[SelectionDAG] ComputeNumSignBits of TRUNCATE operations

Add basic ComputeNumSignBits support for TRUNCATE ops for cases where the source's number of sign bits overlaps with the truncated size.

Improves X86 SIGN_EXTEND_IN_REG vector cases which were needlessly sign extending boolean vector results.

Differential Revision: https://reviews.llvm.org/D26851

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@287635 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index c3230726b1ea9adee0c70a5e9ee4e7c3d2f6cdef..f2bb3795a2d848799602f9d845c9902de29563e5 100644 (file)
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2843,10 +2843,14 @@ unsigned SelectionDAG::ComputeNumSignBits(SDValue Op, unsigned Depth) const {
     Tmp = ComputeNumSignBits(Op.getOperand(0), Depth+1);
     if (Tmp == 1) return 1;  // Early out.
     return std::min(Tmp, Tmp2)-1;
-  case ISD::TRUNCATE:
-    // FIXME: it's tricky to do anything useful for this, but it is an important
-    // case for targets like X86.
+  case ISD::TRUNCATE: {
+    // Check if the sign bits of source go down as far as the truncated value.
+    unsigned NumSrcBits = Op.getOperand(0).getScalarValueSizeInBits();
+    unsigned NumSrcSignBits = ComputeNumSignBits(Op.getOperand(0), Depth + 1);
+    if (NumSrcSignBits > (NumSrcBits - VTBits))
+      return NumSrcSignBits - (NumSrcBits - VTBits);
     break;
+  }
   case ISD::EXTRACT_ELEMENT: {
     const int KnownSign = ComputeNumSignBits(Op.getOperand(0), Depth+1);
     const int BitWidth = Op.getValueSizeInBits();

diff --git a/test/CodeGen/X86/avx512-mask-op.ll b/test/CodeGen/X86/avx512-mask-op.ll
index 6378353bd1c9bf6d13bd79367606ab4599459d43..4696cbe4118f7f5022d6ed98ae6aa3255b4cffaa 100644 (file)
--- a/test/CodeGen/X86/avx512-mask-op.ll
+++ b/test/CodeGen/X86/avx512-mask-op.ll
@@ -246,12 +246,8 @@ define <4 x i32> @test4(<4 x i64> %x, <4 x i64> %y, <4 x i64> %x1, <4 x i64> %y1
 ; KNL:       ## BB#0:
 ; KNL-NEXT:    vpcmpgtq %ymm1, %ymm0, %ymm0
 ; KNL-NEXT:    vpmovqd %zmm0, %ymm0
-; KNL-NEXT:    vpslld $31, %xmm0, %xmm0
-; KNL-NEXT:    vpsrad $31, %xmm0, %xmm0
 ; KNL-NEXT:    vpcmpgtq %ymm3, %ymm2, %ymm1
 ; KNL-NEXT:    vpmovqd %zmm1, %ymm1
-; KNL-NEXT:    vpslld $31, %xmm1, %xmm1
-; KNL-NEXT:    vpsrad $31, %xmm1, %xmm1
 ; KNL-NEXT:    vpcmpgtd %xmm1, %xmm0, %xmm0
 ; KNL-NEXT:    retq
 ;

diff --git a/test/CodeGen/X86/psubus.ll b/test/CodeGen/X86/psubus.ll
index 731cd2bc8c0ada81aa1da81383b55e6b9f761833..9a11eff5331df228f7444d3814677fd7033c0321 100644 (file)
--- a/test/CodeGen/X86/psubus.ll
+++ b/test/CodeGen/X86/psubus.ll
@@ -601,16 +601,13 @@ define void @test13(i16* nocapture %head, i32* nocapture %w) nounwind {
 ; AVX1-NEXT:    vxorps %xmm3, %xmm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtd %xmm6, %xmm3, %xmm3
 ; AVX1-NEXT:    vpacksswb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw $15, %xmm3, %xmm3
-; AVX1-NEXT:    vpsraw $15, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpsubd %xmm5, %xmm2, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm1, %xmm1
 ; AVX1-NEXT:    vpshufb %xmm2, %xmm0, %xmm0
 ; AVX1-NEXT:    vpunpcklqdq {{.*#+}} xmm0 = xmm0[0],xmm1[0]
-; AVX1-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX1-NEXT:    vpblendvb %xmm3, %xmm1, %xmm0, %xmm0
+; AVX1-NEXT:    vpandn %xmm0, %xmm3, %xmm0
 ; AVX1-NEXT:    vmovdqu %xmm0, (%rdi)
 ; AVX1-NEXT:    vzeroupper
 ; AVX1-NEXT:    retq
@@ -625,13 +622,10 @@ define void @test13(i16* nocapture %head, i32* nocapture %w) nounwind {
 ; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpacksswb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsllw $15, %xmm2, %xmm2
-; AVX2-NEXT:    vpsraw $15, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
-; AVX2-NEXT:    vpxor %xmm1, %xmm1, %xmm1
-; AVX2-NEXT:    vpblendvb %xmm2, %xmm1, %xmm0, %xmm0
+; AVX2-NEXT:    vpandn %xmm0, %xmm2, %xmm0
 ; AVX2-NEXT:    vmovdqu %xmm0, (%rdi)
 ; AVX2-NEXT:    vzeroupper
 ; AVX2-NEXT:    retq
@@ -984,8 +978,6 @@ define void @test15(i16* nocapture %head, i32* nocapture %w) nounwind {
 ; AVX1-NEXT:    vxorps %xmm3, %xmm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm6, %xmm3
 ; AVX1-NEXT:    vpacksswb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw $15, %xmm3, %xmm3
-; AVX1-NEXT:    vpsraw $15, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpsubd %xmm5, %xmm2, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
@@ -1007,8 +999,6 @@ define void @test15(i16* nocapture %head, i32* nocapture %w) nounwind {
 ; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpacksswb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsllw $15, %xmm2, %xmm2
-; AVX2-NEXT:    vpsraw $15, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]
@@ -1118,8 +1108,6 @@ define void @test16(i16* nocapture %head, i32* nocapture %w) nounwind {
 ; AVX1-NEXT:    vxorps %xmm3, %xmm0, %xmm3
 ; AVX1-NEXT:    vpcmpgtd %xmm3, %xmm6, %xmm3
 ; AVX1-NEXT:    vpacksswb %xmm4, %xmm3, %xmm3
-; AVX1-NEXT:    vpsllw $15, %xmm3, %xmm3
-; AVX1-NEXT:    vpsraw $15, %xmm3, %xmm3
 ; AVX1-NEXT:    vpsubd %xmm0, %xmm1, %xmm0
 ; AVX1-NEXT:    vpsubd %xmm5, %xmm2, %xmm1
 ; AVX1-NEXT:    vmovdqa {{.*#+}} xmm2 = [0,1,4,5,8,9,12,13,8,9,12,13,12,13,14,15]
@@ -1141,8 +1129,6 @@ define void @test16(i16* nocapture %head, i32* nocapture %w) nounwind {
 ; AVX2-NEXT:    vpcmpgtd %ymm3, %ymm2, %ymm2
 ; AVX2-NEXT:    vextracti128 $1, %ymm2, %xmm3
 ; AVX2-NEXT:    vpacksswb %xmm3, %xmm2, %xmm2
-; AVX2-NEXT:    vpsllw $15, %xmm2, %xmm2
-; AVX2-NEXT:    vpsraw $15, %xmm2, %xmm2
 ; AVX2-NEXT:    vpsubd %ymm0, %ymm1, %ymm0
 ; AVX2-NEXT:    vpshufb {{.*#+}} ymm0 = ymm0[0,1,4,5,8,9,12,13],zero,zero,zero,zero,zero,zero,zero,zero,ymm0[16,17,20,21,24,25,28,29],zero,zero,zero,zero,zero,zero,zero,zero
 ; AVX2-NEXT:    vpermq {{.*#+}} ymm0 = ymm0[0,2,2,3]