diff options
Diffstat (limited to 'llvm/test/Transforms/InstCombine/select-extractelement.ll')
| -rw-r--r-- | llvm/test/Transforms/InstCombine/select-extractelement.ll | 55 |
1 files changed, 23 insertions, 32 deletions
diff --git a/llvm/test/Transforms/InstCombine/select-extractelement.ll b/llvm/test/Transforms/InstCombine/select-extractelement.ll index 10ce220346b..79d0b47f97d 100644 --- a/llvm/test/Transforms/InstCombine/select-extractelement.ll +++ b/llvm/test/Transforms/InstCombine/select-extractelement.ll @@ -5,9 +5,8 @@ declare void @v4float_user(<4 x float>) #0 define float @extract_one_select(<4 x float> %a, <4 x float> %b, i32 %c) #0 { ; CHECK-LABEL: @extract_one_select( ; CHECK-NEXT: [[CMP:%.*]] = icmp eq i32 %c, 0 -; CHECK-NEXT: [[B_ELT:%.*]] = extractelement <4 x float> %b, i32 2 -; CHECK-NEXT: [[A_ELT:%.*]] = extractelement <4 x float> %a, i32 2 -; CHECK-NEXT: [[EXTRACT:%.*]] = select i1 [[CMP]], float [[B_ELT]], float [[A_ELT]] +; CHECK-NEXT: [[SEL:%.*]] = select i1 [[CMP]], <4 x float> %b, <4 x float> %a +; CHECK-NEXT: [[EXTRACT:%.*]] = extractelement <4 x float> [[SEL]], i32 2 ; CHECK-NEXT: ret float [[EXTRACT]] ; %cmp = icmp ne i32 %c, 0 @@ -64,14 +63,14 @@ define float @extract_one_vselect_user(<4 x float> %a, <4 x float> %b, <4 x i32> ret float %extract } -; Extract from a vector select +; Do not convert the vector select into a scalar select. That would increase +; the instruction count and potentially obfuscate a vector min/max idiom. + define float @extract_one_vselect(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { ; CHECK-LABEL: @extract_one_vselect( ; CHECK-NEXT: [[CMP:%.*]] = icmp eq <4 x i32> %c, zeroinitializer -; CHECK-NEXT: [[CMP_ELT:%.*]] = extractelement <4 x i1> [[CMP]], i32 0 -; CHECK-NEXT: [[B_ELT:%.*]] = extractelement <4 x float> %b, i32 0 -; CHECK-NEXT: [[A_ELT:%.*]] = extractelement <4 x float> %a, i32 0 -; CHECK-NEXT: [[EXTRACT:%.*]] = select i1 [[CMP_ELT]], float [[B_ELT]], float [[A_ELT]] +; CHECK-NEXT: [[SELECT:%.*]] = select <4 x i1> [[CMP]], <4 x float> %b, <4 x float> %a +; CHECK-NEXT: [[EXTRACT:%.*]] = extractelement <4 x float> [[SELECT]], i32 0 ; CHECK-NEXT: ret float [[EXTRACT]] ; %cmp = icmp ne <4 x i32> %c, zeroinitializer @@ -97,36 +96,28 @@ define <2 x float> @extract_two_vselect(<4 x float> %a, <4 x float> %b, <4 x i32 ret <2 x float> %build2 } -; All the vector selects should be decomposed into scalar selects +; The vector selects are not decomposed into scalar selects because that would increase +; the instruction count. Extract+insert is converted to non-lane-crossing shuffles. ; Test multiple extractelements define <4 x float> @simple_vector_select(<4 x float> %a, <4 x float> %b, <4 x i32> %c) #0 { ; CHECK-LABEL: @simple_vector_select( ; CHECK-NEXT: entry: ; CHECK-NEXT: [[TMP0:%.*]] = extractelement <4 x i32> %c, i32 0 ; CHECK-NEXT: [[TOBOOL:%.*]] = icmp eq i32 [[TMP0]], 0 -; CHECK-NEXT: [[B_ELT:%.*]] = extractelement <4 x float> %b, i32 0 -; CHECK-NEXT: [[A_ELT:%.*]] = extractelement <4 x float> %a, i32 0 -; CHECK-NEXT: [[TMP1:%.*]] = select i1 [[TOBOOL]], float [[B_ELT]], float [[A_ELT]] -; CHECK-NEXT: [[TMP2:%.*]] = insertelement <4 x float> undef, float [[TMP1]], i32 0 -; CHECK-NEXT: [[TMP3:%.*]] = extractelement <4 x i32> %c, i32 1 -; CHECK-NEXT: [[TOBOOL1:%.*]] = icmp eq i32 [[TMP3]], 0 -; CHECK-NEXT: [[B_ELT1:%.*]] = extractelement <4 x float> %b, i32 1 -; CHECK-NEXT: [[A_ELT2:%.*]] = extractelement <4 x float> %a, i32 1 -; CHECK-NEXT: [[TMP4:%.*]] = select i1 [[TOBOOL1]], float [[B_ELT1]], float [[A_ELT2]] -; CHECK-NEXT: [[TMP5:%.*]] = insertelement <4 x float> [[TMP2]], float [[TMP4]], i32 1 -; CHECK-NEXT: [[TMP6:%.*]] = extractelement <4 x i32> %c, i32 2 -; CHECK-NEXT: [[TOBOOL6:%.*]] = icmp eq i32 [[TMP6]], 0 -; CHECK-NEXT: [[B_ELT3:%.*]] = extractelement <4 x float> %b, i32 2 -; CHECK-NEXT: [[A_ELT4:%.*]] = extractelement <4 x float> %a, i32 2 -; CHECK-NEXT: [[TMP7:%.*]] = select i1 [[TOBOOL6]], float [[B_ELT3]], float [[A_ELT4]] -; CHECK-NEXT: [[TMP8:%.*]] = insertelement <4 x float> [[TMP5]], float [[TMP7]], i32 2 -; CHECK-NEXT: [[TMP9:%.*]] = extractelement <4 x i32> %c, i32 3 -; CHECK-NEXT: [[TOBOOL11:%.*]] = icmp eq i32 [[TMP9]], 0 -; CHECK-NEXT: [[B_ELT5:%.*]] = extractelement <4 x float> %b, i32 3 -; CHECK-NEXT: [[A_ELT6:%.*]] = extractelement <4 x float> %a, i32 3 -; CHECK-NEXT: [[TMP10:%.*]] = select i1 [[TOBOOL11]], float [[B_ELT5]], float [[A_ELT6]] -; CHECK-NEXT: [[TMP11:%.*]] = insertelement <4 x float> [[TMP8]], float [[TMP10]], i32 3 -; CHECK-NEXT: ret <4 x float> [[TMP11]] +; CHECK-NEXT: [[A_SINK:%.*]] = select i1 [[TOBOOL]], <4 x float> %b, <4 x float> %a +; CHECK-NEXT: [[TMP1:%.*]] = extractelement <4 x i32> %c, i32 1 +; CHECK-NEXT: [[TOBOOL1:%.*]] = icmp eq i32 [[TMP1]], 0 +; CHECK-NEXT: [[A_SINK1:%.*]] = select i1 [[TOBOOL1]], <4 x float> %b, <4 x float> %a +; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <4 x float> [[A_SINK]], <4 x float> [[A_SINK1]], <4 x i32> <i32 0, i32 5, i32 undef, i32 undef> +; CHECK-NEXT: [[TMP3:%.*]] = extractelement <4 x i32> %c, i32 2 +; CHECK-NEXT: [[TOBOOL6:%.*]] = icmp eq i32 [[TMP3]], 0 +; CHECK-NEXT: [[A_SINK2:%.*]] = select i1 [[TOBOOL6]], <4 x float> %b, <4 x float> %a +; CHECK-NEXT: [[TMP4:%.*]] = shufflevector <4 x float> [[TMP2]], <4 x float> [[A_SINK2]], <4 x i32> <i32 0, i32 1, i32 6, i32 undef> +; CHECK-NEXT: [[TMP5:%.*]] = extractelement <4 x i32> %c, i32 3 +; CHECK-NEXT: [[TOBOOL11:%.*]] = icmp eq i32 [[TMP5]], 0 +; CHECK-NEXT: [[A_SINK3:%.*]] = select i1 [[TOBOOL11]], <4 x float> %b, <4 x float> %a +; CHECK-NEXT: [[TMP6:%.*]] = shufflevector <4 x float> [[TMP4]], <4 x float> [[A_SINK3]], <4 x i32> <i32 0, i32 1, i32 2, i32 7> +; CHECK-NEXT: ret <4 x float> [[TMP6]] ; entry: %0 = extractelement <4 x i32> %c, i32 0 |
