[llvm] 0c0bac9 - [InstCombine] Add additional tests for arm intrinsic alignment (NFC)

[Nikita Popov via llvm-commits]

Author: Nikita Popov
Date: 2024-09-02T14:43:49+02:00
New Revision: 0c0bac94c08e73d4c35b454ba02317f2db313f93

URL: https://github.com/llvm/llvm-project/commit/0c0bac94c08e73d4c35b454ba02317f2db313f93
DIFF: https://github.com/llvm/llvm-project/commit/0c0bac94c08e73d4c35b454ba02317f2db313f93.diff

LOG: [InstCombine] Add additional tests for arm intrinsic alignment (NFC)

Added: 
    

Modified: 
    llvm/test/Transforms/InstCombine/ARM/neon-intrinsics.ll

Removed: 
    


################################################################################
diff  --git a/llvm/test/Transforms/InstCombine/ARM/neon-intrinsics.ll b/llvm/test/Transforms/InstCombine/ARM/neon-intrinsics.ll
index 1c5287b7042fb7..68dda24b537471 100644
--- a/llvm/test/Transforms/InstCombine/ARM/neon-intrinsics.ll
+++ b/llvm/test/Transforms/InstCombine/ARM/neon-intrinsics.ll
@@ -1,25 +1,67 @@
+; NOTE: Assertions have been autogenerated by utils/update_test_checks.py UTC_ARGS: --version 5
 ; RUN: opt < %s -passes=instcombine -mtriple=arm -S | FileCheck %s
 
 ; The alignment arguments for NEON load/store intrinsics can be increased
 ; by instcombine.  Check for this.
 
-; CHECK: vld4.v2i32.p0({{.*}}, i32 32)
-; CHECK: vst4.p0.v2i32({{.*}}, i32 16)
-
 @x = common global [8 x i32] zeroinitializer, align 32
 @y = common global [8 x i32] zeroinitializer, align 16
 
-%struct.__neon_int32x2x4_t = type { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> }
-
-define void @test() nounwind ssp {
-  %tmp1 = call %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32.p0(ptr @x, i32 1)
-  %tmp2 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 0
-  %tmp3 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 1
-  %tmp4 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 2
-  %tmp5 = extractvalue %struct.__neon_int32x2x4_t %tmp1, 3
+define void @test() {
+; CHECK-LABEL: define void @test() {
+; CHECK-NEXT:    [[TMP1:%.*]] = call { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } @llvm.arm.neon.vld4.v2i32.p0(ptr nonnull @x, i32 32)
+; CHECK-NEXT:    [[TMP2:%.*]] = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } [[TMP1]], 0
+; CHECK-NEXT:    [[TMP3:%.*]] = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } [[TMP1]], 1
+; CHECK-NEXT:    [[TMP4:%.*]] = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } [[TMP1]], 2
+; CHECK-NEXT:    [[TMP5:%.*]] = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } [[TMP1]], 3
+; CHECK-NEXT:    call void @llvm.arm.neon.vst4.p0.v2i32(ptr nonnull @y, <2 x i32> [[TMP2]], <2 x i32> [[TMP3]], <2 x i32> [[TMP4]], <2 x i32> [[TMP5]], i32 16)
+; CHECK-NEXT:    ret void
+;
+  %tmp1 = call { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } @llvm.arm.neon.vld4.v2i32.p0(ptr @x, i32 1)
+  %tmp2 = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %tmp1, 0
+  %tmp3 = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %tmp1, 1
+  %tmp4 = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %tmp1, 2
+  %tmp5 = extractvalue { <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32> } %tmp1, 3
   call void @llvm.arm.neon.vst4.p0.v2i32(ptr @y, <2 x i32> %tmp2, <2 x i32> %tmp3, <2 x i32> %tmp4, <2 x i32> %tmp5, i32 1)
   ret void
 }
 
-declare %struct.__neon_int32x2x4_t @llvm.arm.neon.vld4.v2i32.p0(ptr, i32) nounwind readonly
-declare void @llvm.arm.neon.vst4.p0.v2i32(ptr, <2 x i32>, <2 x i32>, <2 x i32>, <2 x i32>, i32) nounwind
+define { <4 x i16>, <4 x i16> } @test_vld1x2_no_align(ptr align 16 %a) {
+; CHECK-LABEL: define { <4 x i16>, <4 x i16> } @test_vld1x2_no_align(
+; CHECK-SAME: ptr align 16 [[A:%.*]]) {
+; CHECK-NEXT:    [[TMP:%.*]] = call { <4 x i16>, <4 x i16> } @llvm.arm.neon.vld1x2.v4i16.p0(ptr [[A]])
+; CHECK-NEXT:    ret { <4 x i16>, <4 x i16> } [[TMP]]
+;
+  %tmp = call { <4 x i16>, <4 x i16> } @llvm.arm.neon.vld1x2.v4i16.p0(ptr %a)
+  ret { <4 x i16>, <4 x i16> } %tmp
+}
+
+define { <4 x i16>, <4 x i16> } @test_vld1x2_lower_align(ptr align 16 %a) {
+; CHECK-LABEL: define { <4 x i16>, <4 x i16> } @test_vld1x2_lower_align(
+; CHECK-SAME: ptr align 16 [[A:%.*]]) {
+; CHECK-NEXT:    [[TMP:%.*]] = call { <4 x i16>, <4 x i16> } @llvm.arm.neon.vld1x2.v4i16.p0(ptr align 8 [[A]])
+; CHECK-NEXT:    ret { <4 x i16>, <4 x i16> } [[TMP]]
+;
+  %tmp = call { <4 x i16>, <4 x i16> } @llvm.arm.neon.vld1x2.v4i16.p0(ptr align 8 %a)
+  ret { <4 x i16>, <4 x i16> } %tmp
+}
+
+define { <4 x i16>, <4 x i16> } @test_vld1x2_higher_align(ptr align 8 %a) {
+; CHECK-LABEL: define { <4 x i16>, <4 x i16> } @test_vld1x2_higher_align(
+; CHECK-SAME: ptr align 8 [[A:%.*]]) {
+; CHECK-NEXT:    [[TMP:%.*]] = call { <4 x i16>, <4 x i16> } @llvm.arm.neon.vld1x2.v4i16.p0(ptr align 16 [[A]])
+; CHECK-NEXT:    ret { <4 x i16>, <4 x i16> } [[TMP]]
+;
+  %tmp = call { <4 x i16>, <4 x i16> } @llvm.arm.neon.vld1x2.v4i16.p0(ptr align 16 %a)
+  ret { <4 x i16>, <4 x i16> } %tmp
+}
+
+define void @test_vst1x2_no_align(ptr align 16 %a, <4 x i16> %b0, <4 x i16> %b1) {
+; CHECK-LABEL: define void @test_vst1x2_no_align(
+; CHECK-SAME: ptr align 16 [[A:%.*]], <4 x i16> [[B0:%.*]], <4 x i16> [[B1:%.*]]) {
+; CHECK-NEXT:    call void @llvm.arm.neon.vst1x2.p0.v4i16(ptr [[A]], <4 x i16> [[B0]], <4 x i16> [[B1]])
+; CHECK-NEXT:    ret void
+;
+  call void @llvm.arm.neon.vst1x2.p0.v4i16(ptr %a, <4 x i16> %b0, <4 x i16> %b1)
+  ret void
+}