From: Simon Pilgrim Date: Mon, 14 Jan 2019 15:43:34 +0000 (+0000) Subject: [DAGCombiner] Add (sub_sat x, x) -> 0 combine X-Git-Url: https://granicus.if.org/sourcecode?a=commitdiff_plain;h=e84a99a487d6fb8b97b99e577870bb742bb26a46;p=llvm [DAGCombiner] Add (sub_sat x, x) -> 0 combine git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@351073 91177308-0d34-0410-b5e6-96231b3b80d8 --- diff --git a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp index d564bda7987..b1770920f06 100644 --- a/lib/CodeGen/SelectionDAG/DAGCombiner.cpp +++ b/lib/CodeGen/SelectionDAG/DAGCombiner.cpp @@ -2800,6 +2800,10 @@ SDValue DAGCombiner::visitSUBSAT(SDNode *N) { if (N0.isUndef() || N1.isUndef()) return DAG.getConstant(0, DL, VT); + // fold (sub_sat x, x) -> 0 + if (N0 == N1) + return DAG.getConstant(0, DL, VT); + if (DAG.isConstantIntBuildVectorOrConstantInt(N0) && DAG.isConstantIntBuildVectorOrConstantInt(N1)) { // fold (sub_sat c1, c2) -> c3 diff --git a/test/CodeGen/X86/combine-sub-ssat.ll b/test/CodeGen/X86/combine-sub-ssat.ll index d68a1deef8e..4cce35552ec 100644 --- a/test/CodeGen/X86/combine-sub-ssat.ll +++ b/test/CodeGen/X86/combine-sub-ssat.ll @@ -96,12 +96,6 @@ define i32 @combine_self_i32(i32 %a0) { ; CHECK-LABEL: combine_self_i32: ; CHECK: # %bb.0: ; CHECK-NEXT: xorl %eax, %eax -; CHECK-NEXT: movl %edi, %ecx -; CHECK-NEXT: subl %edi, %ecx -; CHECK-NEXT: setns %al -; CHECK-NEXT: addl $2147483647, %eax # imm = 0x7FFFFFFF -; CHECK-NEXT: subl %edi, %edi -; CHECK-NEXT: cmovnol %edi, %eax ; CHECK-NEXT: retq %1 = call i32 @llvm.ssub.sat.i32(i32 %a0, i32 %a0) ret i32 %1 @@ -110,12 +104,12 @@ define i32 @combine_self_i32(i32 %a0) { define <8 x i16> @combine_self_v8i16(<8 x i16> %a0) { ; SSE-LABEL: combine_self_v8i16: ; SSE: # %bb.0: -; SSE-NEXT: psubsw %xmm0, %xmm0 +; SSE-NEXT: xorps %xmm0, %xmm0 ; SSE-NEXT: retq ; ; AVX-LABEL: combine_self_v8i16: ; AVX: # %bb.0: -; AVX-NEXT: vpsubsw %xmm0, %xmm0, %xmm0 +; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 ; AVX-NEXT: retq %1 = call <8 x i16> @llvm.ssub.sat.v8i16(<8 x i16> %a0, <8 x i16> %a0) ret <8 x i16> %1 diff --git a/test/CodeGen/X86/combine-sub-usat.ll b/test/CodeGen/X86/combine-sub-usat.ll index ee519194207..e26d0d66799 100644 --- a/test/CodeGen/X86/combine-sub-usat.ll +++ b/test/CodeGen/X86/combine-sub-usat.ll @@ -96,8 +96,6 @@ define i32 @combine_self_i32(i32 %a0) { ; CHECK-LABEL: combine_self_i32: ; CHECK: # %bb.0: ; CHECK-NEXT: xorl %eax, %eax -; CHECK-NEXT: subl %edi, %edi -; CHECK-NEXT: cmovael %edi, %eax ; CHECK-NEXT: retq %1 = call i32 @llvm.usub.sat.i32(i32 %a0, i32 %a0) ret i32 %1 @@ -106,12 +104,12 @@ define i32 @combine_self_i32(i32 %a0) { define <8 x i16> @combine_self_v8i16(<8 x i16> %a0) { ; SSE-LABEL: combine_self_v8i16: ; SSE: # %bb.0: -; SSE-NEXT: psubusw %xmm0, %xmm0 +; SSE-NEXT: xorps %xmm0, %xmm0 ; SSE-NEXT: retq ; ; AVX-LABEL: combine_self_v8i16: ; AVX: # %bb.0: -; AVX-NEXT: vpsubusw %xmm0, %xmm0, %xmm0 +; AVX-NEXT: vxorps %xmm0, %xmm0, %xmm0 ; AVX-NEXT: retq %1 = call <8 x i16> @llvm.usub.sat.v8i16(<8 x i16> %a0, <8 x i16> %a0) ret <8 x i16> %1