[TargetLowering] Add ISD::EXTRACT_VECTOR_ELT support to SimplifyDemandedBits

Let SimplifyDemandedBits attempt to simplify all elements of a vector extraction.

Part of PR39689.

llvm-svn: 348839

diff --git a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
index 1f5ea31..6d0d392 100644 (file)
--- a/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
+++ b/llvm/lib/CodeGen/SelectionDAG/TargetLowering.cpp
@@ -1220,6 +1220,25 @@ bool TargetLowering::SimplifyDemandedBits(SDValue Op,
     Known.Zero |= ~InMask;
     break;
   }
+  case ISD::EXTRACT_VECTOR_ELT: {
+    // Demand the bits from every vector element.
+    SDValue Src = Op.getOperand(0);
+    unsigned EltBitWidth = Src.getScalarValueSizeInBits();
+
+    // If BitWidth > EltBitWidth the value is anyext:ed. So we do not know
+    // anything about the extended bits.
+    APInt DemandedSrcBits = DemandedBits;
+    if (BitWidth > EltBitWidth)
+      DemandedSrcBits = DemandedSrcBits.trunc(EltBitWidth);
+
+    if (SimplifyDemandedBits(Src, DemandedSrcBits, Known2, TLO, Depth + 1))
+        return true;
+
+    Known = Known2;
+    if (BitWidth > EltBitWidth)
+      Known = Known.zext(BitWidth);
+    break;
+  }
   case ISD::BITCAST: {
     SDValue Src = Op.getOperand(0);
     EVT SrcVT = Src.getValueType();

diff --git a/llvm/test/CodeGen/X86/known-signbits-vector.ll b/llvm/test/CodeGen/X86/known-signbits-vector.ll
index 306021f..e02ab5b 100644 (file)
--- a/llvm/test/CodeGen/X86/known-signbits-vector.ll
+++ b/llvm/test/CodeGen/X86/known-signbits-vector.ll
@@ -74,11 +74,9 @@ define float @signbits_ashr_extract_sitofp_0(<2 x i64> %a0) nounwind {
 ;
 ; X64-LABEL: signbits_ashr_extract_sitofp_0:
 ; X64:       # %bb.0:
-; X64-NEXT:    vpsrad $31, %xmm0, %xmm1
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[1,1,3,3]
-; X64-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
+; X64-NEXT:    vpsrlq $32, %xmm0, %xmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
-; X64-NEXT:    vcvtsi2ssl %eax, %xmm2, %xmm0
+; X64-NEXT:    vcvtsi2ssl %eax, %xmm1, %xmm0
 ; X64-NEXT:    retq
   %1 = ashr <2 x i64> %a0, <i64 32, i64 32>
   %2 = extractelement <2 x i64> %1, i32 0
@@ -183,9 +181,7 @@ define float @signbits_ashr_insert_ashr_extract_sitofp(i64 %a0, i64 %a1) nounwin
 ; X64-NEXT:    vmovq %rsi, %xmm0
 ; X64-NEXT:    vmovq %rdi, %xmm1
 ; X64-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm1[0],xmm0[0]
-; X64-NEXT:    vpsrad $3, %xmm0, %xmm1
 ; X64-NEXT:    vpsrlq $3, %xmm0, %xmm0
-; X64-NEXT:    vpblendw {{.*#+}} xmm0 = xmm0[0,1],xmm1[2,3],xmm0[4,5],xmm1[6,7]
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vcvtsi2ssl %eax, %xmm2, %xmm0
 ; X64-NEXT:    retq
@@ -282,10 +278,6 @@ define float @signbits_ashr_sext_sextinreg_and_extract_sitofp(<2 x i64> %a0, <2
 ; X64-NEXT:    vpsubq %xmm2, %xmm0, %xmm0
 ; X64-NEXT:    movslq %edi, %rax
 ; X64-NEXT:    vpinsrq $0, %rax, %xmm1, %xmm1
-; X64-NEXT:    vpsllq $20, %xmm1, %xmm1
-; X64-NEXT:    vpsrad $20, %xmm1, %xmm2
-; X64-NEXT:    vpsrlq $20, %xmm1, %xmm1
-; X64-NEXT:    vpblendw {{.*#+}} xmm1 = xmm1[0,1],xmm2[2,3],xmm1[4,5],xmm2[6,7]
 ; X64-NEXT:    vpand %xmm1, %xmm0, %xmm0
 ; X64-NEXT:    vmovq %xmm0, %rax
 ; X64-NEXT:    vcvtsi2ssl %eax, %xmm3, %xmm0

diff --git a/llvm/test/CodeGen/X86/pr35918.ll b/llvm/test/CodeGen/X86/pr35918.ll
index 1d422ad..6ce9706 100644 (file)
--- a/llvm/test/CodeGen/X86/pr35918.ll
+++ b/llvm/test/CodeGen/X86/pr35918.ll
@@ -16,7 +16,7 @@ define void @fetch_r16g16_snorm_unorm8(<4 x i8>*, i8*, i32, i32, { [2048 x i32],
 ; X86-SKYLAKE-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; X86-SKYLAKE-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm0
 ; X86-SKYLAKE-NEXT:    vpsrld $7, %xmm0, %xmm0
-; X86-SKYLAKE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4],zero,zero,xmm0[u,u,u,u,u,u,u,u,u,u,u,u]
+; X86-SKYLAKE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4],zero,xmm0[u,u,u,u,u,u,u,u,u,u,u,u,u]
 ; X86-SKYLAKE-NEXT:    vmovd %xmm0, %ecx
 ; X86-SKYLAKE-NEXT:    orl $-16777216, %ecx # imm = 0xFF000000
 ; X86-SKYLAKE-NEXT:    movl %ecx, (%eax)
@@ -54,7 +54,7 @@ define void @fetch_r16g16_snorm_unorm8(<4 x i8>*, i8*, i32, i32, { [2048 x i32],
 ; X64-SKYLAKE-NEXT:    vpxor %xmm1, %xmm1, %xmm1
 ; X64-SKYLAKE-NEXT:    vpmaxsd %xmm1, %xmm0, %xmm0
 ; X64-SKYLAKE-NEXT:    vpsrld $7, %xmm0, %xmm0
-; X64-SKYLAKE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4],zero,zero,xmm0[u,u,u,u,u,u,u,u,u,u,u,u]
+; X64-SKYLAKE-NEXT:    vpshufb {{.*#+}} xmm0 = xmm0[0,4],zero,xmm0[u,u,u,u,u,u,u,u,u,u,u,u,u]
 ; X64-SKYLAKE-NEXT:    vmovd %xmm0, %eax
 ; X64-SKYLAKE-NEXT:    orl $-16777216, %eax # imm = 0xFF000000
 ; X64-SKYLAKE-NEXT:    movl %eax, (%rdi)