Skip to content

[SLP][NFC] Add non-power-of-2 buildvector test#206332

Open
harrisonGPU wants to merge 1 commit into
mainfrom
users/harrisonGPU/SLP-non-power-2-test
Open

[SLP][NFC] Add non-power-of-2 buildvector test#206332
harrisonGPU wants to merge 1 commit into
mainfrom
users/harrisonGPU/SLP-non-power-2-test

Conversation

@harrisonGPU

Copy link
Copy Markdown
Member

For this PR: #206259

@llvmorg-github-actions

Copy link
Copy Markdown

@llvm/pr-subscribers-llvm-transforms

Author: Harrison Hao (harrisonGPU)

Changes

For this PR: #206259


Full diff: https://github.com/llvm/llvm-project/pull/206332.diff

1 Files Affected:

  • (added) llvm/test/Transforms/SLPVectorizer/non-power-of-2-buildvector.ll (+112)
diff --git a/llvm/test/Transforms/SLPVectorizer/non-power-of-2-buildvector.ll b/llvm/test/Transforms/SLPVectorizer/non-power-of-2-buildvector.ll
new file mode 100644
index 0000000000000..950a83f0bf4a8
--- /dev/null
+++ b/llvm/test/Transforms/SLPVectorizer/non-power-of-2-buildvector.ll
@@ -0,0 +1,112 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 6
+; RUN: opt -passes=slp-vectorizer -S -slp-vectorize-non-power-of-2 < %s | FileCheck %s
+
+define <15 x half> @func(ptr addrspace(1) %buffer) {
+; CHECK-LABEL: define <15 x half> @func(
+; CHECK-SAME: ptr addrspace(1) [[BUFFER:%.*]]) {
+; CHECK-NEXT:  [[_ENTRY:.*:]]
+; CHECK-NEXT:    [[DATA:%.*]] = load <15 x half>, ptr addrspace(1) [[BUFFER]], align 32
+; CHECK-NEXT:    [[TMP0:%.*]] = shufflevector <15 x half> [[DATA]], <15 x half> poison, <8 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7>
+; CHECK-NEXT:    [[TMP1:%.*]] = fcmp olt <8 x half> [[TMP0]], zeroinitializer
+; CHECK-NEXT:    [[TMP19:%.*]] = fmul reassoc nnan nsz arcp contract afn <8 x half> [[TMP0]], splat (half 1.000210e-02)
+; CHECK-NEXT:    [[TMP3:%.*]] = select <8 x i1> [[TMP1]], <8 x half> [[TMP19]], <8 x half> [[TMP0]]
+; CHECK-NEXT:    [[TMP4:%.*]] = shufflevector <8 x half> [[TMP3]], <8 x half> poison, <15 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOT14_VEC_INSERT18211:%.*]] = shufflevector <15 x half> [[DATA]], <15 x half> [[TMP4]], <15 x i32> <i32 15, i32 16, i32 17, i32 18, i32 19, i32 20, i32 21, i32 22, i32 8, i32 9, i32 10, i32 11, i32 12, i32 13, i32 14>
+; CHECK-NEXT:    [[TMP5:%.*]] = extractelement <15 x half> [[DATA]], i64 8
+; CHECK-NEXT:    [[TMP6:%.*]] = fcmp olt half [[TMP5]], 0.000000e+00
+; CHECK-NEXT:    [[TMP7:%.*]] = fmul reassoc nnan nsz arcp contract afn half [[TMP5]], 1.000210e-02
+; CHECK-NEXT:    [[TMP8:%.*]] = select i1 [[TMP6]], half [[TMP7]], half [[TMP5]]
+; CHECK-NEXT:    [[DOT16_VEC_INSERT1823:%.*]] = insertelement <15 x half> [[DOT14_VEC_INSERT18211]], half [[TMP8]], i64 8
+; CHECK-NEXT:    [[TMP9:%.*]] = shufflevector <15 x half> [[DATA]], <15 x half> poison, <2 x i32> <i32 9, i32 10>
+; CHECK-NEXT:    [[TMP10:%.*]] = fcmp olt <2 x half> [[TMP9]], zeroinitializer
+; CHECK-NEXT:    [[TMP11:%.*]] = fmul reassoc nnan nsz arcp contract afn <2 x half> [[TMP9]], splat (half 1.000210e-02)
+; CHECK-NEXT:    [[TMP12:%.*]] = select <2 x i1> [[TMP10]], <2 x half> [[TMP11]], <2 x half> [[TMP9]]
+; CHECK-NEXT:    [[TMP13:%.*]] = shufflevector <2 x half> [[TMP12]], <2 x half> poison, <15 x i32> <i32 0, i32 1, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[DOT20_VEC_INSERT18273:%.*]] = shufflevector <15 x half> [[DOT16_VEC_INSERT1823]], <15 x half> [[TMP13]], <15 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 15, i32 16, i32 11, i32 12, i32 13, i32 14>
+; CHECK-NEXT:    [[TMP14:%.*]] = shufflevector <15 x half> [[DATA]], <15 x half> poison, <4 x i32> <i32 11, i32 12, i32 13, i32 14>
+; CHECK-NEXT:    [[TMP15:%.*]] = fcmp olt <4 x half> [[TMP14]], zeroinitializer
+; CHECK-NEXT:    [[TMP16:%.*]] = fmul reassoc nnan nsz arcp contract afn <4 x half> [[TMP14]], splat (half 1.000210e-02)
+; CHECK-NEXT:    [[TMP17:%.*]] = select <4 x i1> [[TMP15]], <4 x half> [[TMP16]], <4 x half> [[TMP14]]
+; CHECK-NEXT:    [[TMP18:%.*]] = shufflevector <4 x half> [[TMP17]], <4 x half> poison, <15 x i32> <i32 0, i32 1, i32 2, i32 3, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison, i32 poison>
+; CHECK-NEXT:    [[TMP2:%.*]] = shufflevector <15 x half> [[DOT20_VEC_INSERT18273]], <15 x half> [[TMP18]], <15 x i32> <i32 0, i32 1, i32 2, i32 3, i32 4, i32 5, i32 6, i32 7, i32 8, i32 9, i32 10, i32 15, i32 16, i32 17, i32 18>
+; CHECK-NEXT:    ret <15 x half> [[TMP2]]
+;
+.entry:
+  %data = load <15 x half>, ptr addrspace(1) %buffer, align 32
+  %0 = extractelement <15 x half> %data, i64 0
+  %1 = fcmp olt half %0, 0.000000e+00
+  %2 = fmul reassoc nnan nsz arcp contract afn half %0, 1.000210e-02
+  %3 = select i1 %1, half %2, half %0
+  %.0.vec.insert1807 = insertelement <15 x half> %data, half %3, i64 0
+  %4 = extractelement <15 x half> %data, i64 1
+  %5 = fcmp olt half %4, 0.000000e+00
+  %6 = fmul reassoc nnan nsz arcp contract afn half %4, 1.000210e-02
+  %7 = select i1 %5, half %6, half %4
+  %.2.vec.insert1809 = insertelement <15 x half> %.0.vec.insert1807, half %7, i64 1
+  %8 = extractelement <15 x half> %data, i64 2
+  %9 = fcmp olt half %8, 0.000000e+00
+  %10 = fmul reassoc nnan nsz arcp contract afn half %8, 1.000210e-02
+  %11 = select i1 %9, half %10, half %8
+  %.4.vec.insert1811 = insertelement <15 x half> %.2.vec.insert1809, half %11, i64 2
+  %12 = extractelement <15 x half> %data, i64 3
+  %13 = fcmp olt half %12, 0.000000e+00
+  %14 = fmul reassoc nnan nsz arcp contract afn half %12, 1.000210e-02
+  %15 = select i1 %13, half %14, half %12
+  %.6.vec.insert1813 = insertelement <15 x half> %.4.vec.insert1811, half %15, i64 3
+  %16 = extractelement <15 x half> %data, i64 4
+  %17 = fcmp olt half %16, 0.000000e+00
+  %18 = fmul reassoc nnan nsz arcp contract afn half %16, 1.000210e-02
+  %19 = select i1 %17, half %18, half %16
+  %.8.vec.insert1815 = insertelement <15 x half> %.6.vec.insert1813, half %19, i64 4
+  %20 = extractelement <15 x half> %data, i64 5
+  %21 = fcmp olt half %20, 0.000000e+00
+  %22 = fmul reassoc nnan nsz arcp contract afn half %20, 1.000210e-02
+  %23 = select i1 %21, half %22, half %20
+  %.10.vec.insert1817 = insertelement <15 x half> %.8.vec.insert1815, half %23, i64 5
+  %24 = extractelement <15 x half> %data, i64 6
+  %25 = fcmp olt half %24, 0.000000e+00
+  %26 = fmul reassoc nnan nsz arcp contract afn half %24, 1.000210e-02
+  %27 = select i1 %25, half %26, half %24
+  %.12.vec.insert1819 = insertelement <15 x half> %.10.vec.insert1817, half %27, i64 6
+  %28 = extractelement <15 x half> %data, i64 7
+  %29 = fcmp olt half %28, 0.000000e+00
+  %30 = fmul reassoc nnan nsz arcp contract afn half %28, 1.000210e-02
+  %31 = select i1 %29, half %30, half %28
+  %.14.vec.insert1821 = insertelement <15 x half> %.12.vec.insert1819, half %31, i64 7
+  %32 = extractelement <15 x half> %data, i64 8
+  %33 = fcmp olt half %32, 0.000000e+00
+  %34 = fmul reassoc nnan nsz arcp contract afn half %32, 1.000210e-02
+  %35 = select i1 %33, half %34, half %32
+  %.16.vec.insert1823 = insertelement <15 x half> %.14.vec.insert1821, half %35, i64 8
+  %36 = extractelement <15 x half> %data, i64 9
+  %37 = fcmp olt half %36, 0.000000e+00
+  %38 = fmul reassoc nnan nsz arcp contract afn half %36, 1.000210e-02
+  %39 = select i1 %37, half %38, half %36
+  %.18.vec.insert1825 = insertelement <15 x half> %.16.vec.insert1823, half %39, i64 9
+  %40 = extractelement <15 x half> %data, i64 10
+  %41 = fcmp olt half %40, 0.000000e+00
+  %42 = fmul reassoc nnan nsz arcp contract afn half %40, 1.000210e-02
+  %43 = select i1 %41, half %42, half %40
+  %.20.vec.insert1827 = insertelement <15 x half> %.18.vec.insert1825, half %43, i64 10
+  %44 = extractelement <15 x half> %data, i64 11
+  %45 = fcmp olt half %44, 0.000000e+00
+  %46 = fmul reassoc nnan nsz arcp contract afn half %44, 1.000210e-02
+  %47 = select i1 %45, half %46, half %44
+  %.22.vec.insert1829 = insertelement <15 x half> %.20.vec.insert1827, half %47, i64 11
+  %48 = extractelement <15 x half> %data, i64 12
+  %49 = fcmp olt half %48, 0.000000e+00
+  %50 = fmul reassoc nnan nsz arcp contract afn half %48, 1.000210e-02
+  %51 = select i1 %49, half %50, half %48
+  %.24.vec.insert1831 = insertelement <15 x half> %.22.vec.insert1829, half %51, i64 12
+  %52 = extractelement <15 x half> %data, i64 13
+  %53 = fcmp olt half %52, 0.000000e+00
+  %54 = fmul reassoc nnan nsz arcp contract afn half %52, 1.000210e-02
+  %55 = select i1 %53, half %54, half %52
+  %.26.vec.insert1833 = insertelement <15 x half> %.24.vec.insert1831, half %55, i64 13
+  %56 = extractelement <15 x half> %data, i64 14
+  %57 = fcmp olt half %56, 0.000000e+00
+  %58 = fmul reassoc nnan nsz arcp contract afn half %56, 1.000210e-02
+  %59 = select i1 %57, half %58, half %56
+  %.28.vec.insert1835 = insertelement <15 x half> %.26.vec.insert1833, half %59, i64 14
+  ret <15 x half> %.28.vec.insert1835
+}

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant