Doomhowl-Interactive/llvm-project
0
0
Fork 0
Code Issues Pull Requests Actions 5 Packages Projects Releases Wiki Activity
Files
main
llvm-project/llvm/test/Transforms/LoopVectorize/AArch64/vector-call-linear-args.ll
David Sherwood 1d9775f684 [LV] Change VPLane::getAsRuntimeExpr to use constant 64-bit indices (#193206) 
The canonical form preferred by instcombine is to use 64-bit values for
the index when it is a constant. We should try to do the same where
possible in the loop vectoriser as this reduces churn in the compiler.

It also makes other work easier, such as removing extra unnecessary
passes on the RUN line in the test directory which I plan to do
afterwards.
2026-04-22 11:33:42 +01:00

407 lines
21 KiB
LLVM
Raw Permalink Blame History

; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter "call.*(foo|bar|baz|quux|goo)|extractelement" --version 2
; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=1 -S | FileCheck %s --check-prefixes=NEON
; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=2 -S | FileCheck %s --check-prefixes=NEON_INTERLEAVE
; RUN: opt < %s -mattr=+sve -passes=loop-vectorize -force-vector-interleave=1 -S | FileCheck %s --check-prefixes=SVE_OR_NEON
; RUN: opt < %s -mattr=+sve -passes=loop-vectorize -force-vector-interleave=2 -S -tail-folding-policy=must-fold-tail | FileCheck %s --check-prefixes=SVE_OR_NEON_INTERLEAVE
; RUN: opt < %s -mattr=+sve -passes=loop-vectorize -force-vector-interleave=1 -S -tail-folding-policy=must-fold-tail | FileCheck %s --check-prefixes=SVE_TF
; RUN: opt < %s -mattr=+sve -passes=loop-vectorize -force-vector-interleave=2 -S -tail-folding-policy=must-fold-tail | FileCheck %s --check-prefixes=SVE_TF_INTERLEAVE
target triple = "aarch64-unknown-linux-gnu"
; A call whose argument can remain a scalar because it's sequential and only the
; starting value is required.
define void @test_linear8(ptr noalias %a, ptr readnone %b, i64 %n) {
; NEON-LABEL: define void @test_linear8
; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
; NEON: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR0:[0-9]+]]
;
; NEON_INTERLEAVE-LABEL: define void @test_linear8
; NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR0:[0-9]+]]
;
; SVE_OR_NEON-LABEL: define void @test_linear8
; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; SVE_OR_NEON: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3:[0-9]+]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_linear8
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; SVE_OR_NEON_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR1:[0-9]+]]
;
; SVE_TF-LABEL: define void @test_linear8
; SVE_TF-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; SVE_TF: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR1:[0-9]+]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_linear8
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
; SVE_TF_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR1:[0-9]+]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%gepb = getelementptr i64, ptr %b, i64 %indvars.iv
%data = call i64 @foo(ptr %gepb) #0
%gepa = getelementptr inbounds i64, ptr %a, i64 %indvars.iv
store i64 %data, ptr %gepa
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
define void @test_vector_linear4(ptr noalias %a, ptr readnone %b, ptr readonly %c, i64 %n) {
; NEON-LABEL: define void @test_vector_linear4
; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) {
; NEON: [[DATA:%.*]] = call i32 @baz(i32 [[INPUT:%.*]], ptr [[GEPB:%.*]]) #[[ATTR1:[0-9]+]]
;
; NEON_INTERLEAVE-LABEL: define void @test_vector_linear4
; NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @baz(i32 [[INPUT:%.*]], ptr [[GEPB:%.*]]) #[[ATTR1:[0-9]+]]
;
; SVE_OR_NEON-LABEL: define void @test_vector_linear4
; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON: [[DATA:%.*]] = call i32 @baz(i32 [[INPUT:%.*]], ptr [[GEPB:%.*]]) #[[ATTR4:[0-9]+]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_vector_linear4
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @baz(i32 [[INPUT:%.*]], ptr [[GEPB:%.*]]) #[[ATTR2:[0-9]+]]
;
; SVE_TF-LABEL: define void @test_vector_linear4
; SVE_TF-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF: [[DATA:%.*]] = call i32 @baz(i32 [[INPUT:%.*]], ptr [[GEPB:%.*]]) #[[ATTR2:[0-9]+]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_vector_linear4
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF_INTERLEAVE: [[DATA:%.*]] = call i32 @baz(i32 [[INPUT:%.*]], ptr [[GEPB:%.*]]) #[[ATTR2:[0-9]+]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%gepc = getelementptr i32, ptr %c, i64 %indvars.iv
%input = load i32, ptr %gepc, align 8
%gepb = getelementptr i32, ptr %b, i64 %indvars.iv
%data = call i32 @baz(i32 %input, ptr %gepb) #1
%gepa = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
store i32 %data, ptr %gepa, align 8
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
define void @test_linear8_bad_stride(ptr noalias %a, ptr readnone %b, i64 %n) {
; NEON-LABEL: define void @test_linear8_bad_stride
; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
; NEON: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR2:[0-9]+]]
;
; NEON_INTERLEAVE-LABEL: define void @test_linear8_bad_stride
; NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR2:[0-9]+]]
;
; SVE_OR_NEON-LABEL: define void @test_linear8_bad_stride
; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR5:[0-9]+]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_linear8_bad_stride
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3:[0-9]+]]
;
; SVE_TF-LABEL: define void @test_linear8_bad_stride
; SVE_TF-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3:[0-9]+]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_linear8_bad_stride
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3:[0-9]+]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%gepb = getelementptr i64, ptr %b, i64 %indvars.iv
%data = call i64 @foo(ptr %gepb) #2
%gepa = getelementptr inbounds i64, ptr %a, i64 %indvars.iv
store i64 %data, ptr %gepa
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
define void @test_linear16_wide_stride(ptr noalias %a, ptr readnone %b, i64 %n) {
; NEON-LABEL: define void @test_linear16_wide_stride
; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
; NEON: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR2]]
;
; NEON_INTERLEAVE-LABEL: define void @test_linear16_wide_stride
; NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR2]]
;
; SVE_OR_NEON-LABEL: define void @test_linear16_wide_stride
; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR5]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_linear16_wide_stride
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3]]
;
; SVE_TF-LABEL: define void @test_linear16_wide_stride
; SVE_TF-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_linear16_wide_stride
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF_INTERLEAVE: [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%double = mul i64 %indvars.iv, 2
%gepb = getelementptr i64, ptr %b, i64 %double
%data = call i64 @foo(ptr %gepb) #2
%gepa = getelementptr inbounds i64, ptr %a, i64 %indvars.iv
store i64 %data, ptr %gepa
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
define void @test_linear4_linear8(ptr noalias %a, ptr readnone %b, ptr readonly %c, i64 %n) {
; NEON-LABEL: define void @test_linear4_linear8
; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) {
; NEON: [[DATA:%.*]] = call i32 @quux(ptr [[GEPC:%.*]], ptr [[GEPB:%.*]]) #[[ATTR3:[0-9]+]]
;
; NEON_INTERLEAVE-LABEL: define void @test_linear4_linear8
; NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @quux(ptr [[GEPC:%.*]], ptr [[GEPB:%.*]]) #[[ATTR3:[0-9]+]]
;
; SVE_OR_NEON-LABEL: define void @test_linear4_linear8
; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON: [[DATA:%.*]] = call i32 @quux(ptr [[GEPC:%.*]], ptr [[GEPB:%.*]]) #[[ATTR6:[0-9]+]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_linear4_linear8
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @quux(ptr [[GEPC:%.*]], ptr [[GEPB:%.*]]) #[[ATTR4:[0-9]+]]
;
; SVE_TF-LABEL: define void @test_linear4_linear8
; SVE_TF-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF: [[DATA:%.*]] = call i32 @quux(ptr [[GEPC:%.*]], ptr [[GEPB:%.*]]) #[[ATTR4:[0-9]+]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_linear4_linear8
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], ptr readonly [[C:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF_INTERLEAVE: [[DATA:%.*]] = call i32 @quux(ptr [[GEPC:%.*]], ptr [[GEPB:%.*]]) #[[ATTR4:[0-9]+]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%gepc = getelementptr i32, ptr %c, i64 %indvars.iv
%gepb = getelementptr i64, ptr %b, i64 %indvars.iv
%data = call i32 @quux(ptr %gepc, ptr %gepb) #3
%gepa = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
store i32 %data, ptr %gepa, align 8
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
define void @test_linear3_non_ptr(ptr noalias %a, i64 %n) {
; NEON-LABEL: define void @test_linear3_non_ptr
; NEON-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) {
; NEON: [[TMP1:%.*]] = extractelement <4 x i32> [[TMP0:%.*]], i64 0
; NEON: [[TMP2:%.*]] = call <4 x i32> @vec_bar_linear3_nomask_neon(i32 [[TMP1]])
; NEON: [[DATA:%.*]] = call i32 @bar(i32 [[TREBLED:%.*]]) #[[ATTR4:[0-9]+]]
;
; NEON_INTERLEAVE-LABEL: define void @test_linear3_non_ptr
; NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: [[TMP2:%.*]] = extractelement <4 x i32> [[TMP0:%.*]], i64 0
; NEON_INTERLEAVE: [[TMP3:%.*]] = call <4 x i32> @vec_bar_linear3_nomask_neon(i32 [[TMP2]])
; NEON_INTERLEAVE: [[TMP4:%.*]] = extractelement <4 x i32> [[TMP1:%.*]], i64 0
; NEON_INTERLEAVE: [[TMP5:%.*]] = call <4 x i32> @vec_bar_linear3_nomask_neon(i32 [[TMP4]])
; NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @bar(i32 [[TREBLED:%.*]]) #[[ATTR4:[0-9]+]]
;
; SVE_OR_NEON-LABEL: define void @test_linear3_non_ptr
; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON: [[TMP7:%.*]] = extractelement <vscale x 4 x i32> [[TMP6:%.*]], i64 0
; SVE_OR_NEON: [[TMP8:%.*]] = call <vscale x 4 x i32> @vec_bar_linear3_nomask_sve(i32 [[TMP7]])
; SVE_OR_NEON: [[DATA:%.*]] = call i32 @bar(i32 [[TREBLED:%.*]]) #[[ATTR7:[0-9]+]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_linear3_non_ptr
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @bar(i32 [[TREBLED:%.*]]) #[[ATTR5:[0-9]+]]
;
; SVE_TF-LABEL: define void @test_linear3_non_ptr
; SVE_TF-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF: [[DATA:%.*]] = call i32 @bar(i32 [[TREBLED:%.*]]) #[[ATTR5:[0-9]+]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_linear3_non_ptr
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF_INTERLEAVE: [[DATA:%.*]] = call i32 @bar(i32 [[TREBLED:%.*]]) #[[ATTR5:[0-9]+]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%little.iv = trunc i64 %indvars.iv to i32
%trebled = mul i32 %little.iv, 3
%data = call i32 @bar(i32 %trebled) #4
%gepa = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
store i32 %data, ptr %gepa
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
define void @test_linearn5_non_ptr_neg_stride(ptr noalias %a, i64 %n) {
; NEON-LABEL: define void @test_linearn5_non_ptr_neg_stride
; NEON-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) {
; NEON: [[TMP1:%.*]] = extractelement <4 x i32> [[TMP0:%.*]], i64 0
; NEON: [[TMP2:%.*]] = call <4 x i32> @vec_bar_linearn5_nomask_neon(i32 [[TMP1]])
; NEON: [[DATA:%.*]] = call i32 @bar(i32 [[NEGSTRIDE:%.*]]) #[[ATTR5:[0-9]+]]
;
; NEON_INTERLEAVE-LABEL: define void @test_linearn5_non_ptr_neg_stride
; NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: [[TMP2:%.*]] = extractelement <4 x i32> [[TMP0:%.*]], i64 0
; NEON_INTERLEAVE: [[TMP3:%.*]] = call <4 x i32> @vec_bar_linearn5_nomask_neon(i32 [[TMP2]])
; NEON_INTERLEAVE: [[TMP4:%.*]] = extractelement <4 x i32> [[TMP1:%.*]], i64 0
; NEON_INTERLEAVE: [[TMP5:%.*]] = call <4 x i32> @vec_bar_linearn5_nomask_neon(i32 [[TMP4]])
; NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @bar(i32 [[NEGSTRIDE:%.*]]) #[[ATTR5:[0-9]+]]
;
; SVE_OR_NEON-LABEL: define void @test_linearn5_non_ptr_neg_stride
; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON: [[TMP7:%.*]] = extractelement <vscale x 4 x i32> [[TMP6:%.*]], i64 0
; SVE_OR_NEON: [[TMP8:%.*]] = call <vscale x 4 x i32> @vec_bar_linearn5_nomask_sve(i32 [[TMP7]])
; SVE_OR_NEON: [[DATA:%.*]] = call i32 @bar(i32 [[NEGSTRIDE:%.*]]) #[[ATTR8:[0-9]+]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_linearn5_non_ptr_neg_stride
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON_INTERLEAVE: [[DATA:%.*]] = call i32 @bar(i32 [[NEGSTRIDE:%.*]]) #[[ATTR6:[0-9]+]]
;
; SVE_TF-LABEL: define void @test_linearn5_non_ptr_neg_stride
; SVE_TF-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF: [[DATA:%.*]] = call i32 @bar(i32 [[NEGSTRIDE:%.*]]) #[[ATTR6:[0-9]+]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_linearn5_non_ptr_neg_stride
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[A:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF_INTERLEAVE: [[DATA:%.*]] = call i32 @bar(i32 [[NEGSTRIDE:%.*]]) #[[ATTR6:[0-9]+]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%little.iv = trunc i64 %indvars.iv to i32
%negstride = mul i32 %little.iv, -5
%data = call i32 @bar(i32 %negstride) #5
%gepa = getelementptr inbounds i32, ptr %a, i64 %indvars.iv
store i32 %data, ptr %gepa
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
define void @test_linear8_return_void(ptr noalias %in, ptr noalias %out, i64 %n) {
; NEON-LABEL: define void @test_linear8_return_void
; NEON-SAME: (ptr noalias [[IN:%.*]], ptr noalias [[OUT:%.*]], i64 [[N:%.*]]) {
; NEON: call void @goo(i64 [[NUM:%.*]], ptr [[GEP_OUT:%.*]]) #[[ATTR6:[0-9]+]]
;
; NEON_INTERLEAVE-LABEL: define void @test_linear8_return_void
; NEON_INTERLEAVE-SAME: (ptr noalias [[IN:%.*]], ptr noalias [[OUT:%.*]], i64 [[N:%.*]]) {
; NEON_INTERLEAVE: call void @goo(i64 [[NUM:%.*]], ptr [[GEP_OUT:%.*]]) #[[ATTR6:[0-9]+]]
;
; SVE_OR_NEON-LABEL: define void @test_linear8_return_void
; SVE_OR_NEON-SAME: (ptr noalias [[IN:%.*]], ptr noalias [[OUT:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON: call void @goo(i64 [[NUM:%.*]], ptr [[GEP_OUT:%.*]]) #[[ATTR9:[0-9]+]]
;
; SVE_OR_NEON_INTERLEAVE-LABEL: define void @test_linear8_return_void
; SVE_OR_NEON_INTERLEAVE-SAME: (ptr noalias [[IN:%.*]], ptr noalias [[OUT:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_OR_NEON_INTERLEAVE: call void @goo(i64 [[NUM:%.*]], ptr [[GEP_OUT:%.*]]) #[[ATTR7:[0-9]+]]
;
; SVE_TF-LABEL: define void @test_linear8_return_void
; SVE_TF-SAME: (ptr noalias [[IN:%.*]], ptr noalias [[OUT:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF: call void @goo(i64 [[NUM:%.*]], ptr [[GEP_OUT:%.*]]) #[[ATTR7:[0-9]+]]
;
; SVE_TF_INTERLEAVE-LABEL: define void @test_linear8_return_void
; SVE_TF_INTERLEAVE-SAME: (ptr noalias [[IN:%.*]], ptr noalias [[OUT:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
; SVE_TF_INTERLEAVE: call void @goo(i64 [[NUM:%.*]], ptr [[GEP_OUT:%.*]]) #[[ATTR7:[0-9]+]]
;
entry:
br label %for.body
for.body:
%indvars.iv = phi i64 [ 0, %entry ], [ %indvars.iv.next, %for.body ]
%gep.in = getelementptr i64, ptr %in, i64 %indvars.iv
%num = load i64, ptr %gep.in, align 8
%gep.out = getelementptr i64, ptr %out, i64 %indvars.iv
call void @goo(i64 %num, ptr %gep.out) #6
%indvars.iv.next = add nuw nsw i64 %indvars.iv, 1
%exitcond = icmp eq i64 %indvars.iv.next, %n
br i1 %exitcond, label %for.cond.cleanup, label %for.body
for.cond.cleanup:
ret void
}
; Note: Vectorizing pointer arguments is currently disabled as LAA cannot detect
; aliasing from output/input pointers.
declare i64 @foo(ptr)
declare i32 @baz(i32, ptr)
declare i32 @quux(ptr, ptr)
declare i32 @bar(i32)
declare void @goo(i64, ptr)
; neon vector variants of foo
declare <2 x i64> @vec_foo_linear8_nomask_neon(ptr)
declare <2 x i64> @vec_foo_linear16_nomask_neon(ptr)
declare <4 x i32> @vec_baz_vector_linear4_nomask_neon(<4 x i32>, ptr)
declare <4 x i32> @vec_quux_linear4_linear8_nomask_neon(ptr, ptr)
declare <4 x i32> @vec_bar_linear3_nomask_neon(i32)
declare <4 x i32> @vec_bar_linearn5_nomask_neon(i32)
declare void @vec_goo_linear8_nomask_neon(<2 x i64>, ptr)
; scalable vector variants of foo
declare <vscale x 2 x i64> @vec_foo_linear8_mask_sve(ptr, <vscale x 2 x i1>)
declare <vscale x 2 x i64> @vec_foo_linear8_nomask_sve(ptr)
declare <vscale x 2 x i64> @vec_foo_linear16_nomask_sve(ptr)
declare <vscale x 4 x i32> @vec_baz_vector_linear4_nomask_sve(<vscale x 4 x i32>, ptr)
declare <vscale x 4 x i32> @vec_quux_linear4_linear8_mask_sve(ptr, ptr, <vscale x 4 x i1>)
declare <vscale x 4 x i32> @vec_bar_linear3_nomask_sve(i32)
declare <vscale x 4 x i32> @vec_bar_linearn5_nomask_sve(i32)
declare void @vec_goo_linear8_nomask_sve(<vscale x 2 x i64>, ptr)
declare void @vec_goo_linear8_mask_sve(<vscale x 2 x i64>, ptr, <vscale x 2 x i1>)
attributes #0 = { nounwind "vector-function-abi-variant"="_ZGVsNxl8_foo(vec_foo_linear8_nomask_sve),_ZGVsMxl8_foo(vec_foo_linear8_mask_sve),_ZGVnN2l8_foo(vec_foo_linear8_nomask_neon)" }
attributes #1 = { nounwind "vector-function-abi-variant"="_ZGVsNxvl4_baz(vec_baz_vector_linear4_nomask_sve),_ZGVnN4vl4_baz(vec_baz_vector_linear4_nomask_neon)" }
attributes #2 = { nounwind "vector-function-abi-variant"="_ZGVsNxl16_foo(vec_foo_linear16_nomask_sve),_ZGVnN2l16_foo(vec_foo_linear16_nomask_neon)" }
attributes #3 = { nounwind "vector-function-abi-variant"="_ZGVsMxl4l8_quux(vec_quux_linear4_linear8_mask_sve),_ZGVnN4l4l8_quux(vec_quux_linear4_linear8_nomask_neon)" }
attributes #4 = { nounwind "vector-function-abi-variant"="_ZGVsNxl3_bar(vec_bar_linear3_nomask_sve),_ZGVnN4l3_bar(vec_bar_linear3_nomask_neon)" }
attributes #5 = { nounwind "vector-function-abi-variant"="_ZGVsNxln5_bar(vec_bar_linearn5_nomask_sve),_ZGVnN4ln5_bar(vec_bar_linearn5_nomask_neon)" }
attributes #6 = { nounwind "vector-function-abi-variant"="_ZGVsNxvl8_goo(vec_goo_linear8_nomask_sve),_ZGVsMxvl8_goo(vec_goo_linear8_mask_sve),_ZGVsN2vl8_goo(vec_goo_linear8_nomask_neon)" }
Reference in New Issue View Git Blame Copy Permalink