[X86][SSE] Improved i8/i16 to f64 uint2fp vector conversions

Followup to D10433 and D10589 that fixes i8/i16 uint2fp vector conversions by zero extending to i32 and using the sint2fp path (unless the target does actually support uint2fp).

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@241394 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/X86/X86ISelLowering.cpp b/lib/Target/X86/X86ISelLowering.cpp
index b16bd18aefaaf49d99241aa310813c28d3369f95..3f52d2bec87bcb06087ea9381ca8267f2044b903 100644 (file)
--- a/lib/Target/X86/X86ISelLowering.cpp
+++ b/lib/Target/X86/X86ISelLowering.cpp
@@ -1611,6 +1611,7 @@ X86TargetLowering::X86TargetLowering(const X86TargetMachine &TM,
   setTargetDAGCombine(ISD::SIGN_EXTEND);
   setTargetDAGCombine(ISD::SIGN_EXTEND_INREG);
   setTargetDAGCombine(ISD::SINT_TO_FP);
+  setTargetDAGCombine(ISD::UINT_TO_FP);
   setTargetDAGCombine(ISD::SETCC);
   setTargetDAGCombine(ISD::INTRINSIC_WO_CHAIN);
   setTargetDAGCombine(ISD::BUILD_VECTOR);
@@ -24659,6 +24660,31 @@ static SDValue performVectorCompareAndMaskUnaryOpCombine(SDNode *N,
   return SDValue();
 }
 
+static SDValue PerformUINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG,
+                                        const X86Subtarget *Subtarget) {
+  SDValue Op0 = N->getOperand(0);
+  EVT VT = N->getValueType(0);
+  EVT InVT = Op0.getValueType();
+  EVT InSVT = InVT.getScalarType();
+  const TargetLowering &TLI = DAG.getTargetLoweringInfo();
+
+  // UINT_TO_FP(vXi8) -> SINT_TO_FP(ZEXT(vXi8 to vXi32))
+  // UINT_TO_FP(vXi16) -> SINT_TO_FP(ZEXT(vXi16 to vXi32))
+  if (InVT.isVector() && (InSVT == MVT::i8 || InSVT == MVT::i16)) {
+    SDLoc dl(N);
+    EVT DstVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32,
+                                 InVT.getVectorNumElements());
+    SDValue P = DAG.getNode(ISD::ZERO_EXTEND, dl, DstVT, Op0);
+
+    if (TLI.isOperationLegal(ISD::UINT_TO_FP, DstVT))
+      return DAG.getNode(ISD::UINT_TO_FP, dl, VT, P);
+
+    return DAG.getNode(ISD::SINT_TO_FP, dl, VT, P);
+  }
+
+  return SDValue();
+}
+
 static SDValue PerformSINT_TO_FPCombine(SDNode *N, SelectionDAG &DAG,
                                         const X86Subtarget *Subtarget) {
   // First try to optimize away the conversion entirely when it's
@@ -24913,6 +24939,7 @@ SDValue X86TargetLowering::PerformDAGCombine(SDNode *N,
   case ISD::STORE:          return PerformSTORECombine(N, DAG, Subtarget);
   case ISD::MSTORE:         return PerformMSTORECombine(N, DAG, Subtarget);
   case ISD::SINT_TO_FP:     return PerformSINT_TO_FPCombine(N, DAG, Subtarget);
+  case ISD::UINT_TO_FP:     return PerformUINT_TO_FPCombine(N, DAG, Subtarget);
   case ISD::FADD:           return PerformFADDCombine(N, DAG, Subtarget);
   case ISD::FSUB:           return PerformFSUBCombine(N, DAG, Subtarget);
   case X86ISD::FXOR:

diff --git a/test/CodeGen/X86/vec_int_to_fp.ll b/test/CodeGen/X86/vec_int_to_fp.ll
index 68b6a8ebe8d35c03dfb5548dd545fc75bc5e13fa..4a3d08813904027c35f4b71283df6693901c6b26 100644 (file)
--- a/test/CodeGen/X86/vec_int_to_fp.ll
+++ b/test/CodeGen/X86/vec_int_to_fp.ll
@@ -284,36 +284,19 @@ define <2 x double> @uitofp_2vf64_i32(<4 x i32> %a) {
 define <2 x double> @uitofp_2vf64_i16(<8 x i16> %a) {
 ; SSE2-LABEL: uitofp_2vf64_i16:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [1127219200,1160773632,0,0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    movapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
-; SSE2-NEXT:    subpd %xmm3, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm4, %xmm0
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT:    subpd %xmm3, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm2, %xmm1
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[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:    pand .LCPI10_0(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: uitofp_2vf64_i16:
 ; AVX:       # BB#0:
 ; AVX-NEXT:    vpmovzxwq {{.*#+}} xmm0 = xmm0[0],zero,zero,zero,xmm0[1],zero,zero,zero
-; AVX-NEXT:    vmovdqa {{.*#+}} xmm1 = [1127219200,1160773632,0,0]
-; AVX-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX-NEXT:    vmovapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
-; AVX-NEXT:    vsubpd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vhaddpd %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX-NEXT:    vsubpd %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
-; AVX-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm2[0],xmm0[0]
+; AVX-NEXT:    vpand .LCPI10_0(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; 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 = uitofp <2 x i16> %shuf to <2 x double>
@@ -323,37 +306,20 @@ define <2 x double> @uitofp_2vf64_i16(<8 x i16> %a) {
 define <2 x double> @uitofp_2vf64_i8(<16 x i8> %a) {
 ; SSE2-LABEL: uitofp_2vf64_i8:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm1 = [1127219200,1160773632,0,0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[2,3,0,1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    movapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
-; SSE2-NEXT:    subpd %xmm3, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm4, %xmm0
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm1[0],xmm2[1],xmm1[1]
-; SSE2-NEXT:    subpd %xmm3, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm2, %xmm1
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm1[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:    pand .LCPI11_0(%rip), %xmm0
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: uitofp_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:    vmovdqa {{.*#+}} xmm1 = [1127219200,1160773632,0,0]
-; AVX-NEXT:    vpunpckldq {{.*#+}} xmm2 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX-NEXT:    vmovapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
-; AVX-NEXT:    vsubpd %xmm3, %xmm2, %xmm2
-; AVX-NEXT:    vhaddpd %xmm2, %xmm2, %xmm2
-; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
-; AVX-NEXT:    vpunpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; AVX-NEXT:    vsubpd %xmm3, %xmm0, %xmm0
-; AVX-NEXT:    vhaddpd %xmm0, %xmm0, %xmm0
-; AVX-NEXT:    vunpcklpd {{.*#+}} xmm0 = xmm2[0],xmm0[0]
+; AVX-NEXT:    vpand .LCPI11_0(%rip), %xmm0, %xmm0
+; AVX-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,2,2,3]
+; 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 = uitofp <2 x i8> %shuf to <2 x double>
@@ -498,34 +464,11 @@ define <4 x double> @uitofp_4vf64_i16(<8 x i16> %a) {
 ; SSE2-LABEL: uitofp_4vf64_i16:
 ; SSE2:       # BB#0:
 ; SSE2-NEXT:    pxor %xmm1, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm0[0,1,2,1]
 ; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm3 = [1127219200,1160773632,0,0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm0[2,3,0,1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm3[0],xmm0[1],xmm3[1]
-; SSE2-NEXT:    movapd {{.*#+}} xmm4 = [4.503600e+15,1.934281e+25]
-; SSE2-NEXT:    subpd %xmm4, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm5, %xmm0
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm1 = xmm1[0],xmm3[0],xmm1[1],xmm3[1]
-; SSE2-NEXT:    subpd %xmm4, %xmm1
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm1[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm1, %xmm5
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm5[0]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm2[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm2 = xmm1[0,1,2,3,7,5,6,7]
-; SSE2-NEXT:    pand .LCPI14_2(%rip), %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm2[2,3,0,1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm2 = xmm2[0],xmm3[0],xmm2[1],xmm3[1]
-; SSE2-NEXT:    subpd %xmm4, %xmm2
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm2[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm2, %xmm1
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm3[0],xmm5[1],xmm3[1]
-; SSE2-NEXT:    subpd %xmm4, %xmm5
-; SSE2-NEXT:    pshufd {{.*#+}} xmm2 = xmm5[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm5, %xmm2
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm2
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: uitofp_4vf64_i16:
@@ -541,38 +484,13 @@ define <4 x double> @uitofp_4vf64_i16(<8 x i16> %a) {
 define <4 x double> @uitofp_4vf64_i8(<16 x i8> %a) {
 ; SSE2-LABEL: uitofp_4vf64_i8:
 ; SSE2:       # BB#0:
-; SSE2-NEXT:    movdqa %xmm0, %xmm1
-; SSE2-NEXT:    pxor %xmm2, %xmm2
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3],xmm0[4],xmm2[4],xmm0[5],xmm2[5],xmm0[6],xmm2[6],xmm0[7],xmm2[7]
-; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1],xmm0[2],xmm2[2],xmm0[3],xmm2[3]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    movdqa {{.*#+}} xmm2 = [1127219200,1160773632,0,0]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm4 = xmm0[2,3,0,1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm0 = xmm0[0],xmm2[0],xmm0[1],xmm2[1]
-; SSE2-NEXT:    movapd {{.*#+}} xmm3 = [4.503600e+15,1.934281e+25]
-; SSE2-NEXT:    subpd %xmm3, %xmm0
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm0[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm5, %xmm0
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; SSE2-NEXT:    subpd %xmm3, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm4, %xmm5
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm0 = xmm0[0],xmm5[0]
-; SSE2-NEXT:    punpcklbw {{.*#+}} xmm1 = xmm1[0,0,1,1,2,2,3,3,4,4,5,5,6,6,7,7]
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm1[0,1,2,1]
-; SSE2-NEXT:    pshuflw {{.*#+}} xmm1 = xmm1[2,1,2,3,4,5,6,7]
-; SSE2-NEXT:    pshufhw {{.*#+}} xmm4 = xmm1[0,1,2,3,7,5,6,7]
-; SSE2-NEXT:    pand .LCPI15_2(%rip), %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm5 = xmm4[2,3,0,1]
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm4 = xmm4[0],xmm2[0],xmm4[1],xmm2[1]
-; SSE2-NEXT:    subpd %xmm3, %xmm4
-; SSE2-NEXT:    pshufd {{.*#+}} xmm1 = xmm4[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm4, %xmm1
-; SSE2-NEXT:    punpckldq {{.*#+}} xmm5 = xmm5[0],xmm2[0],xmm5[1],xmm2[1]
-; SSE2-NEXT:    subpd %xmm3, %xmm5
-; SSE2-NEXT:    pshufd {{.*#+}}  xmm2 = xmm5[2,3,0,1]
-; SSE2-NEXT:    addpd %xmm5, %xmm2
-; SSE2-NEXT:    unpcklpd {{.*#+}} xmm1 = xmm1[0],xmm2[0]
+; SSE2-NEXT:    pxor %xmm1, %xmm1
+; SSE2-NEXT:    punpcklbw {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3],xmm0[4],xmm1[4],xmm0[5],xmm1[5],xmm0[6],xmm1[6],xmm0[7],xmm1[7]
+; SSE2-NEXT:    punpcklwd {{.*#+}} xmm0 = xmm0[0],xmm1[0],xmm0[1],xmm1[1],xmm0[2],xmm1[2],xmm0[3],xmm1[3]
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm2
+; SSE2-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[2,3,0,1]
+; SSE2-NEXT:    cvtdq2pd %xmm0, %xmm1
+; SSE2-NEXT:    movaps %xmm2, %xmm0
 ; SSE2-NEXT:    retq
 ;
 ; AVX-LABEL: uitofp_4vf64_i8: