[X86][SSE] Add initial support for combining (V)PMOVZX with shuffles.

llvm-svn: 288049

diff --git a/llvm/lib/Target/X86/X86ISelLowering.cpp b/llvm/lib/Target/X86/X86ISelLowering.cpp
index 6b6a1d9..bb6faab 100644 (file)
--- a/llvm/lib/Target/X86/X86ISelLowering.cpp
+++ b/llvm/lib/Target/X86/X86ISelLowering.cpp
@@ -5509,6 +5509,15 @@ static bool getFauxShuffleMask(SDValue N, SmallVectorImpl<int> &Mask,
     }
     return true;
   }
+  case X86ISD::VZEXT: {
+    // TODO - add support for VPMOVZX with smaller input vector types.
+    SDValue Op0 = N.getOperand(0);
+    if (VT.getSizeInBits() != Op0.getValueSizeInBits())
+      break;
+    DecodeZeroExtendMask(Op0.getSimpleValueType().getScalarType(), VT, Mask);
+    Ops.push_back(Op0);
+    return true;
+  }
   }
 
   return false;

diff --git a/llvm/test/CodeGen/X86/2011-10-21-widen-cmp.ll b/llvm/test/CodeGen/X86/2011-10-21-widen-cmp.ll
index f60eb5a..9e6e2f7 100644 (file)
--- a/llvm/test/CodeGen/X86/2011-10-21-widen-cmp.ll
+++ b/llvm/test/CodeGen/X86/2011-10-21-widen-cmp.ll
@@ -9,9 +9,7 @@ define void @cmp_2_floats(<2 x float> %a, <2 x float> %b) {
 ; CHECK:       # BB#0: # %entry
 ; CHECK-NEXT:    movaps %xmm0, %xmm2
 ; CHECK-NEXT:    cmpordps %xmm0, %xmm0
-; CHECK-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; CHECK-NEXT:    psllq $32, %xmm0
-; CHECK-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
+; CHECK-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0,1],zero,xmm0[1]
 ; CHECK-NEXT:    pslld $31, %xmm0
 ; CHECK-NEXT:    blendvps %xmm2, %xmm1
 ; CHECK-NEXT:    movlps %xmm1, (%rax)
@@ -26,7 +24,7 @@ entry:
 define void @cmp_2_doubles(<2 x double> %a, <2 x double> %b) {
 ; CHECK-LABEL: cmp_2_doubles:
 ; CHECK:       # BB#0: # %entry
-; CHECK-NEXT:    movapd        %xmm0, %xmm2
+; CHECK-NEXT:    movapd %xmm0, %xmm2
 ; CHECK-NEXT:    cmpordpd %xmm0, %xmm0
 ; CHECK-NEXT:    blendvpd %xmm2, %xmm1
 ; CHECK-NEXT:    movapd %xmm1, (%rax)

diff --git a/llvm/test/CodeGen/X86/avx2-vbroadcast.ll b/llvm/test/CodeGen/X86/avx2-vbroadcast.ll
index 029c21a..0d1aaac 100644 (file)
--- a/llvm/test/CodeGen/X86/avx2-vbroadcast.ll
+++ b/llvm/test/CodeGen/X86/avx2-vbroadcast.ll
@@ -279,15 +279,15 @@ define <16 x i16> @broadcast_mem_v4i16_v16i16(<4 x i16>* %ptr) {
 ; X32-AVX2-LABEL: broadcast_mem_v4i16_v16i16:
 ; X32-AVX2:       ## BB#0:
 ; X32-AVX2-NEXT:    movl {{[0-9]+}}(%esp), %eax
-; X32-AVX2-NEXT:    vpmovzxwd {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero
-; X32-AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; X32-AVX2-NEXT:    vmovsd {{.*#+}} xmm0 = mem[0],zero
+; X32-AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6,7,4,5,6,7,6,7],zero,zero
 ; X32-AVX2-NEXT:    vpbroadcastq %xmm0, %ymm0
 ; X32-AVX2-NEXT:    retl
 ;
 ; X64-AVX2-LABEL: broadcast_mem_v4i16_v16i16:
 ; X64-AVX2:       ## BB#0:
-; X64-AVX2-NEXT:    vpmovzxwd {{.*#+}} xmm0 = mem[0],zero,mem[1],zero,mem[2],zero,mem[3],zero
-; X64-AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
+; X64-AVX2-NEXT:    vmovq {{.*#+}} xmm0 = mem[0],zero
+; X64-AVX2-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,1,2,3,4,5,6,7,4,5,6,7,6,7],zero,zero
 ; X64-AVX2-NEXT:    vpbroadcastq %xmm0, %ymm0
 ; X64-AVX2-NEXT:    retq
 ;

diff --git a/llvm/test/CodeGen/X86/avx512-cvt.ll b/llvm/test/CodeGen/X86/avx512-cvt.ll
index 75ac796..884362e 100644 (file)
--- a/llvm/test/CodeGen/X86/avx512-cvt.ll
+++ b/llvm/test/CodeGen/X86/avx512-cvt.ll
@@ -923,9 +923,7 @@ define <2 x float> @sitofp_2i1_float(<2 x float> %a) {
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vxorps %xmm1, %xmm1, %xmm1
 ; KNL-NEXT:    vcmpltps %xmm0, %xmm1, %xmm0
-; KNL-NEXT:    vpmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; KNL-NEXT:    vpsllq $32, %xmm0, %xmm0
-; KNL-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
+; KNL-NEXT:    vinsertps {{.*#+}} xmm0 = xmm0[0,1],zero,xmm0[1]
 ; KNL-NEXT:    vcvtdq2ps %xmm0, %xmm0
 ; KNL-NEXT:    retq
 ;

diff --git a/llvm/test/CodeGen/X86/sse-minmax.ll b/llvm/test/CodeGen/X86/sse-minmax.ll
index b9d9e31..de85d8a 100644 (file)
--- a/llvm/test/CodeGen/X86/sse-minmax.ll
+++ b/llvm/test/CodeGen/X86/sse-minmax.ll
@@ -1337,9 +1337,7 @@ define <2 x float> @test_maxps_illegal_v2f32(<2 x float> %x, <2 x float> %y)  {
 ; STRICT-NEXT:    movaps %xmm0, %xmm2
 ; STRICT-NEXT:    movaps %xmm1, %xmm0
 ; STRICT-NEXT:    cmpleps %xmm2, %xmm0
-; STRICT-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; STRICT-NEXT:    psllq $32, %xmm0
-; STRICT-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
+; STRICT-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0,1],zero,xmm0[1]
 ; STRICT-NEXT:    pslld $31, %xmm0
 ; STRICT-NEXT:    blendvps %xmm2, %xmm1
 ; STRICT-NEXT:    movaps %xmm1, %xmm0
@@ -1360,9 +1358,7 @@ define <2 x float> @test_minps_illegal_v2f32(<2 x float> %x, <2 x float> %y)  {
 ; STRICT:       # BB#0:
 ; STRICT-NEXT:    movaps %xmm0, %xmm2
 ; STRICT-NEXT:    cmpleps %xmm1, %xmm0
-; STRICT-NEXT:    pmovzxdq {{.*#+}} xmm0 = xmm0[0],zero,xmm0[1],zero
-; STRICT-NEXT:    psllq $32, %xmm0
-; STRICT-NEXT:    pshufd {{.*#+}} xmm0 = xmm0[1,3,2,3]
+; STRICT-NEXT:    insertps {{.*#+}} xmm0 = xmm0[0,1],zero,xmm0[1]
 ; STRICT-NEXT:    pslld $31, %xmm0
 ; STRICT-NEXT:    blendvps %xmm2, %xmm1
 ; STRICT-NEXT:    movaps %xmm1, %xmm0