[llvm] [LV] Linear argument tests for vectorization of function calls (PR #73936)

[Maciej Gabka via llvm-commits]

================
@@ -0,0 +1,253 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --filter "call" --version 2
+; RUN: opt < %s -passes=loop-vectorize -force-vector-interleave=1 -S | FileCheck %s --check-prefixes=NEON
+; 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=1 -S -prefer-predicate-over-epilogue=predicate-dont-vectorize | FileCheck %s --check-prefixes=SVE_TF
+
+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_linear(ptr noalias %a, ptr readnone %b, i64 %n) {
+; NEON-LABEL: define void @test_linear
+; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
+; NEON:    [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR0:[0-9]+]]
+;
+; SVE_OR_NEON-LABEL: define void @test_linear
+; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0:[0-9]+]] {
+; SVE_OR_NEON:    [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR1:[0-9]+]]
+;
+; SVE_TF-LABEL: define void @test_linear
+; 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]+]]
+;
+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_linear_with_vector(ptr noalias %a, ptr readnone %b, ptr readonly %c, i64 %n) {
+; NEON-LABEL: define void @test_linear_with_vector
+; 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]+]]
+;
+; SVE_OR_NEON-LABEL: define void @test_linear_with_vector
+; 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:%.*]]) #[[ATTR2:[0-9]+]]
+;
+; SVE_TF-LABEL: define void @test_linear_with_vector
+; 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]+]]
+;
+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_linear_bad_stride(ptr noalias %a, ptr readnone %b, i64 %n) {
+; NEON-LABEL: define void @test_linear_bad_stride
+; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
+; NEON:    [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR2:[0-9]+]]
+;
+; SVE_OR_NEON-LABEL: define void @test_linear_bad_stride
+; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
+; SVE_OR_NEON:    [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3:[0-9]+]]
+;
+; SVE_TF-LABEL: define void @test_linear_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]+]]
+;
+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_linear_wide_stride(ptr noalias %a, ptr readnone %b, i64 %n) {
+; NEON-LABEL: define void @test_linear_wide_stride
+; NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) {
+; NEON:    [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR2]]
+;
+; SVE_OR_NEON-LABEL: define void @test_linear_wide_stride
+; SVE_OR_NEON-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
+; SVE_OR_NEON:    [[DATA:%.*]] = call i64 @foo(ptr [[GEPB:%.*]]) #[[ATTR3]]
+;
+; SVE_TF-LABEL: define void @test_linear_wide_stride
+; SVE_TF-SAME: (ptr noalias [[A:%.*]], ptr readnone [[B:%.*]], i64 [[N:%.*]]) #[[ATTR0]] {
+; SVE_TF:    [[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_linear_mixed_types(ptr noalias %a, ptr readnone %b, ptr readonly %c, i64 %n) {
----------------
mgabka wrote:

test_linear4_linear8

https://github.com/llvm/llvm-project/pull/73936