Index: llvm/test/Transforms/InstCombine/fast-basictest-instcombine.ll =================================================================== --- llvm/test/Transforms/InstCombine/fast-basictest-instcombine.ll +++ llvm/test/Transforms/InstCombine/fast-basictest-instcombine.ll @@ -1,11 +1,19 @@ ; NOTE: Assertions have been autogenerated by utils/update_test_checks.py -; RUN: opt < %s -reassociate -gvn -instcombine -S | FileCheck %s +; +; Test numbering remains continuous across: +; - InstCombine/fast-basictest-instcombine.ll +; - Reassociate/fast-basictest-reassociate-gvn.ll +; - Reassociate/fast-basictest-reassociate-instcombine.ll +; - Reassociate/fast-basictest-reassociate.ll +; +; RUN: opt < %s -instcombine -S | FileCheck %s ; With reassociation, constant folding can eliminate the 12 and -12 constants. + define float @test1(float %arg) { ; CHECK-LABEL: @test1( -; CHECK-NEXT: [[ARG_NEG:%.*]] = fneg fast float [[ARG:%.*]] -; CHECK-NEXT: ret float [[ARG_NEG]] +; CHECK-NEXT: [[TMP1:%.*]] = fneg fast float [[ARG:%.*]] +; CHECK-NEXT: ret float [[TMP1]] ; %t1 = fsub fast float -1.200000e+01, %arg %t2 = fadd fast float %t1, 1.200000e+01 @@ -36,22 +44,15 @@ ret float %t2 } -define float @test2(float %reg109, float %reg1111) { -; CHECK-LABEL: @test2( -; CHECK-NEXT: [[REG115:%.*]] = fadd float [[REG109:%.*]], -3.000000e+01 -; CHECK-NEXT: [[REG116:%.*]] = fadd float [[REG115]], [[REG1111:%.*]] -; CHECK-NEXT: [[REG117:%.*]] = fadd float [[REG116]], 3.000000e+01 -; CHECK-NEXT: ret float [[REG117]] -; - %reg115 = fadd float %reg109, -3.000000e+01 - %reg116 = fadd float %reg115, %reg1111 - %reg117 = fadd float %reg116, 3.000000e+01 - ret float %reg117 -} +; ((a + (-3)) + b) + 3 -> a + b +; That only works with both instcombine and reassociate passes enabled. +; Check that instcombine is not enough. define float @test3(float %reg109, float %reg1111) { ; CHECK-LABEL: @test3( -; CHECK-NEXT: [[REG117:%.*]] = fadd fast float [[REG109:%.*]], [[REG1111:%.*]] +; CHECK-NEXT: [[REG115:%.*]] = fadd fast float [[REG109:%.*]], -3.000000e+01 +; CHECK-NEXT: [[REG116:%.*]] = fadd fast float [[REG115]], [[REG1111:%.*]] +; CHECK-NEXT: [[REG117:%.*]] = fadd fast float [[REG116]], 3.000000e+01 ; CHECK-NEXT: ret float [[REG117]] ; %reg115 = fadd fast float %reg109, -3.000000e+01 @@ -60,6 +61,19 @@ ret float %reg117 } +define float @test3_no_FMF(float %reg109, float %reg1111) { +; CHECK-LABEL: @test3_no_FMF( +; CHECK-NEXT: [[REG115:%.*]] = fadd float [[REG109:%.*]], -3.000000e+01 +; CHECK-NEXT: [[REG116:%.*]] = fadd float [[REG115]], [[REG1111:%.*]] +; CHECK-NEXT: [[REG117:%.*]] = fadd float [[REG116]], 3.000000e+01 +; CHECK-NEXT: ret float [[REG117]] +; + %reg115 = fadd float %reg109, -3.000000e+01 + %reg116 = fadd float %reg115, %reg1111 + %reg117 = fadd float %reg116, 3.000000e+01 + ret float %reg117 +} + define float @test3_reassoc(float %reg109, float %reg1111) { ; CHECK-LABEL: @test3_reassoc( ; CHECK-NEXT: [[REG115:%.*]] = fadd reassoc float [[REG109:%.*]], -3.000000e+01 @@ -73,125 +87,12 @@ ret float %reg117 } -@fe = external global float -@fa = external global float -@fb = external global float -@fc = external global float -@ff = external global float - -define void @test4() { -; CHECK-LABEL: @test4( -; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 -; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 -; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 -; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] -; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] -; CHECK-NEXT: store float [[T2]], float* @fe, align 4 -; CHECK-NEXT: store float [[T2]], float* @ff, align 4 -; CHECK-NEXT: ret void -; - %A = load float, float* @fa - %B = load float, float* @fb - %C = load float, float* @fc - %t1 = fadd fast float %A, %B - %t2 = fadd fast float %t1, %C - %t3 = fadd fast float %C, %A - %t4 = fadd fast float %t3, %B - ; e = (a+b)+c; - store float %t2, float* @fe - ; f = (a+c)+b - store float %t4, float* @ff - ret void -} - -define void @test5() { -; CHECK-LABEL: @test5( -; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 -; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 -; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 -; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] -; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] -; CHECK-NEXT: store float [[T2]], float* @fe, align 4 -; CHECK-NEXT: store float [[T2]], float* @ff, align 4 -; CHECK-NEXT: ret void -; - %A = load float, float* @fa - %B = load float, float* @fb - %C = load float, float* @fc - %t1 = fadd fast float %A, %B - %t2 = fadd fast float %t1, %C - %t3 = fadd fast float %C, %A - %t4 = fadd fast float %t3, %B - ; e = c+(a+b) - store float %t2, float* @fe - ; f = (c+a)+b - store float %t4, float* @ff - ret void -} - -define void @test6() { -; CHECK-LABEL: @test6( -; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 -; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 -; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 -; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] -; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] -; CHECK-NEXT: store float [[T2]], float* @fe, align 4 -; CHECK-NEXT: store float [[T2]], float* @ff, align 4 -; CHECK-NEXT: ret void -; - %A = load float, float* @fa - %B = load float, float* @fb - %C = load float, float* @fc - %t1 = fadd fast float %B, %A - %t2 = fadd fast float %t1, %C - %t3 = fadd fast float %C, %A - %t4 = fadd fast float %t3, %B - ; e = c+(b+a) - store float %t2, float* @fe - ; f = (c+a)+b - store float %t4, float* @ff - ret void -} - -define float @test7(float %A, float %B, float %C) { -; CHECK-LABEL: @test7( -; CHECK-NEXT: [[REASS_ADD1:%.*]] = fadd fast float [[C:%.*]], [[B:%.*]] -; CHECK-NEXT: [[REASS_MUL2:%.*]] = fmul fast float [[A:%.*]], [[A]] -; CHECK-NEXT: [[REASS_MUL:%.*]] = fmul fast float [[REASS_MUL2]], [[REASS_ADD1]] -; CHECK-NEXT: ret float [[REASS_MUL]] -; - %aa = fmul fast float %A, %A - %aab = fmul fast float %aa, %B - %ac = fmul fast float %A, %C - %aac = fmul fast float %ac, %A - %r = fadd fast float %aab, %aac - ret float %r -} - -define float @test7_reassoc(float %A, float %B, float %C) { -; CHECK-LABEL: @test7_reassoc( -; CHECK-NEXT: [[AA:%.*]] = fmul reassoc float [[A:%.*]], [[A]] -; CHECK-NEXT: [[AAB:%.*]] = fmul reassoc float [[AA]], [[B:%.*]] -; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc float [[A]], [[A]] -; CHECK-NEXT: [[AAC:%.*]] = fmul reassoc float [[TMP1]], [[C:%.*]] -; CHECK-NEXT: [[R:%.*]] = fadd reassoc float [[AAB]], [[AAC]] -; CHECK-NEXT: ret float [[R]] -; - %aa = fmul reassoc float %A, %A - %aab = fmul reassoc float %aa, %B - %ac = fmul reassoc float %A, %C - %aac = fmul reassoc float %ac, %A - %r = fadd reassoc float %aab, %aac - ret float %r -} - ; (-X)*Y + Z -> Z-X*Y define float @test8(float %X, float %Y, float %Z) { ; CHECK-LABEL: @test8( -; CHECK-NEXT: [[A:%.*]] = fmul fast float [[Y:%.*]], [[X:%.*]] -; CHECK-NEXT: [[C:%.*]] = fsub fast float [[Z:%.*]], [[A]] +; CHECK-NEXT: [[TMP1:%.*]] = fmul fast float [[X:%.*]], [[Y:%.*]] +; CHECK-NEXT: [[C:%.*]] = fsub fast float [[Z:%.*]], [[TMP1]] ; CHECK-NEXT: ret float [[C]] ; %A = fsub fast float 0.0, %X @@ -212,6 +113,19 @@ ret float %C } +define float @test8_reassoc_nsz(float %X, float %Y, float %Z) { +; CHECK-LABEL: @test8_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz float [[X:%.*]], [[Y:%.*]] +; CHECK-NEXT: [[C:%.*]] = fsub reassoc nsz float [[Z:%.*]], [[TMP1]] +; CHECK-NEXT: ret float [[C]] +; + %A = fsub reassoc nsz float 0.0, %X + %B = fmul reassoc nsz float %A, %Y + %C = fadd reassoc nsz float %B, %Z + ret float %C +} + +; Verify that fold is not done only with 'reassoc' ('nsz' is required) define float @test8_reassoc(float %X, float %Y, float %Z) { ; CHECK-LABEL: @test8_reassoc( ; CHECK-NEXT: [[A:%.*]] = fsub reassoc float 0.000000e+00, [[X:%.*]] @@ -227,8 +141,8 @@ define float @test9(float %X) { ; CHECK-LABEL: @test9( -; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[X:%.*]], 9.400000e+01 -; CHECK-NEXT: ret float [[FACTOR]] +; CHECK-NEXT: [[TMP1:%.*]] = fmul fast float [[X:%.*]], 9.400000e+01 +; CHECK-NEXT: ret float [[TMP1]] ; %Y = fmul fast float %X, 4.700000e+01 %Z = fadd fast float %Y, %Y @@ -261,10 +175,11 @@ ; Side note: (x + x + x) and (3*x) each have only a single rounding. So ; transforming x+x+x to 3*x is always safe, even without any FMF. ; To avoid that special-case, we have the addition of 'x' four times, here. + define float @test10(float %X) { ; CHECK-LABEL: @test10( -; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[X:%.*]], 4.000000e+00 -; CHECK-NEXT: ret float [[FACTOR]] +; CHECK-NEXT: [[TMP1:%.*]] = fmul fast float [[X:%.*]], 4.000000e+00 +; CHECK-NEXT: ret float [[TMP1]] ; %Y = fadd fast float %X ,%X %Z = fadd fast float %Y, %X @@ -300,8 +215,8 @@ define float @test11(float %W) { ; CHECK-LABEL: @test11( -; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[W:%.*]], 3.810000e+02 -; CHECK-NEXT: ret float [[FACTOR]] +; CHECK-NEXT: [[Z:%.*]] = fmul fast float [[W:%.*]], 3.810000e+02 +; CHECK-NEXT: ret float [[Z]] ; %X = fmul fast float %W, 127.0 %Y = fadd fast float %X ,%X @@ -327,7 +242,7 @@ ; CHECK-LABEL: @test11_reassoc( ; CHECK-NEXT: [[X:%.*]] = fmul reassoc float [[W:%.*]], 1.270000e+02 ; CHECK-NEXT: [[Y:%.*]] = fadd reassoc float [[X]], [[X]] -; CHECK-NEXT: [[Z:%.*]] = fadd reassoc float [[X]], [[Y]] +; CHECK-NEXT: [[Z:%.*]] = fadd reassoc float [[Y]], [[X]] ; CHECK-NEXT: ret float [[Z]] ; %X = fmul reassoc float %W, 127.0 @@ -338,9 +253,9 @@ define float @test12(float %X) { ; CHECK-LABEL: @test12( -; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[X:%.*]], -3.000000e+00 -; CHECK-NEXT: [[Z:%.*]] = fadd fast float [[FACTOR]], 6.000000e+00 -; CHECK-NEXT: ret float [[Z]] +; CHECK-NEXT: [[TMP1:%.*]] = fmul fast float [[X:%.*]], 3.000000e+00 +; CHECK-NEXT: [[TMP2:%.*]] = fsub fast float 6.000000e+00, [[TMP1]] +; CHECK-NEXT: ret float [[TMP2]] ; %A = fsub fast float 1.000000e+00, %X %B = fsub fast float 2.000000e+00, %X @@ -353,8 +268,8 @@ ; Check again with 'reassoc' and 'nsz' ('nsz' not technically required). define float @test12_reassoc_nsz(float %X) { ; CHECK-LABEL: @test12_reassoc_nsz( -; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz float [[X:%.*]], -3.000000e+00 -; CHECK-NEXT: [[TMP2:%.*]] = fadd reassoc nsz float [[TMP1]], 6.000000e+00 +; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz float [[X:%.*]], 3.000000e+00 +; CHECK-NEXT: [[TMP2:%.*]] = fsub reassoc nsz float 6.000000e+00, [[TMP1]] ; CHECK-NEXT: ret float [[TMP2]] ; %A = fsub reassoc nsz float 1.000000e+00, %X @@ -372,7 +287,7 @@ ; CHECK-NEXT: [[B:%.*]] = fsub reassoc float 2.000000e+00, [[X]] ; CHECK-NEXT: [[C:%.*]] = fsub reassoc float 3.000000e+00, [[X]] ; CHECK-NEXT: [[Y:%.*]] = fadd reassoc float [[A]], [[B]] -; CHECK-NEXT: [[Z:%.*]] = fadd reassoc float [[C]], [[Y]] +; CHECK-NEXT: [[Z:%.*]] = fadd reassoc float [[Y]], [[C]] ; CHECK-NEXT: ret float [[Z]] ; %A = fsub reassoc float 1.000000e+00, %X @@ -385,9 +300,9 @@ define float @test13(float %X1, float %X2, float %X3) { ; CHECK-LABEL: @test13( -; CHECK-NEXT: [[REASS_ADD:%.*]] = fsub fast float [[X3:%.*]], [[X2:%.*]] -; CHECK-NEXT: [[REASS_MUL:%.*]] = fmul fast float [[REASS_ADD]], [[X1:%.*]] -; CHECK-NEXT: ret float [[REASS_MUL]] +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float [[X3:%.*]], [[X2:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul fast float [[TMP1]], [[X1:%.*]] +; CHECK-NEXT: ret float [[D]] ; %A = fsub fast float 0.000000e+00, %X1 %B = fmul fast float %A, %X2 ; -X1*X2 @@ -409,6 +324,21 @@ ret float %D } +define float @test13_reassoc_nsz(float %X1, float %X2, float %X3) { +; CHECK-LABEL: @test13_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fsub reassoc nsz float [[X3:%.*]], [[X2:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul reassoc nsz float [[TMP1]], [[X1:%.*]] +; CHECK-NEXT: ret float [[D]] +; + %A = fsub reassoc nsz float 0.000000e+00, %X1 + %B = fmul reassoc nsz float %A, %X2 ; -X1*X2 + %C = fmul reassoc nsz float %X1, %X3 ; X1*X3 + %D = fadd reassoc nsz float %B, %C ; -X1*X2 + X1*X3 -> X1*(X3-X2) + ret float %D +} + +; TODO: check if 'nsz' is technically required. Currently the optimization +; is not done with only 'reassoc' without 'nsz'. define float @test13_reassoc(float %X1, float %X2, float %X3) { ; CHECK-LABEL: @test13_reassoc( ; CHECK-NEXT: [[A:%.*]] = fsub reassoc float 0.000000e+00, [[X1:%.*]] @@ -424,43 +354,46 @@ ret float %D } +; (x1 * 47) + (x2 * -47) => (x1 - x2) * 47 +; That only works with both instcombine and reassociate passes enabled. +; Check that instcombine is not enough. + define float @test14(float %X1, float %X2) { ; CHECK-LABEL: @test14( -; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float [[X1:%.*]], [[X2:%.*]] -; CHECK-NEXT: [[D1:%.*]] = fmul fast float [[TMP1]], 4.700000e+01 -; CHECK-NEXT: ret float [[D1]] +; CHECK-NEXT: [[B:%.*]] = fmul fast float [[X1:%.*]], 4.700000e+01 +; CHECK-NEXT: [[C:%.*]] = fmul fast float [[X2:%.*]], -4.700000e+01 +; CHECK-NEXT: [[D:%.*]] = fadd fast float [[B]], [[C]] +; CHECK-NEXT: ret float [[D]] ; %B = fmul fast float %X1, 47. ; X1*47 %C = fmul fast float %X2, -47. ; X2*-47 - %D = fadd fast float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + %D = fadd fast float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) ret float %D } -; (x1 * 47) + (x2 * -47) => (x1 - x2) * 47 -; Check again with 'reassoc' and 'nsz' ('nsz' not technically required). define float @test14_reassoc_nsz(float %X1, float %X2) { ; CHECK-LABEL: @test14_reassoc_nsz( -; CHECK-NEXT: [[TMP1:%.*]] = fsub reassoc nsz float [[X1:%.*]], [[X2:%.*]] -; CHECK-NEXT: [[D1:%.*]] = fmul reassoc nsz float [[TMP1]], 4.700000e+01 -; CHECK-NEXT: ret float [[D1]] +; CHECK-NEXT: [[B:%.*]] = fmul reassoc nsz float [[X1:%.*]], 4.700000e+01 +; CHECK-NEXT: [[C:%.*]] = fmul reassoc nsz float [[X2:%.*]], -4.700000e+01 +; CHECK-NEXT: [[D:%.*]] = fadd reassoc nsz float [[B]], [[C]] +; CHECK-NEXT: ret float [[D]] ; %B = fmul reassoc nsz float %X1, 47. ; X1*47 %C = fmul reassoc nsz float %X2, -47. ; X2*-47 - %D = fadd reassoc nsz float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + %D = fadd reassoc nsz float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) ret float %D } -; TODO: This doesn't require 'nsz'. It should fold to ((x1 - x2) * 47.0) define float @test14_reassoc(float %X1, float %X2) { ; CHECK-LABEL: @test14_reassoc( ; CHECK-NEXT: [[B:%.*]] = fmul reassoc float [[X1:%.*]], 4.700000e+01 -; CHECK-NEXT: [[C:%.*]] = fmul reassoc float [[X2:%.*]], 4.700000e+01 -; CHECK-NEXT: [[D1:%.*]] = fsub reassoc float [[B]], [[C]] -; CHECK-NEXT: ret float [[D1]] +; CHECK-NEXT: [[C:%.*]] = fmul reassoc float [[X2:%.*]], -4.700000e+01 +; CHECK-NEXT: [[D:%.*]] = fadd reassoc float [[B]], [[C]] +; CHECK-NEXT: ret float [[D]] ; %B = fmul reassoc float %X1, 47. ; X1*47 %C = fmul reassoc float %X2, -47. ; X2*-47 - %D = fadd reassoc float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + %D = fadd reassoc float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) ret float %D } @@ -485,10 +418,15 @@ } ; (b+(a+1234))+-a -> b+1234 +; That only works with both instcombine and reassociate passes enabled. +; Check that instcombine is not enough. + define float @test16(float %b, float %a) { ; CHECK-LABEL: @test16( -; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[B:%.*]], 1.234000e+03 -; CHECK-NEXT: ret float [[TMP1]] +; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[A:%.*]], 1.234000e+03 +; CHECK-NEXT: [[TMP2:%.*]] = fadd fast float [[TMP1]], [[B:%.*]] +; CHECK-NEXT: [[TMP3:%.*]] = fsub fast float [[TMP2]], [[A]] +; CHECK-NEXT: ret float [[TMP3]] ; %1 = fadd fast float %a, 1234.0 %2 = fadd fast float %b, %1 @@ -499,8 +437,10 @@ define float @test16_unary_fneg(float %b, float %a) { ; CHECK-LABEL: @test16_unary_fneg( -; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[B:%.*]], 1.234000e+03 -; CHECK-NEXT: ret float [[TMP1]] +; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[A:%.*]], 1.234000e+03 +; CHECK-NEXT: [[TMP2:%.*]] = fadd fast float [[TMP1]], [[B:%.*]] +; CHECK-NEXT: [[TMP3:%.*]] = fsub fast float [[TMP2]], [[A]] +; CHECK-NEXT: ret float [[TMP3]] ; %1 = fadd fast float %a, 1234.0 %2 = fadd fast float %b, %1 @@ -509,12 +449,26 @@ ret float %4 } +define float @test16_reassoc_nsz(float %b, float %a) { +; CHECK-LABEL: @test16_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc nsz float [[A:%.*]], 1.234000e+03 +; CHECK-NEXT: [[TMP2:%.*]] = fadd reassoc nsz float [[TMP1]], [[B:%.*]] +; CHECK-NEXT: [[TMP3:%.*]] = fsub reassoc nsz float [[TMP2]], [[A]] +; CHECK-NEXT: ret float [[TMP3]] +; + %1 = fadd reassoc nsz float %a, 1234.0 + %2 = fadd reassoc nsz float %b, %1 + %3 = fsub reassoc nsz float 0.0, %a + %4 = fadd reassoc nsz float %2, %3 + ret float %4 +} + define float @test16_reassoc(float %b, float %a) { ; CHECK-LABEL: @test16_reassoc( ; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc float [[A:%.*]], 1.234000e+03 ; CHECK-NEXT: [[TMP2:%.*]] = fadd reassoc float [[TMP1]], [[B:%.*]] ; CHECK-NEXT: [[TMP3:%.*]] = fsub reassoc float 0.000000e+00, [[A]] -; CHECK-NEXT: [[TMP4:%.*]] = fadd reassoc float [[TMP3]], [[TMP2]] +; CHECK-NEXT: [[TMP4:%.*]] = fadd reassoc float [[TMP2]], [[TMP3]] ; CHECK-NEXT: ret float [[TMP4]] ; %1 = fadd reassoc float %a, 1234.0 @@ -524,13 +478,16 @@ ret float %4 } -; Test that we can turn things like X*-(Y*Z) -> X*-1*Y*Z. +; X*-(Y*Z) -> X*-1*Y*Z +; That only works with both instcombine and reassociate passes enabled. +; Check that instcombine is not enough. define float @test17(float %a, float %b, float %z) { ; CHECK-LABEL: @test17( -; CHECK-NEXT: [[E:%.*]] = fmul fast float [[A:%.*]], 1.234500e+04 -; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[B:%.*]] -; CHECK-NEXT: [[G:%.*]] = fmul fast float [[F]], [[Z:%.*]] +; CHECK-NEXT: [[C:%.*]] = fneg fast float [[Z:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul fast float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[D]], [[C]] +; CHECK-NEXT: [[G:%.*]] = fmul fast float [[E]], -1.234500e+04 ; CHECK-NEXT: ret float [[G]] ; %c = fsub fast float 0.000000e+00, %z @@ -543,10 +500,11 @@ define float @test17_unary_fneg(float %a, float %b, float %z) { ; CHECK-LABEL: @test17_unary_fneg( -; CHECK-NEXT: [[D:%.*]] = fmul fast float [[A:%.*]], 1.234500e+04 -; CHECK-NEXT: [[E:%.*]] = fmul fast float [[D]], [[B:%.*]] -; CHECK-NEXT: [[TMP1:%.*]] = fmul fast float [[E]], [[Z:%.*]] -; CHECK-NEXT: ret float [[TMP1]] +; CHECK-NEXT: [[C:%.*]] = fneg fast float [[Z:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul fast float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[D]], [[C]] +; CHECK-NEXT: [[G:%.*]] = fmul fast float [[E]], -1.234500e+04 +; CHECK-NEXT: ret float [[G]] ; %c = fneg fast float %z %d = fmul fast float %a, %b @@ -556,11 +514,27 @@ ret float %g } +define float @test17_reassoc_nsz(float %a, float %b, float %z) { +; CHECK-LABEL: @test17_reassoc_nsz( +; CHECK-NEXT: [[C:%.*]] = fneg reassoc nsz float [[Z:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul reassoc nsz float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[E:%.*]] = fmul reassoc nsz float [[D]], [[C]] +; CHECK-NEXT: [[G:%.*]] = fmul reassoc nsz float [[E]], -1.234500e+04 +; CHECK-NEXT: ret float [[G]] +; + %c = fsub reassoc nsz float 0.000000e+00, %z + %d = fmul reassoc nsz float %a, %b + %e = fmul reassoc nsz float %c, %d + %f = fmul reassoc nsz float %e, 1.234500e+04 + %g = fsub reassoc nsz float 0.000000e+00, %f + ret float %g +} + define float @test17_reassoc(float %a, float %b, float %z) { ; CHECK-LABEL: @test17_reassoc( ; CHECK-NEXT: [[C:%.*]] = fsub reassoc float 0.000000e+00, [[Z:%.*]] ; CHECK-NEXT: [[D:%.*]] = fmul reassoc float [[A:%.*]], [[B:%.*]] -; CHECK-NEXT: [[E:%.*]] = fmul reassoc float [[D]], [[C]] +; CHECK-NEXT: [[E:%.*]] = fmul reassoc float [[C]], [[D]] ; CHECK-NEXT: [[F:%.*]] = fmul reassoc float [[E]], 1.234500e+04 ; CHECK-NEXT: [[G:%.*]] = fsub reassoc float 0.000000e+00, [[F]] ; CHECK-NEXT: ret float [[G]] @@ -575,8 +549,8 @@ define float @test18(float %a, float %b, float %z) { ; CHECK-LABEL: @test18( -; CHECK-NEXT: [[E:%.*]] = fmul fast float [[A:%.*]], 4.000000e+01 -; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[Z:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = fmul fast float [[Z:%.*]], 4.000000e+01 +; CHECK-NEXT: [[F:%.*]] = fmul fast float [[TMP1]], [[A:%.*]] ; CHECK-NEXT: ret float [[F]] ; %d = fmul fast float %z, 4.000000e+01 @@ -588,8 +562,8 @@ define float @test18_unary_fneg(float %a, float %b, float %z) { ; CHECK-LABEL: @test18_unary_fneg( -; CHECK-NEXT: [[E:%.*]] = fmul fast float [[A:%.*]], 4.000000e+01 -; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[Z:%.*]] +; CHECK-NEXT: [[TMP1:%.*]] = fmul fast float [[Z:%.*]], 4.000000e+01 +; CHECK-NEXT: [[F:%.*]] = fmul fast float [[TMP1]], [[A:%.*]] ; CHECK-NEXT: ret float [[F]] ; %d = fmul fast float %z, 4.000000e+01 @@ -599,6 +573,20 @@ ret float %f } +define float @test18_reassoc_nsz(float %a, float %b, float %z) { +; CHECK-LABEL: @test18_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fmul reassoc nsz float [[Z:%.*]], 4.000000e+01 +; CHECK-NEXT: [[F:%.*]] = fmul reassoc nsz float [[TMP1]], [[A:%.*]] +; CHECK-NEXT: ret float [[F]] +; + %d = fmul reassoc nsz float %z, 4.000000e+01 + %c = fsub reassoc nsz float 0.000000e+00, %d + %e = fmul reassoc nsz float %a, %c + %f = fsub reassoc nsz float 0.000000e+00, %e + ret float %f +} + +; Verify the fold is not done with only 'reassoc' ('nsz' is required). define float @test18_reassoc(float %a, float %b, float %z) { ; CHECK-LABEL: @test18_reassoc( ; CHECK-NEXT: [[D:%.*]] = fmul reassoc float [[Z:%.*]], 4.000000e+01 @@ -643,9 +631,14 @@ } ; With sub reassociation, constant folding can eliminate the 12 and -12 constants. +; That only works with both instcombine and reassociate passes enabled. +; Check that instcombine is not enough. + define float @test19(float %A, float %B) { ; CHECK-LABEL: @test19( -; CHECK-NEXT: [[Z:%.*]] = fsub fast float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[X:%.*]] = fadd fast float [[A:%.*]], -1.200000e+01 +; CHECK-NEXT: [[Y:%.*]] = fsub fast float [[X]], [[B:%.*]] +; CHECK-NEXT: [[Z:%.*]] = fadd fast float [[Y]], 1.200000e+01 ; CHECK-NEXT: ret float [[Z]] ; %X = fadd fast float -1.200000e+01, %A @@ -668,6 +661,7 @@ } ; With sub reassociation, constant folding can eliminate the uses of %a. + define float @test20(float %a, float %b, float %c) nounwind { ; CHECK-LABEL: @test20( ; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[B:%.*]], [[C:%.*]] @@ -680,6 +674,19 @@ ret float %t7 } +define float @test20_reassoc_nsz(float %a, float %b, float %c) nounwind { +; CHECK-LABEL: @test20_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc nsz float [[B:%.*]], [[C:%.*]] +; CHECK-NEXT: [[T7:%.*]] = fneg reassoc nsz float [[TMP1]] +; CHECK-NEXT: ret float [[T7]] +; + %t3 = fsub reassoc nsz float %a, %b + %t5 = fsub reassoc nsz float %t3, %c + %t7 = fsub reassoc nsz float %t5, %a + ret float %t7 +} + +; Verify the fold is not done with only 'reassoc' ('nsz' is required). define float @test20_reassoc(float %a, float %b, float %c) nounwind { ; CHECK-LABEL: @test20_reassoc( ; CHECK-NEXT: [[T3:%.*]] = fsub reassoc float [[A:%.*]], [[B:%.*]] @@ -692,4 +699,3 @@ %t7 = fsub reassoc float %t5, %a ret float %t7 } - Index: llvm/test/Transforms/Reassociate/fast-basictest-reassociate-gvn.ll =================================================================== --- /dev/null +++ llvm/test/Transforms/Reassociate/fast-basictest-reassociate-gvn.ll @@ -0,0 +1,102 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; +; Test cases in this file are intended to be run with both reassociate and +; gvn passes enabled. +; +; Test numbering remains continuous across: +; - InstCombine/fast-basictest-instcombine.ll +; - Reassociate/fast-basictest-reassociate-gvn.ll +; - Reassociate/fast-basictest-reassociate-instcombine.ll +; - Reassociate/fast-basictest-reassociate.ll +; +; RUN: opt < %s -reassociate -gvn -S | FileCheck %s + +@fe = external global float +@fa = external global float +@fb = external global float +@fc = external global float +@ff = external global float + +; If two sums of the same operands in different order are counted with 'fast' +; flag and then stored to global variables, we can reuse the same value twice. +; Sums: +; - test4: (a+b)+c and (a+c)+b +; - test5: c+(a+b) and (c+a)+b +; - test6: c+(b+a) and (c+a)+b +; TODO: check if 'reassoc' flag is technically enough for this optimization +; (currently the transformation is not done with 'reassoc' only). + +define void @test4() { +; CHECK-LABEL: @test4( +; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 +; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 +; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 +; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] +; CHECK-NEXT: store float [[T2]], float* @fe, align 4 +; CHECK-NEXT: store float [[T2]], float* @ff, align 4 +; CHECK-NEXT: ret void +; + %A = load float, float* @fa + %B = load float, float* @fb + %C = load float, float* @fc + %t1 = fadd fast float %A, %B + %t2 = fadd fast float %t1, %C + %t3 = fadd fast float %A, %C + %t4 = fadd fast float %t3, %B + ; e = (a+b)+c; + store float %t2, float* @fe + ; f = (a+c)+b + store float %t4, float* @ff + ret void +} + +define void @test5() { +; CHECK-LABEL: @test5( +; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 +; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 +; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 +; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] +; CHECK-NEXT: store float [[T2]], float* @fe, align 4 +; CHECK-NEXT: store float [[T2]], float* @ff, align 4 +; CHECK-NEXT: ret void +; + %A = load float, float* @fa + %B = load float, float* @fb + %C = load float, float* @fc + %t1 = fadd fast float %A, %B + %t2 = fadd fast float %C, %t1 + %t3 = fadd fast float %C, %A + %t4 = fadd fast float %t3, %B + ; e = c+(a+b) + store float %t2, float* @fe + ; f = (c+a)+b + store float %t4, float* @ff + ret void +} + +define void @test6() { +; CHECK-LABEL: @test6( +; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 +; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 +; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 +; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] +; CHECK-NEXT: store float [[T2]], float* @fe, align 4 +; CHECK-NEXT: store float [[T2]], float* @ff, align 4 +; CHECK-NEXT: ret void +; + %A = load float, float* @fa + %B = load float, float* @fb + %C = load float, float* @fc + %t1 = fadd fast float %B, %A + %t2 = fadd fast float %C, %t1 + %t3 = fadd fast float %C, %A + %t4 = fadd fast float %t3, %B + ; e = c+(b+a) + store float %t2, float* @fe + ; f = (c+a)+b + store float %t4, float* @ff + ret void +} Index: llvm/test/Transforms/Reassociate/fast-basictest-reassociate-instcombine.ll =================================================================== --- /dev/null +++ llvm/test/Transforms/Reassociate/fast-basictest-reassociate-instcombine.ll @@ -0,0 +1,289 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; +; Test cases in this file are intended to be run with both reassociate and +; instcombine passes enabled. +; +; Test numbering remains continuous across: +; - InstCombine/fast-basictest-instcombine.ll +; - Reassociate/fast-basictest-reassociate-gvn.ll +; - Reassociate/fast-basictest-reassociate-instcombine.ll +; - Reassociate/fast-basictest-reassociate.ll +; +; RUN: opt < %s -reassociate -instcombine -S | FileCheck %s + +; test2 ... test19 - both reassociate and instcombine passes +; are required to perform a transform + +; ((a + (-3)) + b) + 3 -> a + b + +define float @test3(float %reg109, float %reg1111) { +; CHECK-LABEL: @test3( +; CHECK-NEXT: [[REG117:%.*]] = fadd fast float [[REG109:%.*]], [[REG1111:%.*]] +; CHECK-NEXT: ret float [[REG117]] +; + %reg115 = fadd fast float %reg109, -3.000000e+01 + %reg116 = fadd fast float %reg115, %reg1111 + %reg117 = fadd fast float %reg116, 3.000000e+01 + ret float %reg117 +} + +; Verify that fold is not done without 'fast' +define float @test3_no_FMF(float %reg109, float %reg1111) { +; CHECK-LABEL: @test3_no_FMF( +; CHECK-NEXT: [[REG115:%.*]] = fadd float [[REG109:%.*]], -3.000000e+01 +; CHECK-NEXT: [[REG116:%.*]] = fadd float [[REG115]], [[REG1111:%.*]] +; CHECK-NEXT: [[REG117:%.*]] = fadd float [[REG116]], 3.000000e+01 +; CHECK-NEXT: ret float [[REG117]] +; + %reg115 = fadd float %reg109, -3.000000e+01 + %reg116 = fadd float %reg115, %reg1111 + %reg117 = fadd float %reg116, 3.000000e+01 + ret float %reg117 +} + +define float @test3_reassoc(float %reg109, float %reg1111) { +; CHECK-LABEL: @test3_reassoc( +; CHECK-NEXT: [[REG115:%.*]] = fadd reassoc float [[REG109:%.*]], -3.000000e+01 +; CHECK-NEXT: [[REG116:%.*]] = fadd reassoc float [[REG115]], [[REG1111:%.*]] +; CHECK-NEXT: [[REG117:%.*]] = fadd reassoc float [[REG116]], 3.000000e+01 +; CHECK-NEXT: ret float [[REG117]] +; + %reg115 = fadd reassoc float %reg109, -3.000000e+01 + %reg116 = fadd reassoc float %reg115, %reg1111 + %reg117 = fadd reassoc float %reg116, 3.000000e+01 + ret float %reg117 +} + +; (x1 * 47) + (x2 * -47) => (x1 - x2) * 47 + +define float @test14(float %X1, float %X2) { +; CHECK-LABEL: @test14( +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float [[X1:%.*]], [[X2:%.*]] +; CHECK-NEXT: [[TMP2:%.*]] = fmul fast float [[TMP1]], 4.700000e+01 +; CHECK-NEXT: ret float [[TMP2]] +; + %B = fmul fast float %X1, 47. ; X1*47 + %C = fmul fast float %X2, -47. ; X2*-47 + %D = fadd fast float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + ret float %D +} + +; Check again with 'reassoc' and 'nsz' ('nsz' not technically required). +define float @test14_reassoc_nsz(float %X1, float %X2) { +; CHECK-LABEL: @test14_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fsub reassoc nsz float [[X1:%.*]], [[X2:%.*]] +; CHECK-NEXT: [[TMP2:%.*]] = fmul reassoc nsz float [[TMP1]], 4.700000e+01 +; CHECK-NEXT: ret float [[TMP2]] +; + %B = fmul reassoc nsz float %X1, 47. ; X1*47 + %C = fmul reassoc nsz float %X2, -47. ; X2*-47 + %D = fadd reassoc nsz float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + ret float %D +} + +; TODO: This doesn't require 'nsz'. It should fold to ((x1 - x2) * 47.0) +define float @test14_reassoc(float %X1, float %X2) { +; CHECK-LABEL: @test14_reassoc( +; CHECK-NEXT: [[B:%.*]] = fmul reassoc float [[X1:%.*]], 4.700000e+01 +; CHECK-NEXT: [[C:%.*]] = fmul reassoc float [[X2:%.*]], 4.700000e+01 +; CHECK-NEXT: [[TMP1:%.*]] = fsub reassoc float [[B]], [[C]] +; CHECK-NEXT: ret float [[TMP1]] +; + %B = fmul reassoc float %X1, 47. ; X1*47 + %C = fmul reassoc float %X2, -47. ; X2*-47 + %D = fadd reassoc float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + ret float %D +} + +; (b+(a+1234))+-a -> b+1234 + +define float @test16(float %b, float %a) { +; CHECK-LABEL: @test16( +; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[B:%.*]], 1.234000e+03 +; CHECK-NEXT: ret float [[TMP1]] +; + %1 = fadd fast float %a, 1234.0 + %2 = fadd fast float %b, %1 + %3 = fsub fast float 0.0, %a + %4 = fadd fast float %2, %3 + ret float %4 +} + +define float @test16_unary_fneg(float %b, float %a) { +; CHECK-LABEL: @test16_unary_fneg( +; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[B:%.*]], 1.234000e+03 +; CHECK-NEXT: ret float [[TMP1]] +; + %1 = fadd fast float %a, 1234.0 + %2 = fadd fast float %b, %1 + %3 = fneg fast float %a + %4 = fadd fast float %2, %3 + ret float %4 +} + +; TODO: check if it is possible to perform the optimization without 'fast' +; with 'reassoc' and 'nsz' only. +define float @test16_reassoc_nsz(float %b, float %a) { +; CHECK-LABEL: @test16_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc nsz float [[A:%.*]], 1.234000e+03 +; CHECK-NEXT: [[TMP2:%.*]] = fadd reassoc nsz float [[TMP1]], [[B:%.*]] +; CHECK-NEXT: [[TMP3:%.*]] = fsub reassoc nsz float [[TMP2]], [[A]] +; CHECK-NEXT: ret float [[TMP3]] +; + %1 = fadd reassoc nsz float %a, 1234.0 + %2 = fadd reassoc nsz float %b, %1 + %3 = fsub reassoc nsz float 0.0, %a + %4 = fadd reassoc nsz float %2, %3 + ret float %4 +} + +define float @test16_reassoc(float %b, float %a) { +; CHECK-LABEL: @test16_reassoc( +; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc float [[A:%.*]], 1.234000e+03 +; CHECK-NEXT: [[TMP2:%.*]] = fadd reassoc float [[TMP1]], [[B:%.*]] +; CHECK-NEXT: [[TMP3:%.*]] = fsub reassoc float 0.000000e+00, [[A]] +; CHECK-NEXT: [[TMP4:%.*]] = fadd reassoc float [[TMP3]], [[TMP2]] +; CHECK-NEXT: ret float [[TMP4]] +; + %1 = fadd reassoc float %a, 1234.0 + %2 = fadd reassoc float %b, %1 + %3 = fsub reassoc float 0.0, %a + %4 = fadd reassoc float %2, %3 + ret float %4 +} + +; Test that we can turn things like X*-(Y*Z) -> X*-1*Y*Z. + +define float @test17(float %a, float %b, float %z) { +; CHECK-LABEL: @test17( +; CHECK-NEXT: [[C:%.*]] = fmul fast float [[A:%.*]], 1.234500e+04 +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[C]], [[B:%.*]] +; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[Z:%.*]] +; CHECK-NEXT: ret float [[F]] +; + %c = fsub fast float 0.000000e+00, %z + %d = fmul fast float %a, %b + %e = fmul fast float %c, %d + %f = fmul fast float %e, 1.234500e+04 + %g = fsub fast float 0.000000e+00, %f + ret float %g +} + +define float @test17_unary_fneg(float %a, float %b, float %z) { +; CHECK-LABEL: @test17_unary_fneg( +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[A:%.*]], 1.234500e+04 +; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[B:%.*]] +; CHECK-NEXT: [[G:%.*]] = fmul fast float [[F]], [[Z:%.*]] +; CHECK-NEXT: ret float [[G]] +; + %c = fneg fast float %z + %d = fmul fast float %a, %b + %e = fmul fast float %c, %d + %f = fmul fast float %e, 1.234500e+04 + %g = fneg fast float %f + ret float %g +} + +; TODO: check if it is possible to perform the optimization without 'fast' +; with 'reassoc' and 'nsz' only. +define float @test17_reassoc_nsz(float %a, float %b, float %z) { +; CHECK-LABEL: @test17_reassoc_nsz( +; CHECK-NEXT: [[C:%.*]] = fneg reassoc nsz float [[Z:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul reassoc nsz float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[E:%.*]] = fmul reassoc nsz float [[D]], [[C]] +; CHECK-NEXT: [[G:%.*]] = fmul reassoc nsz float [[E]], -1.234500e+04 +; CHECK-NEXT: ret float [[G]] +; + %c = fsub reassoc nsz float 0.000000e+00, %z + %d = fmul reassoc nsz float %a, %b + %e = fmul reassoc nsz float %c, %d + %f = fmul reassoc nsz float %e, 1.234500e+04 + %g = fsub reassoc nsz float 0.000000e+00, %f + ret float %g +} + +define float @test17_reassoc(float %a, float %b, float %z) { +; CHECK-LABEL: @test17_reassoc( +; CHECK-NEXT: [[C:%.*]] = fsub reassoc float 0.000000e+00, [[Z:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul reassoc float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[E:%.*]] = fmul reassoc float [[D]], [[C]] +; CHECK-NEXT: [[F:%.*]] = fmul reassoc float [[E]], 1.234500e+04 +; CHECK-NEXT: [[G:%.*]] = fsub reassoc float 0.000000e+00, [[F]] +; CHECK-NEXT: ret float [[G]] +; + %c = fsub reassoc float 0.000000e+00, %z + %d = fmul reassoc float %a, %b + %e = fmul reassoc float %c, %d + %f = fmul reassoc float %e, 1.234500e+04 + %g = fsub reassoc float 0.000000e+00, %f + ret float %g +} + +; With sub reassociation, constant folding can eliminate the 12 and -12 constants. + +define float @test19(float %A, float %B) { +; CHECK-LABEL: @test19( +; CHECK-NEXT: [[Z:%.*]] = fsub fast float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: ret float [[Z]] +; + %X = fadd fast float -1.200000e+01, %A + %Y = fsub fast float %X, %B + %Z = fadd fast float %Y, 1.200000e+01 + ret float %Z +} + +define float @test19_reassoc(float %A, float %B) { +; CHECK-LABEL: @test19_reassoc( +; CHECK-NEXT: [[X:%.*]] = fadd reassoc float [[A:%.*]], -1.200000e+01 +; CHECK-NEXT: [[Y:%.*]] = fsub reassoc float [[X]], [[B:%.*]] +; CHECK-NEXT: [[Z:%.*]] = fadd reassoc float [[Y]], 1.200000e+01 +; CHECK-NEXT: ret float [[Z]] +; + %X = fadd reassoc float -1.200000e+01, %A + %Y = fsub reassoc float %X, %B + %Z = fadd reassoc float %Y, 1.200000e+01 + ret float %Z +} + +; test20 - check that the bug described in the revision does not appear: +; https://reviews.llvm.org/D72521 + +; With sub reassociation, constant folding can eliminate the uses of %a. + +define float @test20(float %a, float %b, float %c) nounwind { +; CHECK-LABEL: @test20( +; CHECK-NEXT: [[TMP1:%.*]] = fadd fast float [[B:%.*]], [[C:%.*]] +; CHECK-NEXT: [[T7:%.*]] = fneg fast float [[TMP1]] +; CHECK-NEXT: ret float [[T7]] +; + %t3 = fsub fast float %a, %b + %t5 = fsub fast float %t3, %c + %t7 = fsub fast float %t5, %a + ret float %t7 +} + +define float @test20_reassoc_nsz(float %a, float %b, float %c) nounwind { +; CHECK-LABEL: @test20_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc nsz float [[B:%.*]], [[C:%.*]] +; CHECK-NEXT: [[T7:%.*]] = fneg reassoc nsz float [[TMP1]] +; CHECK-NEXT: ret float [[T7]] +; + %t3 = fsub reassoc nsz float %a, %b + %t5 = fsub reassoc nsz float %t3, %c + %t7 = fsub reassoc nsz float %t5, %a + ret float %t7 +} + +; Verify the fold is not done with only 'reassoc' ('nsz' is required). +define float @test20_reassoc(float %a, float %b, float %c) nounwind { +; CHECK-LABEL: @test20_reassoc( +; CHECK-NEXT: [[T3:%.*]] = fsub reassoc float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[T5:%.*]] = fsub reassoc float [[T3]], [[C:%.*]] +; CHECK-NEXT: [[T7:%.*]] = fsub reassoc float [[T5]], [[A]] +; CHECK-NEXT: ret float [[T7]] +; + %t3 = fsub reassoc float %a, %b + %t5 = fsub reassoc float %t3, %c + %t7 = fsub reassoc float %t5, %a + ret float %t7 +} Index: llvm/test/Transforms/Reassociate/fast-basictest-reassociate.ll =================================================================== --- /dev/null +++ llvm/test/Transforms/Reassociate/fast-basictest-reassociate.ll @@ -0,0 +1,588 @@ +; NOTE: Assertions have been autogenerated by utils/update_test_checks.py +; +; Test numbering remains continuous across: +; - InstCombine/fast-basictest-instcombine.ll +; - Reassociate/fast-basictest-reassociate-gvn.ll +; - Reassociate/fast-basictest-reassociate-instcombine.ll +; - Reassociate/fast-basictest-reassociate.ll +; +; RUN: opt < %s -reassociate -S | FileCheck %s + +; ((a + (-3)) + b) + 3 -> a + b +; That only works with both instcombine and reassociate passes enabled. +; Check that reassociate is not enough. + +; TODO: check if we can eliminate zero add. +define float @test3(float %reg109, float %reg1111) { +; CHECK-LABEL: @test3( +; CHECK-NEXT: [[REG116:%.*]] = fadd fast float [[REG109:%.*]], 0.000000e+00 +; CHECK-NEXT: [[REG117:%.*]] = fadd fast float [[REG116]], [[REG1111:%.*]] +; CHECK-NEXT: ret float [[REG117]] +; + %reg115 = fadd fast float %reg109, -3.000000e+01 + %reg116 = fadd fast float %reg115, %reg1111 + %reg117 = fadd fast float %reg116, 3.000000e+01 + ret float %reg117 +} + +define float @test3_no_FMF(float %reg109, float %reg1111) { +; CHECK-LABEL: @test3_no_FMF( +; CHECK-NEXT: [[REG115:%.*]] = fadd float [[REG109:%.*]], -3.000000e+01 +; CHECK-NEXT: [[REG116:%.*]] = fadd float [[REG115]], [[REG1111:%.*]] +; CHECK-NEXT: [[REG117:%.*]] = fadd float [[REG116]], 3.000000e+01 +; CHECK-NEXT: ret float [[REG117]] +; + %reg115 = fadd float %reg109, -3.000000e+01 + %reg116 = fadd float %reg115, %reg1111 + %reg117 = fadd float %reg116, 3.000000e+01 + ret float %reg117 +} + +define float @test3_reassoc(float %reg109, float %reg1111) { +; CHECK-LABEL: @test3_reassoc( +; CHECK-NEXT: [[REG115:%.*]] = fadd reassoc float [[REG109:%.*]], -3.000000e+01 +; CHECK-NEXT: [[REG116:%.*]] = fadd reassoc float [[REG115]], [[REG1111:%.*]] +; CHECK-NEXT: [[REG117:%.*]] = fadd reassoc float [[REG116]], 3.000000e+01 +; CHECK-NEXT: ret float [[REG117]] +; + %reg115 = fadd reassoc float %reg109, -3.000000e+01 + %reg116 = fadd reassoc float %reg115, %reg1111 + %reg117 = fadd reassoc float %reg116, 3.000000e+01 + ret float %reg117 +} + +@fe = external global float +@fa = external global float +@fb = external global float +@fc = external global float +@ff = external global float + +; If two sums of the same operands in different order are counted with 'fast' +; flag and then stored to global variables, we can reuse the same value twice. +; Sums: +; - test4: (a+b)+c and (a+c)+b +; - test5: c+(a+b) and (c+a)+b +; - test6: c+(b+a) and (c+a)+b +; +; That only works with both gvn and reassociate passes enabled. +; Check that reassociate is not enough. + +define void @test4() { +; CHECK-LABEL: @test4( +; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 +; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 +; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 +; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] +; CHECK-NEXT: [[T3:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T4:%.*]] = fadd fast float [[T3]], [[C]] +; CHECK-NEXT: store float [[T2]], float* @fe, align 4 +; CHECK-NEXT: store float [[T4]], float* @ff, align 4 +; CHECK-NEXT: ret void +; + %A = load float, float* @fa + %B = load float, float* @fb + %C = load float, float* @fc + %t1 = fadd fast float %A, %B + %t2 = fadd fast float %t1, %C + %t3 = fadd fast float %A, %C + %t4 = fadd fast float %t3, %B + ; e = (a+b)+c; + store float %t2, float* @fe + ; f = (a+c)+b + store float %t4, float* @ff + ret void +} + +define void @test5() { +; CHECK-LABEL: @test5( +; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 +; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 +; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 +; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] +; CHECK-NEXT: [[T3:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T4:%.*]] = fadd fast float [[T3]], [[C]] +; CHECK-NEXT: store float [[T2]], float* @fe, align 4 +; CHECK-NEXT: store float [[T4]], float* @ff, align 4 +; CHECK-NEXT: ret void +; + %A = load float, float* @fa + %B = load float, float* @fb + %C = load float, float* @fc + %t1 = fadd fast float %A, %B + %t2 = fadd fast float %C, %t1 + %t3 = fadd fast float %C, %A + %t4 = fadd fast float %t3, %B + ; e = c+(a+b) + store float %t2, float* @fe + ; f = (c+a)+b + store float %t4, float* @ff + ret void +} + +define void @test6() { +; CHECK-LABEL: @test6( +; CHECK-NEXT: [[A:%.*]] = load float, float* @fa, align 4 +; CHECK-NEXT: [[B:%.*]] = load float, float* @fb, align 4 +; CHECK-NEXT: [[C:%.*]] = load float, float* @fc, align 4 +; CHECK-NEXT: [[T1:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T2:%.*]] = fadd fast float [[T1]], [[C]] +; CHECK-NEXT: [[T3:%.*]] = fadd fast float [[B]], [[A]] +; CHECK-NEXT: [[T4:%.*]] = fadd fast float [[T3]], [[C]] +; CHECK-NEXT: store float [[T2]], float* @fe, align 4 +; CHECK-NEXT: store float [[T4]], float* @ff, align 4 +; CHECK-NEXT: ret void +; + %A = load float, float* @fa + %B = load float, float* @fb + %C = load float, float* @fc + %t1 = fadd fast float %B, %A + %t2 = fadd fast float %C, %t1 + %t3 = fadd fast float %C, %A + %t4 = fadd fast float %t3, %B + ; e = c+(b+a) + store float %t2, float* @fe + ; f = (c+a)+b + store float %t4, float* @ff + ret void +} + +define float @test7(float %A, float %B, float %C) { +; CHECK-LABEL: @test7( +; CHECK-NEXT: [[REASS_ADD1:%.*]] = fadd fast float [[C:%.*]], [[B:%.*]] +; CHECK-NEXT: [[REASS_MUL2:%.*]] = fmul fast float [[A:%.*]], [[A]] +; CHECK-NEXT: [[REASS_MUL:%.*]] = fmul fast float [[REASS_MUL2]], [[REASS_ADD1]] +; CHECK-NEXT: ret float [[REASS_MUL]] +; + %aa = fmul fast float %A, %A + %aab = fmul fast float %aa, %B + %ac = fmul fast float %A, %C + %aac = fmul fast float %ac, %A + %r = fadd fast float %aab, %aac + ret float %r +} + +define float @test7_reassoc(float %A, float %B, float %C) { +; CHECK-LABEL: @test7_reassoc( +; CHECK-NEXT: [[AA:%.*]] = fmul reassoc float [[A:%.*]], [[A]] +; CHECK-NEXT: [[AAB:%.*]] = fmul reassoc float [[AA]], [[B:%.*]] +; CHECK-NEXT: [[AC:%.*]] = fmul reassoc float [[A]], [[C:%.*]] +; CHECK-NEXT: [[AAC:%.*]] = fmul reassoc float [[A]], [[AC]] +; CHECK-NEXT: [[R:%.*]] = fadd reassoc float [[AAB]], [[AAC]] +; CHECK-NEXT: ret float [[R]] +; + %aa = fmul reassoc float %A, %A + %aab = fmul reassoc float %aa, %B + %ac = fmul reassoc float %A, %C + %aac = fmul reassoc float %ac, %A + %r = fadd reassoc float %aab, %aac + ret float %r +} + +; (-X)*Y + Z -> Z-X*Y +; TODO: check why IR transformation of test8 with 'fast' math flag +; is worse than without it (and even without transformation) + +define float @test8(float %X, float %Y, float %Z) { +; CHECK-LABEL: @test8( +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float 0.000000e+00, 0.000000e+00 +; CHECK-NEXT: [[A:%.*]] = fmul fast float [[Y:%.*]], [[X:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul fast float [[A]], 1.000000e+00 +; CHECK-NEXT: [[TMP2:%.*]] = fsub fast float [[Z:%.*]], [[B]] +; CHECK-NEXT: ret float [[TMP2]] +; + %A = fsub fast float 0.0, %X + %B = fmul fast float %A, %Y + %C = fadd fast float %B, %Z + ret float %C +} + +define float @test8_unary_fneg(float %X, float %Y, float %Z) { +; CHECK-LABEL: @test8_unary_fneg( +; CHECK-NEXT: [[TMP1:%.*]] = fneg fast float 0.000000e+00 +; CHECK-NEXT: [[A:%.*]] = fmul fast float [[Y:%.*]], [[X:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul fast float [[A]], 1.000000e+00 +; CHECK-NEXT: [[TMP2:%.*]] = fsub fast float [[Z:%.*]], [[B]] +; CHECK-NEXT: ret float [[TMP2]] +; + %A = fneg fast float %X + %B = fmul fast float %A, %Y + %C = fadd fast float %B, %Z + ret float %C +} + +define float @test8_reassoc_nsz(float %X, float %Y, float %Z) { +; CHECK-LABEL: @test8_reassoc_nsz( +; CHECK-NEXT: [[A:%.*]] = fsub reassoc nsz float 0.000000e+00, [[X:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul reassoc nsz float [[A]], [[Y:%.*]] +; CHECK-NEXT: [[C:%.*]] = fadd reassoc nsz float [[B]], [[Z:%.*]] +; CHECK-NEXT: ret float [[C]] +; + %A = fsub reassoc nsz float 0.0, %X + %B = fmul reassoc nsz float %A, %Y + %C = fadd reassoc nsz float %B, %Z + ret float %C +} + +; Verify that fold is not done only with 'reassoc' ('nsz' is required) +define float @test8_reassoc(float %X, float %Y, float %Z) { +; CHECK-LABEL: @test8_reassoc( +; CHECK-NEXT: [[A:%.*]] = fsub reassoc float 0.000000e+00, [[X:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul reassoc float [[A]], [[Y:%.*]] +; CHECK-NEXT: [[C:%.*]] = fadd reassoc float [[B]], [[Z:%.*]] +; CHECK-NEXT: ret float [[C]] +; + %A = fsub reassoc float 0.0, %X + %B = fmul reassoc float %A, %Y + %C = fadd reassoc float %B, %Z + ret float %C +} + +define float @test9(float %X) { +; CHECK-LABEL: @test9( +; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[X:%.*]], 9.400000e+01 +; CHECK-NEXT: ret float [[FACTOR]] +; + %Y = fmul fast float %X, 4.700000e+01 + %Z = fadd fast float %Y, %Y + ret float %Z +} + +; Side note: (x + x + x) and (3*x) each have only a single rounding. So +; transforming x+x+x to 3*x is always safe, even without any FMF. +; To avoid that special-case, we have the addition of 'x' four times, here. +define float @test10(float %X) { +; CHECK-LABEL: @test10( +; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[X:%.*]], 4.000000e+00 +; CHECK-NEXT: ret float [[FACTOR]] +; + %Y = fadd fast float %X ,%X + %Z = fadd fast float %Y, %X + %W = fadd fast float %Z, %X + ret float %W +} + +define float @test11(float %W) { +; CHECK-LABEL: @test11( +; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[W:%.*]], 3.810000e+02 +; CHECK-NEXT: ret float [[FACTOR]] +; + %X = fmul fast float %W, 127.0 + %Y = fadd fast float %X ,%X + %Z = fadd fast float %Y, %X + ret float %Z +} + +define float @test12(float %X) { +; CHECK-LABEL: @test12( +; CHECK-NEXT: [[TMP1:%.*]] = fneg fast float 0.000000e+00 +; CHECK-NEXT: [[FACTOR:%.*]] = fmul fast float [[X:%.*]], -3.000000e+00 +; CHECK-NEXT: [[Z:%.*]] = fadd fast float [[FACTOR]], 6.000000e+00 +; CHECK-NEXT: ret float [[Z]] +; + %A = fsub fast float 1.000000e+00, %X + %B = fsub fast float 2.000000e+00, %X + %C = fsub fast float 3.000000e+00, %X + %Y = fadd fast float %A ,%B + %Z = fadd fast float %Y, %C + ret float %Z +} + +; TODO: check why IR transformation of test13 with 'fast' math flag +; is worse than without it (and even without transformation) + +define float @test13(float %X1, float %X2, float %X3) { +; CHECK-LABEL: @test13( +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float 0.000000e+00, 0.000000e+00 +; CHECK-NEXT: [[A:%.*]] = fmul fast float [[X2:%.*]], [[X1:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul fast float [[A]], 1.000000e+00 +; CHECK-NEXT: [[C:%.*]] = fmul fast float [[X3:%.*]], [[X1]] +; CHECK-NEXT: [[TMP2:%.*]] = fsub fast float [[C]], [[B]] +; CHECK-NEXT: ret float [[TMP2]] +; + %A = fsub fast float 0.000000e+00, %X1 + %B = fmul fast float %A, %X2 ; -X1*X2 + %C = fmul fast float %X1, %X3 ; X1*X3 + %D = fadd fast float %B, %C ; -X1*X2 + X1*X3 -> X1*(X3-X2) + ret float %D +} + +define float @test13_unary_fneg(float %X1, float %X2, float %X3) { +; CHECK-LABEL: @test13_unary_fneg( +; CHECK-NEXT: [[TMP1:%.*]] = fneg fast float 0.000000e+00 +; CHECK-NEXT: [[A:%.*]] = fmul fast float [[X2:%.*]], [[X1:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul fast float [[A]], 1.000000e+00 +; CHECK-NEXT: [[C:%.*]] = fmul fast float [[X3:%.*]], [[X1]] +; CHECK-NEXT: [[TMP2:%.*]] = fsub fast float [[C]], [[B]] +; CHECK-NEXT: ret float [[TMP2]] +; + %A = fneg fast float %X1 + %B = fmul fast float %A, %X2 ; -X1*X2 + %C = fmul fast float %X1, %X3 ; X1*X3 + %D = fadd fast float %B, %C ; -X1*X2 + X1*X3 -> X1*(X3-X2) + ret float %D +} + +define float @test13_reassoc_nsz(float %X1, float %X2, float %X3) { +; CHECK-LABEL: @test13_reassoc_nsz( +; CHECK-NEXT: [[A:%.*]] = fsub reassoc nsz float 0.000000e+00, [[X1:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul reassoc nsz float [[A]], [[X2:%.*]] +; CHECK-NEXT: [[C:%.*]] = fmul reassoc nsz float [[X1]], [[X3:%.*]] +; CHECK-NEXT: [[D:%.*]] = fadd reassoc nsz float [[B]], [[C]] +; CHECK-NEXT: ret float [[D]] +; + %A = fsub reassoc nsz float 0.000000e+00, %X1 + %B = fmul reassoc nsz float %A, %X2 ; -X1*X2 + %C = fmul reassoc nsz float %X1, %X3 ; X1*X3 + %D = fadd reassoc nsz float %B, %C ; -X1*X2 + X1*X3 -> X1*(X3-X2) + ret float %D +} + +; TODO: check if 'nsz' is technically required. Currently the optimization +; is not done with only 'reassoc' without 'nsz'. +define float @test13_reassoc(float %X1, float %X2, float %X3) { +; CHECK-LABEL: @test13_reassoc( +; CHECK-NEXT: [[A:%.*]] = fsub reassoc float 0.000000e+00, [[X1:%.*]] +; CHECK-NEXT: [[B:%.*]] = fmul reassoc float [[A]], [[X2:%.*]] +; CHECK-NEXT: [[C:%.*]] = fmul reassoc float [[X1]], [[X3:%.*]] +; CHECK-NEXT: [[D:%.*]] = fadd reassoc float [[B]], [[C]] +; CHECK-NEXT: ret float [[D]] +; + %A = fsub reassoc float 0.000000e+00, %X1 + %B = fmul reassoc float %A, %X2 ; -X1*X2 + %C = fmul reassoc float %X1, %X3 ; X1*X3 + %D = fadd reassoc float %B, %C ; -X1*X2 + X1*X3 -> X1*(X3-X2) + ret float %D +} + +; (x1 * 47) + (x2 * -47) => (x1 - x2) * 47 +; That only works with both instcombine and reassociate passes enabled. +; Check that reassociate is not enough. + +define float @test14(float %X1, float %X2) { +; CHECK-LABEL: @test14( +; CHECK-NEXT: [[B:%.*]] = fmul fast float [[X1:%.*]], 4.700000e+01 +; CHECK-NEXT: [[C:%.*]] = fmul fast float [[X2:%.*]], 4.700000e+01 +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float [[B]], [[C]] +; CHECK-NEXT: ret float [[TMP1]] +; + %B = fmul fast float %X1, 47. ; X1*47 + %C = fmul fast float %X2, -47. ; X2*-47 + %D = fadd fast float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + ret float %D +} + +define float @test14_reassoc_nsz(float %X1, float %X2) { +; CHECK-LABEL: @test14_reassoc_nsz( +; CHECK-NEXT: [[B:%.*]] = fmul reassoc nsz float [[X1:%.*]], 4.700000e+01 +; CHECK-NEXT: [[C:%.*]] = fmul reassoc nsz float [[X2:%.*]], 4.700000e+01 +; CHECK-NEXT: [[TMP1:%.*]] = fsub reassoc nsz float [[B]], [[C]] +; CHECK-NEXT: ret float [[TMP1]] +; + %B = fmul reassoc nsz float %X1, 47. ; X1*47 + %C = fmul reassoc nsz float %X2, -47. ; X2*-47 + %D = fadd reassoc nsz float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + ret float %D +} + +define float @test14_reassoc(float %X1, float %X2) { +; CHECK-LABEL: @test14_reassoc( +; CHECK-NEXT: [[B:%.*]] = fmul reassoc float [[X1:%.*]], 4.700000e+01 +; CHECK-NEXT: [[C:%.*]] = fmul reassoc float [[X2:%.*]], 4.700000e+01 +; CHECK-NEXT: [[TMP1:%.*]] = fsub reassoc float [[B]], [[C]] +; CHECK-NEXT: ret float [[TMP1]] +; + %B = fmul reassoc float %X1, 47. ; X1*47 + %C = fmul reassoc float %X2, -47. ; X2*-47 + %D = fadd reassoc float %B, %C ; X1*47 + X2*-47 -> 47*(X1-X2) + ret float %D +} + +define float @test15(float %arg) { +; CHECK-LABEL: @test15( +; CHECK-NEXT: [[T2:%.*]] = fmul fast float [[ARG:%.*]], 1.440000e+02 +; CHECK-NEXT: ret float [[T2]] +; + %t1 = fmul fast float 1.200000e+01, %arg + %t2 = fmul fast float %t1, 1.200000e+01 + ret float %t2 +} + +; TODO: check if we can transform the code with 'reassoc' only. +; The same IR is transformed to one fmul in instcombine pass. +define float @test15_reassoc(float %arg) { +; CHECK-LABEL: @test15_reassoc( +; CHECK-NEXT: [[T1:%.*]] = fmul reassoc float [[ARG:%.*]], 1.200000e+01 +; CHECK-NEXT: [[T2:%.*]] = fmul reassoc float [[T1]], 1.200000e+01 +; CHECK-NEXT: ret float [[T2]] +; + %t1 = fmul reassoc float 1.200000e+01, %arg + %t2 = fmul reassoc float %t1, 1.200000e+01 + ret float %t2 +} + +; (b+(a+1234))+-a -> b+1234 +; That only works with both instcombine and reassociate passes enabled. +; Check that reassociate is not enough. + +; TODO: check if we can remove dead fsub. +define float @test16(float %b, float %a) { +; CHECK-LABEL: @test16( +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float 0.000000e+00, [[A:%.*]] +; CHECK-NEXT: [[TMP2:%.*]] = fadd fast float [[B:%.*]], 1.234000e+03 +; CHECK-NEXT: ret float [[TMP2]] +; + %1 = fadd fast float %a, 1234.0 + %2 = fadd fast float %b, %1 + %3 = fsub fast float 0.0, %a + %4 = fadd fast float %2, %3 + ret float %4 +} + +; TODO: check if we can remove dead fneg. +define float @test16_unary_fneg(float %b, float %a) { +; CHECK-LABEL: @test16_unary_fneg( +; CHECK-NEXT: [[TMP1:%.*]] = fneg fast float [[A:%.*]] +; CHECK-NEXT: [[TMP2:%.*]] = fadd fast float [[B:%.*]], 1.234000e+03 +; CHECK-NEXT: ret float [[TMP2]] +; + %1 = fadd fast float %a, 1234.0 + %2 = fadd fast float %b, %1 + %3 = fneg fast float %a + %4 = fadd fast float %2, %3 + ret float %4 +} + +define float @test16_reassoc_nsz(float %b, float %a) { +; CHECK-LABEL: @test16_reassoc_nsz( +; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc nsz float [[A:%.*]], 1.234000e+03 +; CHECK-NEXT: [[TMP2:%.*]] = fadd reassoc nsz float [[B:%.*]], [[TMP1]] +; CHECK-NEXT: [[TMP3:%.*]] = fsub reassoc nsz float 0.000000e+00, [[A]] +; CHECK-NEXT: [[TMP4:%.*]] = fadd reassoc nsz float [[TMP3]], [[TMP2]] +; CHECK-NEXT: ret float [[TMP4]] +; + %1 = fadd reassoc nsz float %a, 1234.0 + %2 = fadd reassoc nsz float %b, %1 + %3 = fsub reassoc nsz float 0.0, %a + %4 = fadd reassoc nsz float %2, %3 + ret float %4 +} + +define float @test16_reassoc(float %b, float %a) { +; CHECK-LABEL: @test16_reassoc( +; CHECK-NEXT: [[TMP1:%.*]] = fadd reassoc float [[A:%.*]], 1.234000e+03 +; CHECK-NEXT: [[TMP2:%.*]] = fadd reassoc float [[B:%.*]], [[TMP1]] +; CHECK-NEXT: [[TMP3:%.*]] = fsub reassoc float 0.000000e+00, [[A]] +; CHECK-NEXT: [[TMP4:%.*]] = fadd reassoc float [[TMP3]], [[TMP2]] +; CHECK-NEXT: ret float [[TMP4]] +; + %1 = fadd reassoc float %a, 1234.0 + %2 = fadd reassoc float %b, %1 + %3 = fsub reassoc float 0.0, %a + %4 = fadd reassoc float %2, %3 + ret float %4 +} + +; Test that we can turn things like X*-(Y*Z) -> X*-1*Y*Z. +; That only works with both instcombine and reassociate passes enabled. +; Check that reassociate is not enough. + +define float @test17(float %a, float %b, float %z) { +; CHECK-LABEL: @test17( +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float 0.000000e+00, 0.000000e+00 +; CHECK-NEXT: [[C:%.*]] = fmul fast float [[A:%.*]], 1.234500e+04 +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[C]], [[B:%.*]] +; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[Z:%.*]] +; CHECK-NEXT: [[TMP2:%.*]] = fadd fast float [[F]], 0.000000e+00 +; CHECK-NEXT: ret float [[TMP2]] +; + %c = fsub fast float 0.000000e+00, %z + %d = fmul fast float %a, %b + %e = fmul fast float %c, %d + %f = fmul fast float %e, 1.234500e+04 + %g = fsub fast float 0.000000e+00, %f + ret float %g +} + +define float @test17_unary_fneg(float %a, float %b, float %z) { +; CHECK-LABEL: @test17_unary_fneg( +; CHECK-NEXT: [[TMP1:%.*]] = fneg fast float 0.000000e+00 +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[A:%.*]], 1.234500e+04 +; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[B:%.*]] +; CHECK-NEXT: [[G:%.*]] = fmul fast float [[F]], [[Z:%.*]] +; CHECK-NEXT: ret float [[G]] +; + %c = fneg fast float %z + %d = fmul fast float %a, %b + %e = fmul fast float %c, %d + %f = fmul fast float %e, 1.234500e+04 + %g = fneg fast float %f + ret float %g +} + +define float @test17_reassoc(float %a, float %b, float %z) { +; CHECK-LABEL: @test17_reassoc( +; CHECK-NEXT: [[C:%.*]] = fsub reassoc float 0.000000e+00, [[Z:%.*]] +; CHECK-NEXT: [[D:%.*]] = fmul reassoc float [[A:%.*]], [[B:%.*]] +; CHECK-NEXT: [[E:%.*]] = fmul reassoc float [[D]], [[C]] +; CHECK-NEXT: [[F:%.*]] = fmul reassoc float [[E]], 1.234500e+04 +; CHECK-NEXT: [[G:%.*]] = fsub reassoc float 0.000000e+00, [[F]] +; CHECK-NEXT: ret float [[G]] +; + %c = fsub reassoc float 0.000000e+00, %z + %d = fmul reassoc float %a, %b + %e = fmul reassoc float %c, %d + %f = fmul reassoc float %e, 1.234500e+04 + %g = fsub reassoc float 0.000000e+00, %f + ret float %g +} + +; TODO: check if we can remove: +; - fsub fast 0, 0 +; - fadd fast x, 0 +; ... as 'fast' implies 'nsz' +define float @test18(float %a, float %b, float %z) { +; CHECK-LABEL: @test18( +; CHECK-NEXT: [[TMP1:%.*]] = fsub fast float 0.000000e+00, 0.000000e+00 +; CHECK-NEXT: [[C:%.*]] = fmul fast float [[A:%.*]], 4.000000e+01 +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[C]], [[Z:%.*]] +; CHECK-NEXT: [[TMP2:%.*]] = fadd fast float [[E]], 0.000000e+00 +; CHECK-NEXT: ret float [[TMP2]] +; + %d = fmul fast float %z, 4.000000e+01 + %c = fsub fast float 0.000000e+00, %d + %e = fmul fast float %a, %c + %f = fsub fast float 0.000000e+00, %e + ret float %f +} + +; TODO: check if we can remove fneg fast 0 as 'fast' implies 'nsz' +define float @test18_unary_fneg(float %a, float %b, float %z) { +; CHECK-LABEL: @test18_unary_fneg( +; CHECK-NEXT: [[TMP1:%.*]] = fneg fast float 0.000000e+00 +; CHECK-NEXT: [[E:%.*]] = fmul fast float [[A:%.*]], 4.000000e+01 +; CHECK-NEXT: [[F:%.*]] = fmul fast float [[E]], [[Z:%.*]] +; CHECK-NEXT: ret float [[F]] +; + %d = fmul fast float %z, 4.000000e+01 + %c = fneg fast float %d + %e = fmul fast float %a, %c + %f = fneg fast float %e + ret float %f +} + +; With sub reassociation, constant folding can eliminate the 12 and -12 constants. +; TODO: check if we can remove fadd fast x, 0 as 'fast' implies 'nsz' +define float @test19(float %A, float %B) { +; CHECK-LABEL: @test19( +; CHECK-NEXT: [[B_NEG:%.*]] = fneg fast float [[B:%.*]] +; CHECK-NEXT: [[Y:%.*]] = fadd fast float [[A:%.*]], 0.000000e+00 +; CHECK-NEXT: [[Z:%.*]] = fadd fast float [[Y]], [[B_NEG]] +; CHECK-NEXT: ret float [[Z]] +; + %X = fadd fast float -1.200000e+01, %A + %Y = fsub fast float %X, %B + %Z = fadd fast float %Y, 1.200000e+01 + ret float %Z +}