtsi2sdq %rax, %xmm1
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
+; SSE2-NEXT:    psrad $16, %xmm0
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: sitofp_2vf64_i16:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX-NEXT:    vmovq %xmm0, %rax
-; AVX-NEXT:    movswq %ax, %rax
-; AVX-NEXT:    vpextrq $1, %xmm0, %rcx
-; AVX-NEXT:    movswq %cx, %rcx
-; AVX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; AVX-NEXT:    vcvtsi2sdq %rcx, %xmm0, %xmm0
-; AVX-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm1
-; AVX-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX-NEXT:    vpmovsxwd %xmm0, %xmm0
+; AVX-NEXT:    vcvtdq2pd %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %shuf = shufflevector <8 x i16> %a, <8 x i16> undef, <2 x i32> <i32 0, i32 1>
   %cvt = sitofp <2 x i16> %shuf to <2 x double>
@@ -86,30 +70,14 @@ define <2 x double> @sitofp_2vf64_i8(<16 x i8> %a) {
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0,0,1,1,2,2,3,3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0,0,1,1]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    movd %xmm1, %rax
-; SSE2-NEXT:    movsbq %al, %rax
-; SSE2-NEXT:    movd %xmm0, %rcx
-; SSE2-NEXT:    movsbq %cl, %rcx
-; SSE2-NEXT:    xorps %xmm0, %xmm0
-; SSE2-NEXT:    cvtsi2sdq %rcx, %xmm0
-; SSE2-NEXT:    xorps %xmm1, %xmm1
-; SSE2-NEXT:    cvtsi2sdq %rax, %xmm1
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[0]
+; SSE2-NEXT:    psrad $24, %xmm0
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: sitofp_2vf64_i8:
 ; AVX:       # BB#0:
-; AVX-NEXT:    vpmovzxbq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,zero,zero,zero,zero,xmm0[1],zero,zero,zero,zero,zero,zero,zero
-; AVX-NEXT:    vmovq %xmm0, %rax
-; AVX-NEXT:    movsbq %al, %rax
-; AVX-NEXT:    vpextrq $1, %xmm0, %rcx
-; AVX-NEXT:    movsbq %cl, %rcx
-; AVX-NEXT:    vxorps %xmm0, %xmm0, %xmm0
-; AVX-NEXT:    vcvtsi2sdq %rcx, %xmm0, %xmm0
-; AVX-NEXT:    vcvtsi2sdq %rax, %xmm0, %xmm1
-; AVX-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm1[0],xmm0[0]
+; AVX-NEXT:    vpmovsxbd %xmm0, %xmm0
+; AVX-NEXT:    vcvtdq2pd %xmm0, %xmm0
 ; AVX-NEXT:    retq
   %shuf = shufflevector <16 x i8> %a, <16 x i8> undef, <2 x i32> <i32 0, i32 1>
   %cvt = sitofp <2 x i8> %shuf to <2 x double>
diff --git a/llvm/test/CodeGen/X86/widen_conv-3.ll b/llvm/test/CodeGen/X86/widen_conv-3.ll
@@ -1,5 +1,5 @@
 ; RUN: llc < %s -march=x86 -mattr=+sse4.2 | FileCheck %s
-; CHECK: cvtsi2ss
+; CHECK: cvtdq2ps
 
 ; sign to float v2i16 to v2f32
 
