From: Simon Pilgrim <llvm-dev@redking.me.uk>
Date: Fri, 11 Nov 2016 10:47:24 +0000 (+0000)
Subject: [SelectionDAG] Add support for vector demandedelts in UDIV opcodes
X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=5809b919f1bc16b9056fb646cfd052651c561dc7;p=llvm

[SelectionDAG] Add support for vector demandedelts in UDIV opcodes

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@286576 91177308-0d34-0410-b5e6-96231b3b80d8
---

diff --git a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
index 93ca120e3cc..48aa1784a9f 100644
--- a/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
+++ b/lib/CodeGen/SelectionDAG/SelectionDAG.cpp
@@ -2192,10 +2192,12 @@ void SelectionDAG::computeKnownBits(SDValue Op, APInt &KnownZero,
     // For the purposes of computing leading zeros we can conservatively
     // treat a udiv as a logical right shift by the power of 2 known to
     // be less than the denominator.
-    computeKnownBits(Op.getOperand(0), KnownZero2, KnownOne2, Depth+1);
+    computeKnownBits(Op.getOperand(0), KnownZero2, KnownOne2, DemandedElts,
+                     Depth + 1);
     unsigned LeadZ = KnownZero2.countLeadingOnes();
 
-    computeKnownBits(Op.getOperand(1), KnownZero2, KnownOne2, Depth+1);
+    computeKnownBits(Op.getOperand(1), KnownZero2, KnownOne2, DemandedElts,
+                     Depth + 1);
     unsigned RHSUnknownLeadingOnes = KnownOne2.countLeadingZeros();
     if (RHSUnknownLeadingOnes != BitWidth)
       LeadZ = std::min(BitWidth,
diff --git a/test/CodeGen/X86/known-bits-vector.ll b/test/CodeGen/X86/known-bits-vector.ll
index 0b3758310fa..a1efb9b176e 100644
--- a/test/CodeGen/X86/known-bits-vector.ll
+++ b/test/CodeGen/X86/known-bits-vector.ll
@@ -242,60 +242,12 @@ define <4 x i32> @knownbits_mask_sub_shuffle_lshr(<4 x i32> %a0) nounwind {
 define <4 x i32> @knownbits_mask_udiv_shuffle_lshr(<4 x i32> %a0, <4 x i32> %a1) nounwind {
 ; X32-LABEL: knownbits_mask_udiv_shuffle_lshr:
 ; X32:       # BB#0:
-; X32-NEXT:    pushl %esi
-; X32-NEXT:    vpand {{\.LCPI.*}}, %xmm0, %xmm0
-; X32-NEXT:    vpextrd $1, %xmm1, %ecx
-; X32-NEXT:    vpextrd $1, %xmm0, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    divl %ecx
-; X32-NEXT:    movl %eax, %ecx
-; X32-NEXT:    vmovd %xmm1, %esi
-; X32-NEXT:    vmovd %xmm0, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    divl %esi
-; X32-NEXT:    vmovd %eax, %xmm2
-; X32-NEXT:    vpinsrd $1, %ecx, %xmm2, %xmm2
-; X32-NEXT:    vpextrd $2, %xmm1, %ecx
-; X32-NEXT:    vpextrd $2, %xmm0, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    divl %ecx
-; X32-NEXT:    vpinsrd $2, %eax, %xmm2, %xmm2
-; X32-NEXT:    vpextrd $3, %xmm1, %ecx
-; X32-NEXT:    vpextrd $3, %xmm0, %eax
-; X32-NEXT:    xorl %edx, %edx
-; X32-NEXT:    divl %ecx
-; X32-NEXT:    vpinsrd $3, %eax, %xmm2, %xmm0
-; X32-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3]
-; X32-NEXT:    vpsrld $22, %xmm0, %xmm0
-; X32-NEXT:    popl %esi
+; X32-NEXT:    vxorps %xmm0, %xmm0, %xmm0
 ; X32-NEXT:    retl
 ;
 ; X64-LABEL: knownbits_mask_udiv_shuffle_lshr:
 ; X64:       # BB#0:
-; X64-NEXT:    vpand {{.*}}(%rip), %xmm0, %xmm0
-; X64-NEXT:    vpextrd $1, %xmm1, %ecx
-; X64-NEXT:    vpextrd $1, %xmm0, %eax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    divl %ecx
-; X64-NEXT:    movl %eax, %ecx
-; X64-NEXT:    vmovd %xmm1, %esi
-; X64-NEXT:    vmovd %xmm0, %eax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    divl %esi
-; X64-NEXT:    vmovd %eax, %xmm2
-; X64-NEXT:    vpinsrd $1, %ecx, %xmm2, %xmm2
-; X64-NEXT:    vpextrd $2, %xmm1, %ecx
-; X64-NEXT:    vpextrd $2, %xmm0, %eax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    divl %ecx
-; X64-NEXT:    vpinsrd $2, %eax, %xmm2, %xmm2
-; X64-NEXT:    vpextrd $3, %xmm1, %ecx
-; X64-NEXT:    vpextrd $3, %xmm0, %eax
-; X64-NEXT:    xorl %edx, %edx
-; X64-NEXT:    divl %ecx
-; X64-NEXT:    vpinsrd $3, %eax, %xmm2, %xmm0
-; X64-NEXT:    vpshufd {{.*#+}} xmm0 = xmm0[0,0,3,3]
-; X64-NEXT:    vpsrld $22, %xmm0, %xmm0
+; X64-NEXT:    vxorps %xmm0, %xmm0, %xmm0
 ; X64-NEXT:    retq
   %1 = and <4 x i32> %a0, <i32 32767, i32 -1, i32 -1, i32 32767>
   %2 = udiv <4 x i32> %1, %a1