[llvm] 97fb91e - [SLP][NFC]Add a test with non-profitable alternate vectorized
Alexey Bataev via llvm-commits
llvm-commits at lists.llvm.org
Tue Mar 12 12:59:56 PDT 2024
Author: Alexey Bataev
Date: 2024-03-12T12:56:35-07:00
New Revision: 97fb91ee665660036f8beffd064b44c6fbbf1b73
URL: https://github.com/llvm/llvm-project/commit/97fb91ee665660036f8beffd064b44c6fbbf1b73
DIFF: https://github.com/llvm/llvm-project/commit/97fb91ee665660036f8beffd064b44c6fbbf1b73.diff
LOG: [SLP][NFC]Add a test with non-profitable alternate vectorized
instructions.
Added:
llvm/test/Transforms/SLPVectorizer/alternate-non-profitable.ll
Modified:
Removed:
################################################################################
diff --git a/llvm/test/Transforms/SLPVectorizer/alternate-non-profitable.ll b/llvm/test/Transforms/SLPVectorizer/alternate-non-profitable.ll
new file mode 100644
index 00000000000000..c6e2cf5543e12e
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/alternate-non-profitable.ll
@@ -0,0 +1,190 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 4
+; RUN: opt -S -passes=slp-vectorizer -slp-threshold=-10000 < %s | FileCheck %s
+
+define <2 x float> @test_fdiv(float %a, i1 %cmp) {
+; CHECK-LABEL: define <2 x float> @test_fdiv(
+; CHECK-SAME: float [[A:%.*]], i1 [[CMP:%.*]]) {
+; CHECK-NEXT: [[TMP1:%.*]] = fdiv float [[A]], 3.000000e+00
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x float> poison, float [[TMP1]], i64 1
+; CHECK-NEXT: [[TMP3:%.*]] = select i1 [[CMP]], <2 x float> <float 7.700000e+01, float 9.900000e+01>, <2 x float> [[TMP2]]
+; CHECK-NEXT: ret <2 x float> [[TMP3]]
+;
+ %1 = fdiv float %a, 3.000000e+00
+ %2 = insertelement <2 x float> poison, float %1, i64 1
+ %3 = select i1 %cmp, <2 x float> <float 7.700000e+01, float 9.900000e+01>, <2 x float> %2
+ ret <2 x float> %3
+}
+
+define <2 x float> @test_frem(float %a, i1 %cmp) {
+; CHECK-LABEL: define <2 x float> @test_frem(
+; CHECK-SAME: float [[A:%.*]], i1 [[CMP:%.*]]) {
+; CHECK-NEXT: [[TMP1:%.*]] = frem float [[A]], 3.000000e+00
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x float> poison, float [[TMP1]], i64 1
+; CHECK-NEXT: [[TMP3:%.*]] = select i1 [[CMP]], <2 x float> <float 7.700000e+01, float 9.900000e+01>, <2 x float> [[TMP2]]
+; CHECK-NEXT: ret <2 x float> [[TMP3]]
+;
+ %1 = frem float %a, 3.000000e+00
+ %2 = insertelement <2 x float> poison, float %1, i64 1
+ %3 = select i1 %cmp, <2 x float> <float 7.700000e+01, float 9.900000e+01>, <2 x float> %2
+ ret <2 x float> %3
+}
+
+define <2 x float> @replace_through_casts(i16 %inp) {
+; CHECK-LABEL: define <2 x float> @replace_through_casts(
+; CHECK-SAME: i16 [[INP:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i16 [[INP]], -10
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i16> poison, i16 [[INP]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i16> [[TMP1]], i16 [[ADD]], i32 1
+; CHECK-NEXT: [[TMP3:%.*]] = uitofp <2 x i16> [[TMP2]] to <2 x float>
+; CHECK-NEXT: [[TMP4:%.*]] = sitofp <2 x i16> [[TMP2]] to <2 x float>
+; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x float> [[TMP3]], <2 x float> [[TMP4]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT: ret <2 x float> [[R]]
+;
+ %add = add nsw i16 %inp, -10
+ %1 = uitofp i16 %inp to float
+ %2 = sitofp i16 %add to float
+ %3 = insertelement <2 x float> poison, float %1, i64 0
+ %r = insertelement <2 x float> %3, float %2, i64 1
+ ret <2 x float> %r
+}
+
+define <2 x float> @replace_through_casts_and_binop(i16 %inp) {
+; CHECK-LABEL: define <2 x float> @replace_through_casts_and_binop(
+; CHECK-SAME: i16 [[INP:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i16 [[INP]], -10
+; CHECK-NEXT: [[MUL:%.*]] = mul nsw i16 [[INP]], 5
+; CHECK-NEXT: [[TMP1:%.*]] = uitofp i16 [[MUL]] to float
+; CHECK-NEXT: [[TMP2:%.*]] = fadd float [[TMP1]], 2.000000e+00
+; CHECK-NEXT: [[TMP3:%.*]] = sitofp i16 [[ADD]] to float
+; CHECK-NEXT: [[TMP4:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
+; CHECK-NEXT: [[R:%.*]] = insertelement <2 x float> [[TMP4]], float [[TMP3]], i64 1
+; CHECK-NEXT: ret <2 x float> [[R]]
+;
+ %add = add nsw i16 %inp, -10
+ %mul = mul nsw i16 %inp, 5
+ %1 = uitofp i16 %mul to float
+ %2 = fadd float %1, 2.000000e+00
+ %3 = sitofp i16 %add to float
+ %4 = insertelement <2 x float> poison, float %2, i64 0
+ %r = insertelement <2 x float> %4, float %3, i64 1
+ ret <2 x float> %r
+}
+
+define <2 x float> @replace_through_casts_and_binop_and_unop(i16 %inp) {
+; CHECK-LABEL: define <2 x float> @replace_through_casts_and_binop_and_unop(
+; CHECK-SAME: i16 [[INP:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i16 [[INP]], -10
+; CHECK-NEXT: [[TMP1:%.*]] = sitofp i16 [[ADD]] to float
+; CHECK-NEXT: [[TMP2:%.*]] = fneg float [[TMP1]]
+; CHECK-NEXT: [[TMP3:%.*]] = uitofp i16 [[ADD]] to float
+; CHECK-NEXT: [[TMP4:%.*]] = fadd float [[TMP3]], 2.000000e+00
+; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x float> poison, float [[TMP4]], i64 0
+; CHECK-NEXT: [[R:%.*]] = insertelement <2 x float> [[TMP5]], float [[TMP2]], i64 1
+; CHECK-NEXT: ret <2 x float> [[R]]
+;
+ %add = add nsw i16 %inp, -10
+ %1 = sitofp i16 %add to float
+ %2 = fneg float %1
+ %3 = uitofp i16 %add to float
+ %4 = fadd float %3, 2.000000e+00
+ %5 = insertelement <2 x float> poison, float %4, i64 0
+ %r = insertelement <2 x float> %5, float %2, i64 1
+ ret <2 x float> %r
+}
+
+define <2 x float> @replace_through_casts_through_splat(i16 %inp) {
+; CHECK-LABEL: define <2 x float> @replace_through_casts_through_splat(
+; CHECK-SAME: i16 [[INP:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i16 [[INP]], -10
+; CHECK-NEXT: [[TMP1:%.*]] = uitofp i16 [[ADD]] to float
+; CHECK-NEXT: [[TMP2:%.*]] = fadd float [[TMP1]], 2.000000e+00
+; CHECK-NEXT: [[TMP3:%.*]] = sitofp i16 [[ADD]] to float
+; CHECK-NEXT: [[TMP4:%.*]] = fneg float [[TMP3]]
+; CHECK-NEXT: [[TMP5:%.*]] = insertelement <2 x float> poison, float [[TMP2]], i64 0
+; CHECK-NEXT: [[R:%.*]] = insertelement <2 x float> [[TMP5]], float [[TMP4]], i64 1
+; CHECK-NEXT: ret <2 x float> [[R]]
+;
+ %add = add nsw i16 %inp, -10
+ %1 = uitofp i16 %add to float
+ %2 = fadd float %1, 2.000000e+00
+ %3 = sitofp i16 %add to float
+ %4 = fneg float %3
+ %5 = insertelement <2 x float> poison, float %2, i64 0
+ %r = insertelement <2 x float> %5, float %4, i64 1
+ ret <2 x float> %r
+}
+
+define <2 x i32> @replace_through_int_casts(i16 %inp, <2 x i16> %dead) {
+; CHECK-LABEL: define <2 x i32> @replace_through_int_casts(
+; CHECK-SAME: i16 [[INP:%.*]], <2 x i16> [[DEAD:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add nsw i16 [[INP]], -10
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i16> poison, i16 [[INP]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i16> [[TMP1]], i16 [[ADD]], i32 1
+; CHECK-NEXT: [[TMP3:%.*]] = zext <2 x i16> [[TMP2]] to <2 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = sext <2 x i16> [[TMP2]] to <2 x i32>
+; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> [[TMP4]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT: ret <2 x i32> [[R]]
+;
+ %add = add nsw i16 %inp, -10
+ %1 = zext i16 %inp to i32
+ %2 = sext i16 %add to i32
+ %3 = insertelement <2 x i32> poison, i32 %1, i64 0
+ %r = insertelement <2 x i32> %3, i32 %2, i64 1
+ ret <2 x i32> %r
+}
+
+define <2 x i32> @replace_through_int_casts_ele0_only(i16 %inp, <2 x i16> %dead) {
+; CHECK-LABEL: define <2 x i32> @replace_through_int_casts_ele0_only(
+; CHECK-SAME: i16 [[INP:%.*]], <2 x i16> [[DEAD:%.*]]) {
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i16> poison, i16 [[INP]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = shufflevector <2 x i16> [[TMP1]], <2 x i16> poison, <2 x i32> zeroinitializer
+; CHECK-NEXT: [[TMP3:%.*]] = zext <2 x i16> [[TMP2]] to <2 x i32>
+; CHECK-NEXT: [[TMP4:%.*]] = sext <2 x i16> [[TMP2]] to <2 x i32>
+; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i32> [[TMP3]], <2 x i32> [[TMP4]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT: ret <2 x i32> [[R]]
+;
+ %2 = sext i16 %inp to i32
+ %4 = zext i16 %inp to i32
+ %5 = insertelement <2 x i32> poison, i32 %4, i64 0
+ %r = insertelement <2 x i32> %5, i32 %2, i64 1
+ ret <2 x i32> %r
+}
+
+define <2 x i8> @replace_through_binop_fail_cant_speculate(i8 %inp, <2 x i8> %d, <2 x i8> %any) {
+; CHECK-LABEL: define <2 x i8> @replace_through_binop_fail_cant_speculate(
+; CHECK-SAME: i8 [[INP:%.*]], <2 x i8> [[D:%.*]], <2 x i8> [[ANY:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = add i8 [[INP]], 5
+; CHECK-NEXT: [[V0:%.*]] = insertelement <2 x i8> poison, i8 [[INP]], i64 0
+; CHECK-NEXT: [[V:%.*]] = insertelement <2 x i8> [[V0]], i8 [[ADD]], i64 1
+; CHECK-NEXT: [[DIV0:%.*]] = sdiv <2 x i8> <i8 -128, i8 -128>, [[V]]
+; CHECK-NEXT: [[TMP1:%.*]] = xor i8 [[INP]], 123
+; CHECK-NEXT: [[R:%.*]] = insertelement <2 x i8> [[DIV0]], i8 [[TMP1]], i64 0
+; CHECK-NEXT: ret <2 x i8> [[R]]
+;
+ %add = add i8 %inp, 5
+ %v0 = insertelement <2 x i8> poison, i8 %inp, i64 0
+ %v = insertelement <2 x i8> %v0, i8 %add, i64 1
+ %div0 = sdiv <2 x i8> <i8 -128, i8 -128>, %v
+ %1 = xor i8 %inp, 123
+ %r = insertelement <2 x i8> %div0, i8 %1, i64 0
+ ret <2 x i8> %r
+}
+
+define <2 x i8> @replace_through_binop_preserve_flags(i8 %inp, <2 x i8> %d, <2 x i8> %any) {
+; CHECK-LABEL: define <2 x i8> @replace_through_binop_preserve_flags(
+; CHECK-SAME: i8 [[INP:%.*]], <2 x i8> [[D:%.*]], <2 x i8> [[ANY:%.*]]) {
+; CHECK-NEXT: [[ADD:%.*]] = xor i8 [[INP]], 5
+; CHECK-NEXT: [[TMP1:%.*]] = insertelement <2 x i8> poison, i8 [[INP]], i32 0
+; CHECK-NEXT: [[TMP2:%.*]] = insertelement <2 x i8> [[TMP1]], i8 [[ADD]], i32 1
+; CHECK-NEXT: [[TMP3:%.*]] = xor <2 x i8> [[TMP2]], <i8 123, i8 1>
+; CHECK-NEXT: [[TMP4:%.*]] = add nsw <2 x i8> [[TMP2]], <i8 123, i8 1>
+; CHECK-NEXT: [[R:%.*]] = shufflevector <2 x i8> [[TMP3]], <2 x i8> [[TMP4]], <2 x i32> <i32 0, i32 3>
+; CHECK-NEXT: ret <2 x i8> [[R]]
+;
+ %add = xor i8 %inp, 5
+ %1 = xor i8 %inp, 123
+ %2 = add nsw i8 %add, 1
+ %3 = insertelement <2 x i8> poison, i8 %1, i64 0
+ %r = insertelement <2 x i8> %3, i8 %2, i64 1
+ ret <2 x i8> %r
+}
More information about the llvm-commits
mailing list