[llvm] r344456 - [InstCombine] add tests for operand complexity canonicalization; NFC
Sanjay Patel via llvm-commits
llvm-commits at lists.llvm.org
Sat Oct 13 09:02:47 PDT 2018
Author: spatel
Date: Sat Oct 13 09:02:47 2018
New Revision: 344456
URL: http://llvm.org/viewvc/llvm-project?rev=344456&view=rev
Log:
[InstCombine] add tests for operand complexity canonicalization; NFC
The tests with undef vector elements demonstrate a hole in
the current pattern matching.
Added:
llvm/trunk/test/Transforms/InstCombine/operand-complexity.ll
Added: llvm/trunk/test/Transforms/InstCombine/operand-complexity.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/Transforms/InstCombine/operand-complexity.ll?rev=344456&view=auto
==============================================================================
--- llvm/trunk/test/Transforms/InstCombine/operand-complexity.ll (added)
+++ llvm/trunk/test/Transforms/InstCombine/operand-complexity.ll Sat Oct 13 09:02:47 2018
@@ -0,0 +1,136 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py
+; RUN: opt < %s -instcombine -S | FileCheck %s
+
+; 'Negate' is considered less complex than a normal binop, so the mul should have the binop as the first operand.
+
+define i8 @neg(i8 %x) {
+; CHECK-LABEL: @neg(
+; CHECK-NEXT: [[BO:%.*]] = udiv i8 [[X:%.*]], 42
+; CHECK-NEXT: [[NEGX:%.*]] = sub i8 0, [[X]]
+; CHECK-NEXT: [[R:%.*]] = mul i8 [[BO]], [[NEGX]]
+; CHECK-NEXT: ret i8 [[R]]
+;
+ %bo = udiv i8 %x, 42
+ %negx = sub i8 0, %x
+ %r = mul i8 %negx, %bo
+ ret i8 %r
+}
+
+define <2 x i8> @neg_vec(<2 x i8> %x) {
+; CHECK-LABEL: @neg_vec(
+; CHECK-NEXT: [[BO:%.*]] = udiv <2 x i8> [[X:%.*]], <i8 42, i8 -42>
+; CHECK-NEXT: [[NEGX:%.*]] = sub <2 x i8> zeroinitializer, [[X]]
+; CHECK-NEXT: [[R:%.*]] = mul <2 x i8> [[BO]], [[NEGX]]
+; CHECK-NEXT: ret <2 x i8> [[R]]
+;
+ %bo = udiv <2 x i8> %x, <i8 42, i8 -42>
+ %negx = sub <2 x i8> <i8 0, i8 0>, %x
+ %r = mul <2 x i8> %negx, %bo
+ ret <2 x i8> %r
+}
+
+define <2 x i8> @neg_vec_undef(<2 x i8> %x) {
+; CHECK-LABEL: @neg_vec_undef(
+; CHECK-NEXT: [[BO:%.*]] = udiv <2 x i8> [[X:%.*]], <i8 42, i8 -42>
+; CHECK-NEXT: [[NEGX:%.*]] = sub <2 x i8> <i8 0, i8 undef>, [[X]]
+; CHECK-NEXT: [[R:%.*]] = mul <2 x i8> [[NEGX]], [[BO]]
+; CHECK-NEXT: ret <2 x i8> [[R]]
+;
+ %bo = udiv <2 x i8> %x, <i8 42, i8 -42>
+ %negx = sub <2 x i8> <i8 0, i8 undef>, %x
+ %r = mul <2 x i8> %negx, %bo
+ ret <2 x i8> %r
+}
+
+; 'Not' is considered less complex than a normal binop, so the mul should have the binop as the first operand.
+
+define i8 @not(i8 %x) {
+; CHECK-LABEL: @not(
+; CHECK-NEXT: [[BO:%.*]] = udiv i8 [[X:%.*]], 42
+; CHECK-NEXT: [[NOTX:%.*]] = xor i8 [[X]], -1
+; CHECK-NEXT: [[R:%.*]] = mul i8 [[BO]], [[NOTX]]
+; CHECK-NEXT: ret i8 [[R]]
+;
+ %bo = udiv i8 %x, 42
+ %notx = xor i8 -1, %x
+ %r = mul i8 %notx, %bo
+ ret i8 %r
+}
+
+define <2 x i8> @not_vec(<2 x i8> %x) {
+; CHECK-LABEL: @not_vec(
+; CHECK-NEXT: [[BO:%.*]] = udiv <2 x i8> [[X:%.*]], <i8 42, i8 -42>
+; CHECK-NEXT: [[NOTX:%.*]] = xor <2 x i8> [[X]], <i8 -1, i8 -1>
+; CHECK-NEXT: [[R:%.*]] = mul <2 x i8> [[BO]], [[NOTX]]
+; CHECK-NEXT: ret <2 x i8> [[R]]
+;
+ %bo = udiv <2 x i8> %x, <i8 42, i8 -42>
+ %notx = xor <2 x i8> <i8 -1, i8 -1>, %x
+ %r = mul <2 x i8> %notx, %bo
+ ret <2 x i8> %r
+}
+
+define <2 x i8> @not_vec_undef(<2 x i8> %x) {
+; CHECK-LABEL: @not_vec_undef(
+; CHECK-NEXT: [[BO:%.*]] = udiv <2 x i8> [[X:%.*]], <i8 42, i8 -42>
+; CHECK-NEXT: [[NOTX:%.*]] = xor <2 x i8> [[X]], <i8 -1, i8 undef>
+; CHECK-NEXT: [[R:%.*]] = mul <2 x i8> [[NOTX]], [[BO]]
+; CHECK-NEXT: ret <2 x i8> [[R]]
+;
+ %bo = udiv <2 x i8> %x, <i8 42, i8 -42>
+ %notx = xor <2 x i8> <i8 -1, i8 undef>, %x
+ %r = mul <2 x i8> %notx, %bo
+ ret <2 x i8> %r
+}
+
+; 'Fneg' is considered less complex than a normal binop, so the fmul should have the binop as the first operand.
+; Extra uses are required to ensure that the fneg is not canonicalized after the fmul.
+
+declare void @use(float)
+declare void @use_vec(<2 x float>)
+
+define float @fneg(float %x) {
+; CHECK-LABEL: @fneg(
+; CHECK-NEXT: [[BO:%.*]] = fdiv float [[X:%.*]], 4.200000e+01
+; CHECK-NEXT: [[FNEGX:%.*]] = fsub float -0.000000e+00, [[X]]
+; CHECK-NEXT: [[R:%.*]] = fmul float [[BO]], [[FNEGX]]
+; CHECK-NEXT: call void @use(float [[FNEGX]])
+; CHECK-NEXT: ret float [[R]]
+;
+ %bo = fdiv float %x, 42.0
+ %fnegx = fsub float -0.0, %x
+ %r = fmul float %fnegx, %bo
+ call void @use(float %fnegx)
+ ret float %r
+}
+
+define <2 x float> @fneg_vec(<2 x float> %x) {
+; CHECK-LABEL: @fneg_vec(
+; CHECK-NEXT: [[BO:%.*]] = fdiv <2 x float> [[X:%.*]], <float 4.200000e+01, float -4.200000e+01>
+; CHECK-NEXT: [[FNEGX:%.*]] = fsub <2 x float> <float -0.000000e+00, float -0.000000e+00>, [[X]]
+; CHECK-NEXT: [[R:%.*]] = fmul <2 x float> [[BO]], [[FNEGX]]
+; CHECK-NEXT: call void @use_vec(<2 x float> [[FNEGX]])
+; CHECK-NEXT: ret <2 x float> [[R]]
+;
+ %bo = fdiv <2 x float> %x, <float 42.0, float -42.0>
+ %fnegx = fsub <2 x float> <float -0.0, float -0.0>, %x
+ %r = fmul <2 x float> %fnegx, %bo
+ call void @use_vec(<2 x float> %fnegx)
+ ret <2 x float> %r
+}
+
+define <2 x float> @fneg_vec_undef(<2 x float> %x) {
+; CHECK-LABEL: @fneg_vec_undef(
+; CHECK-NEXT: [[BO:%.*]] = fdiv <2 x float> [[X:%.*]], <float 4.200000e+01, float -4.200000e+01>
+; CHECK-NEXT: [[FNEGX:%.*]] = fsub <2 x float> <float -0.000000e+00, float undef>, [[X]]
+; CHECK-NEXT: [[R:%.*]] = fmul <2 x float> [[FNEGX]], [[BO]]
+; CHECK-NEXT: call void @use_vec(<2 x float> [[FNEGX]])
+; CHECK-NEXT: ret <2 x float> [[R]]
+;
+ %bo = fdiv <2 x float> %x, <float 42.0, float -42.0>
+ %fnegx = fsub <2 x float> <float -0.0, float undef>, %x
+ %r = fmul <2 x float> %fnegx, %bo
+ call void @use_vec(<2 x float> %fnegx)
+ ret <2 x float> %r
+}
+
More information about the llvm-commits
mailing list