[clang] ca603d2 - [AArch64] Regenerate neon-vcmla.c test. NFC
David Green via cfe-commits
cfe-commits at lists.llvm.org
Mon Jan 6 08:26:47 PST 2025
Author: David Green
Date: 2025-01-06T16:26:41Z
New Revision: ca603d2536f039194141bf3a01e9ee7f60e37406
URL: https://github.com/llvm/llvm-project/commit/ca603d2536f039194141bf3a01e9ee7f60e37406
DIFF: https://github.com/llvm/llvm-project/commit/ca603d2536f039194141bf3a01e9ee7f60e37406.diff
LOG: [AArch64] Regenerate neon-vcmla.c test. NFC
This removes -O1 from the opt pipeline, using just mem2reg,instsimplify
instead. The target is changed so that the auto update script will apply.
Added:
Modified:
clang/test/CodeGen/AArch64/neon-vcmla.c
Removed:
################################################################################
diff --git a/clang/test/CodeGen/AArch64/neon-vcmla.c b/clang/test/CodeGen/AArch64/neon-vcmla.c
index 02171527cc6a32..d860411fe45c50 100644
--- a/clang/test/CodeGen/AArch64/neon-vcmla.c
+++ b/clang/test/CodeGen/AArch64/neon-vcmla.c
@@ -1,444 +1,913 @@
-// RUN: %clang_cc1 -triple arm64-apple-ios -target-feature +neon \
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+// RUN: %clang_cc1 -triple arm64 -target-feature +neon \
// RUN: -target-feature +v8.3a \
// RUN: -target-feature +fullfp16 \
-// RUN: -disable-O0-optnone -emit-llvm -o - %s | opt -S -O1 | FileCheck %s
+// RUN: -disable-O0-optnone -emit-llvm -o - %s | opt -S -passes="mem2reg,instsimplify" | FileCheck %s
// REQUIRES: aarch64-registered-target
#include <arm_neon.h>
-// CHECK-LABEL: @test_vcmla_f16(
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot0.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> %rhs)
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot0.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[RHS]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_F163_I]]
+//
float16x4_t test_vcmla_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot0.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot0.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[RHS]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_F323_I]]
+//
float32x2_t test_vcmla_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_f16(
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot0.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> %rhs)
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot0.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[RHS]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_F163_I]]
+//
float16x8_t test_vcmlaq_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_f32(
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot0.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> %rhs)
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot0.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[RHS]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_F323_I]]
+//
float32x4_t test_vcmlaq_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_f64(
-// CHECK: [[RES:%.*]] = tail call <2 x double> @llvm.aarch64.neon.vcmla.rot0.v2f64(<2 x double> %acc, <2 x double> %lhs, <2 x double> %rhs)
-// CHECK: ret <2 x double> [[RES]]
+// CHECK-LABEL: define dso_local <2 x double> @test_vcmlaq_f64(
+// CHECK-SAME: <2 x double> noundef [[ACC:%.*]], <2 x double> noundef [[LHS:%.*]], <2 x double> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_F643_I:%.*]] = call <2 x double> @llvm.aarch64.neon.vcmla.rot0.v2f64(<2 x double> [[ACC]], <2 x double> [[LHS]], <2 x double> [[RHS]])
+// CHECK-NEXT: ret <2 x double> [[VCMLAQ_F643_I]]
+//
float64x2_t test_vcmlaq_f64(float64x2_t acc, float64x2_t lhs, float64x2_t rhs) {
return vcmlaq_f64(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_rot90_f16(
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot90.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> %rhs)
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot90_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_ROT90_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot90.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[RHS]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT90_F163_I]]
+//
float16x4_t test_vcmla_rot90_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_rot90_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_rot90_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot90.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot90_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_ROT90_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot90.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[RHS]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT90_F323_I]]
+//
float32x2_t test_vcmla_rot90_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_rot90_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot90_f16(
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot90.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> %rhs)
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot90_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT90_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot90.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[RHS]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT90_F163_I]]
+//
float16x8_t test_vcmlaq_rot90_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_rot90_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot90_f32(
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot90.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> %rhs)
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot90_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT90_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot90.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[RHS]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT90_F323_I]]
+//
float32x4_t test_vcmlaq_rot90_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_rot90_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot90_f64(
-// CHECK: [[RES:%.*]] = tail call <2 x double> @llvm.aarch64.neon.vcmla.rot90.v2f64(<2 x double> %acc, <2 x double> %lhs, <2 x double> %rhs)
-// CHECK: ret <2 x double> [[RES]]
+// CHECK-LABEL: define dso_local <2 x double> @test_vcmlaq_rot90_f64(
+// CHECK-SAME: <2 x double> noundef [[ACC:%.*]], <2 x double> noundef [[LHS:%.*]], <2 x double> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT90_F643_I:%.*]] = call <2 x double> @llvm.aarch64.neon.vcmla.rot90.v2f64(<2 x double> [[ACC]], <2 x double> [[LHS]], <2 x double> [[RHS]])
+// CHECK-NEXT: ret <2 x double> [[VCMLAQ_ROT90_F643_I]]
+//
float64x2_t test_vcmlaq_rot90_f64(float64x2_t acc, float64x2_t lhs, float64x2_t rhs) {
return vcmlaq_rot90_f64(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_rot180_f16(
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot180.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> %rhs)
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot180_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_ROT180_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot180.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[RHS]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT180_F163_I]]
+//
float16x4_t test_vcmla_rot180_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_rot180_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_rot180_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot180.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot180_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_ROT180_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot180.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[RHS]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT180_F323_I]]
+//
float32x2_t test_vcmla_rot180_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_rot180_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot180_f16(
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot180.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> %rhs)
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot180_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT180_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot180.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[RHS]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT180_F163_I]]
+//
float16x8_t test_vcmlaq_rot180_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_rot180_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot180_f32(
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot180.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> %rhs)
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot180_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT180_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot180.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[RHS]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT180_F323_I]]
+//
float32x4_t test_vcmlaq_rot180_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_rot180_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot180_f64(
-// CHECK: [[RES:%.*]] = tail call <2 x double> @llvm.aarch64.neon.vcmla.rot180.v2f64(<2 x double> %acc, <2 x double> %lhs, <2 x double> %rhs)
-// CHECK: ret <2 x double> [[RES]]
+// CHECK-LABEL: define dso_local <2 x double> @test_vcmlaq_rot180_f64(
+// CHECK-SAME: <2 x double> noundef [[ACC:%.*]], <2 x double> noundef [[LHS:%.*]], <2 x double> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT180_F643_I:%.*]] = call <2 x double> @llvm.aarch64.neon.vcmla.rot180.v2f64(<2 x double> [[ACC]], <2 x double> [[LHS]], <2 x double> [[RHS]])
+// CHECK-NEXT: ret <2 x double> [[VCMLAQ_ROT180_F643_I]]
+//
float64x2_t test_vcmlaq_rot180_f64(float64x2_t acc, float64x2_t lhs, float64x2_t rhs) {
return vcmlaq_rot180_f64(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_rot270_f16(
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot270.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> %rhs)
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot270_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_ROT270_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot270.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[RHS]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT270_F163_I]]
+//
float16x4_t test_vcmla_rot270_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_rot270_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_rot270_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot270.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot270_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLA_ROT270_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot270.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[RHS]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT270_F323_I]]
+//
float32x2_t test_vcmla_rot270_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_rot270_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot270_f16(
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot270.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> %rhs)
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot270_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT270_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot270.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[RHS]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT270_F163_I]]
+//
float16x8_t test_vcmlaq_rot270_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_rot270_f16(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot270_f32(
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot270.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> %rhs)
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot270_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT270_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot270.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[RHS]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT270_F323_I]]
+//
float32x4_t test_vcmlaq_rot270_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_rot270_f32(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmlaq_rot270_f64(
-// CHECK: [[RES:%.*]] = tail call <2 x double> @llvm.aarch64.neon.vcmla.rot270.v2f64(<2 x double> %acc, <2 x double> %lhs, <2 x double> %rhs)
-// CHECK: ret <2 x double> [[RES]]
+// CHECK-LABEL: define dso_local <2 x double> @test_vcmlaq_rot270_f64(
+// CHECK-SAME: <2 x double> noundef [[ACC:%.*]], <2 x double> noundef [[LHS:%.*]], <2 x double> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[VCMLAQ_ROT270_F643_I:%.*]] = call <2 x double> @llvm.aarch64.neon.vcmla.rot270.v2f64(<2 x double> [[ACC]], <2 x double> [[LHS]], <2 x double> [[RHS]])
+// CHECK-NEXT: ret <2 x double> [[VCMLAQ_ROT270_F643_I]]
+//
float64x2_t test_vcmlaq_rot270_f64(float64x2_t acc, float64x2_t lhs, float64x2_t rhs) {
return vcmlaq_rot270_f64(acc, lhs, rhs);
}
-// CHECK-LABEL: @test_vcmla_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot0.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_lane_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_150:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_150:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_150]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_150]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_150]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_150]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_150]], align 8
+// CHECK-NEXT: [[VCMLA_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot0.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_F163_I]]
+//
float16x4_t test_vcmla_lane_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_lane_f16(acc, lhs, rhs, 1);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <2 x i32> <i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <2 x i32> [[DUP]] to <4 x half>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot0.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP_FLT]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_laneq_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_154:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_154:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_154]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_154]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_154]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_154]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_154]], align 8
+// CHECK-NEXT: [[VCMLA_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot0.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_F163_I]]
+//
float16x4_t test_vcmla_laneq_f16(float16x4_t acc, float16x4_t lhs, float16x8_t rhs) {
return vcmla_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmlaq_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot0.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_lane_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_152:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_152:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_152]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_152]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_152]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[__REINT_152]], align 8
+// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGET_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr [[__REINT_152]], align 8
+// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGET_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_152]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_152]], align 16
+// CHECK-NEXT: [[VCMLAQ_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot0.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_F163_I]]
+//
float16x8_t test_vcmlaq_lane_f16(float16x8_t acc, float16x8_t lhs, float16x4_t rhs) {
return vcmlaq_lane_f16(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <4 x i32> [[DUP]] to <8 x half>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot0.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP_FLT]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_laneq_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_156:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_156:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_156]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_156]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_156]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr [[__REINT_156]], align 16
+// CHECK-NEXT: [[VGETQ_LANE8:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGETQ_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr [[__REINT_156]], align 16
+// CHECK-NEXT: [[VGETQ_LANE13:%.*]] = extractelement <4 x i32> [[TMP3]], i32 3
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGETQ_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_156]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_156]], align 16
+// CHECK-NEXT: [[VCMLAQ_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot0.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_F163_I]]
+//
float16x8_t test_vcmlaq_laneq_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmla_lane_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot0.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_lane_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_182:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_182:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_182]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_182]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_182]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_182]], align 8
+// CHECK-NEXT: [[VCMLA_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot0.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_F323_I]]
+//
float32x2_t test_vcmla_lane_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_lane_f32(acc, lhs, rhs, 0);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_laneq_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <4 x float> %rhs to <2 x i64>
-// CHECK: [[DUP:%.*]] = shufflevector <2 x i64> [[CPLX]], <2 x i64> poison, <1 x i32> <i32 1>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <1 x i64> [[DUP]] to <2 x float>
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot0.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> [[DUP_FLT]])
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_laneq_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_186:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_186:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_186]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_186]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_186]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_186]], align 8
+// CHECK-NEXT: [[VCMLA_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot0.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_F323_I]]
+//
float32x2_t test_vcmla_laneq_f32(float32x2_t acc, float32x2_t lhs, float32x4_t rhs) {
return vcmla_laneq_f32(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_lane_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <2 x float> %rhs to i64
-// CHECK: [[CPLX_VEC:%.*]] = insertelement <2 x i64> poison, i64 [[CPLX]], i64 0
-// CHECK: [[CPLX2:%.*]] = bitcast <2 x i64> [[CPLX_VEC]] to <4 x float>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> [[CPLX2]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot0.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_lane_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_184:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_184:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_184]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_184]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <1 x i64>, ptr [[__REINT_184]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <1 x i64> [[TMP1]], i32 0
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_184]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_184]], align 16
+// CHECK-NEXT: [[VCMLAQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot0.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_F323_I]]
+//
float32x4_t test_vcmlaq_lane_f32(float32x4_t acc, float32x4_t lhs, float32x2_t rhs) {
return vcmlaq_lane_f32(acc, lhs, rhs, 0);
}
-// CHECK-LABEL: @test_vcmlaq_laneq_f32(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> %rhs, <4 x float> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot0.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_laneq_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_188:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_188:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_188]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_188]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr [[__REINT_188]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <2 x i64> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_188]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_188]], align 16
+// CHECK-NEXT: [[VCMLAQ_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot0.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_F323_I]]
+//
float32x4_t test_vcmlaq_laneq_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_laneq_f32(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmla_rot90_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot90.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot90_lane_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_174:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_174:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_174]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_174]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_174]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_174]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_174]], align 8
+// CHECK-NEXT: [[VCMLA_ROT90_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot90.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT90_F163_I]]
+//
float16x4_t test_vcmla_rot90_lane_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_rot90_lane_f16(acc, lhs, rhs, 1);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_rot90_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <2 x i32> <i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <2 x i32> [[DUP]] to <4 x half>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot90.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP_FLT]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot90_laneq_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_178:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_178:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_178]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_178]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_178]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_178]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_178]], align 8
+// CHECK-NEXT: [[VCMLA_ROT90_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot90.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT90_F163_I]]
+//
float16x4_t test_vcmla_rot90_laneq_f16(float16x4_t acc, float16x4_t lhs, float16x8_t rhs) {
return vcmla_rot90_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmlaq_rot90_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot90.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot90_lane_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_176:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_176:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_176]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_176]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_176]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[__REINT_176]], align 8
+// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGET_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr [[__REINT_176]], align 8
+// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGET_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_176]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_176]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT90_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot90.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT90_F163_I]]
+//
float16x8_t test_vcmlaq_rot90_lane_f16(float16x8_t acc, float16x8_t lhs, float16x4_t rhs) {
return vcmlaq_rot90_lane_f16(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_rot90_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <4 x i32> [[DUP]] to <8 x half>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot90.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP_FLT]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot90_laneq_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_180:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_180:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_180]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_180]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_180]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr [[__REINT_180]], align 16
+// CHECK-NEXT: [[VGETQ_LANE8:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGETQ_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr [[__REINT_180]], align 16
+// CHECK-NEXT: [[VGETQ_LANE13:%.*]] = extractelement <4 x i32> [[TMP3]], i32 3
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGETQ_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_180]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_180]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT90_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot90.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT90_F163_I]]
+//
float16x8_t test_vcmlaq_rot90_laneq_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_rot90_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmla_rot90_lane_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot90.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot90_lane_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_206:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_206:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_206]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_206]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_206]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_206]], align 8
+// CHECK-NEXT: [[VCMLA_ROT90_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot90.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT90_F323_I]]
+//
float32x2_t test_vcmla_rot90_lane_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_rot90_lane_f32(acc, lhs, rhs, 0);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_rot90_laneq_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <4 x float> %rhs to <2 x i64>
-// CHECK: [[DUP:%.*]] = shufflevector <2 x i64> [[CPLX]], <2 x i64> poison, <1 x i32> <i32 1>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <1 x i64> [[DUP]] to <2 x float>
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot90.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> [[DUP_FLT]])
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot90_laneq_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_210:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_210:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_210]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_210]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_210]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_210]], align 8
+// CHECK-NEXT: [[VCMLA_ROT90_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot90.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT90_F323_I]]
+//
float32x2_t test_vcmla_rot90_laneq_f32(float32x2_t acc, float32x2_t lhs, float32x4_t rhs) {
return vcmla_rot90_laneq_f32(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_rot90_lane_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <2 x float> %rhs to i64
-// CHECK: [[CPLX_VEC:%.*]] = insertelement <2 x i64> poison, i64 [[CPLX]], i64 0
-// CHECK: [[CPLX2:%.*]] = bitcast <2 x i64> [[CPLX_VEC]] to <4 x float>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> [[CPLX2]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot90.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot90_lane_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_208:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_208:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_208]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_208]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <1 x i64>, ptr [[__REINT_208]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <1 x i64> [[TMP1]], i32 0
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_208]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_208]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT90_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot90.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT90_F323_I]]
+//
float32x4_t test_vcmlaq_rot90_lane_f32(float32x4_t acc, float32x4_t lhs, float32x2_t rhs) {
return vcmlaq_rot90_lane_f32(acc, lhs, rhs, 0);
}
-// CHECK-LABEL: @test_vcmlaq_rot90_laneq_f32(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> %rhs, <4 x float> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot90.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot90_laneq_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_212:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_212:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_212]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_212]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr [[__REINT_212]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <2 x i64> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_212]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_212]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT90_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot90.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT90_F323_I]]
+//
float32x4_t test_vcmlaq_rot90_laneq_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_rot90_laneq_f32(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmla_rot180_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot180.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot180_lane_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_158:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_158:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_158]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_158]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_158]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_158]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_158]], align 8
+// CHECK-NEXT: [[VCMLA_ROT180_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot180.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT180_F163_I]]
+//
float16x4_t test_vcmla_rot180_lane_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_rot180_lane_f16(acc, lhs, rhs, 1);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_rot180_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <2 x i32> <i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <2 x i32> [[DUP]] to <4 x half>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot180.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP_FLT]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot180_laneq_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_162:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_162:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_162]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_162]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_162]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_162]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_162]], align 8
+// CHECK-NEXT: [[VCMLA_ROT180_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot180.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT180_F163_I]]
+//
float16x4_t test_vcmla_rot180_laneq_f16(float16x4_t acc, float16x4_t lhs, float16x8_t rhs) {
return vcmla_rot180_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmlaq_rot180_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot180.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot180_lane_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_160:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_160:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_160]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_160]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_160]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[__REINT_160]], align 8
+// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGET_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr [[__REINT_160]], align 8
+// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGET_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_160]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_160]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT180_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot180.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT180_F163_I]]
+//
float16x8_t test_vcmlaq_rot180_lane_f16(float16x8_t acc, float16x8_t lhs, float16x4_t rhs) {
return vcmlaq_rot180_lane_f16(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_rot180_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <4 x i32> [[DUP]] to <8 x half>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot180.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP_FLT]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot180_laneq_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_164:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_164:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_164]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_164]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_164]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr [[__REINT_164]], align 16
+// CHECK-NEXT: [[VGETQ_LANE8:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGETQ_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr [[__REINT_164]], align 16
+// CHECK-NEXT: [[VGETQ_LANE13:%.*]] = extractelement <4 x i32> [[TMP3]], i32 3
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGETQ_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_164]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_164]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT180_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot180.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT180_F163_I]]
+//
float16x8_t test_vcmlaq_rot180_laneq_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_rot180_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmla_rot180_lane_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot180.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot180_lane_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_190:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_190:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_190]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_190]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_190]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_190]], align 8
+// CHECK-NEXT: [[VCMLA_ROT180_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot180.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT180_F323_I]]
+//
float32x2_t test_vcmla_rot180_lane_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_rot180_lane_f32(acc, lhs, rhs, 0);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_rot180_laneq_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <4 x float> %rhs to <2 x i64>
-// CHECK: [[DUP:%.*]] = shufflevector <2 x i64> [[CPLX]], <2 x i64> poison, <1 x i32> <i32 1>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <1 x i64> [[DUP]] to <2 x float>
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot180.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> [[DUP_FLT]])
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot180_laneq_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_194:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_194:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_194]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_194]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_194]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_194]], align 8
+// CHECK-NEXT: [[VCMLA_ROT180_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot180.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT180_F323_I]]
+//
float32x2_t test_vcmla_rot180_laneq_f32(float32x2_t acc, float32x2_t lhs, float32x4_t rhs) {
return vcmla_rot180_laneq_f32(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_rot180_lane_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <2 x float> %rhs to i64
-// CHECK: [[CPLX_VEC:%.*]] = insertelement <2 x i64> poison, i64 [[CPLX]], i64 0
-// CHECK: [[CPLX2:%.*]] = bitcast <2 x i64> [[CPLX_VEC]] to <4 x float>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> [[CPLX2]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot180.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot180_lane_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_192:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_192:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_192]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_192]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <1 x i64>, ptr [[__REINT_192]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <1 x i64> [[TMP1]], i32 0
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_192]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_192]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT180_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot180.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT180_F323_I]]
+//
float32x4_t test_vcmlaq_rot180_lane_f32(float32x4_t acc, float32x4_t lhs, float32x2_t rhs) {
return vcmlaq_rot180_lane_f32(acc, lhs, rhs, 0);
}
-// CHECK-LABEL: @test_vcmlaq_rot180_laneq_f32(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> %rhs, <4 x float> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot180.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot180_laneq_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_196:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_196:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_196]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_196]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr [[__REINT_196]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <2 x i64> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_196]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_196]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT180_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot180.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT180_F323_I]]
+//
float32x4_t test_vcmlaq_rot180_laneq_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_rot180_laneq_f32(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmla_rot270_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot270.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot270_lane_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_166:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_166:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_166]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_166]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_166]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_166]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_166]], align 8
+// CHECK-NEXT: [[VCMLA_ROT270_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot270.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT270_F163_I]]
+//
float16x4_t test_vcmla_rot270_lane_f16(float16x4_t acc, float16x4_t lhs, float16x4_t rhs) {
return vcmla_rot270_lane_f16(acc, lhs, rhs, 1);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_rot270_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <2 x i32> <i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <2 x i32> [[DUP]] to <4 x half>
-// CHECK: [[RES:%.*]] = tail call <4 x half> @llvm.aarch64.neon.vcmla.rot270.v4f16(<4 x half> %acc, <4 x half> %lhs, <4 x half> [[DUP_FLT]])
-// CHECK: ret <4 x half> [[RES]]
+// CHECK-LABEL: define dso_local <4 x half> @test_vcmla_rot270_laneq_f16(
+// CHECK-SAME: <4 x half> noundef [[ACC:%.*]], <4 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_170:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_170:%.*]] = alloca <2 x i32>, align 8
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_170]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_170]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_170]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i32> [[VECINIT5]], ptr [[__REINT1_170]], align 8
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x half>, ptr [[__REINT1_170]], align 8
+// CHECK-NEXT: [[VCMLA_ROT270_F163_I:%.*]] = call <4 x half> @llvm.aarch64.neon.vcmla.rot270.v4f16(<4 x half> [[ACC]], <4 x half> [[LHS]], <4 x half> [[TMP2]])
+// CHECK-NEXT: ret <4 x half> [[VCMLA_ROT270_F163_I]]
+//
float16x4_t test_vcmla_rot270_laneq_f16(float16x4_t acc, float16x4_t lhs, float16x8_t rhs) {
return vcmla_rot270_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmlaq_rot270_lane_f16(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x half> %rhs, <4 x half> poison, <8 x i32> <i32 2, i32 3, i32 2, i32 3, i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot270.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot270_lane_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <4 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_168:%.*]] = alloca <4 x half>, align 8
+// CHECK-NEXT: [[__REINT1_168:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <4 x half> [[RHS]], ptr [[__REINT_168]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i32>, ptr [[__REINT_168]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <2 x i32> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i32>, ptr [[__REINT_168]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <2 x i32> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <2 x i32>, ptr [[__REINT_168]], align 8
+// CHECK-NEXT: [[VGET_LANE8:%.*]] = extractelement <2 x i32> [[TMP2]], i32 1
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGET_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <2 x i32>, ptr [[__REINT_168]], align 8
+// CHECK-NEXT: [[VGET_LANE13:%.*]] = extractelement <2 x i32> [[TMP3]], i32 1
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGET_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_168]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_168]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT270_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot270.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT270_F163_I]]
+//
float16x8_t test_vcmlaq_rot270_lane_f16(float16x8_t acc, float16x8_t lhs, float16x4_t rhs) {
return vcmlaq_rot270_lane_f16(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_rot270_laneq_f16(
-// CHECK: [[CPLX:%.*]] = bitcast <8 x half> %rhs to <4 x i32>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x i32> [[CPLX]], <4 x i32> poison, <4 x i32> <i32 3, i32 3, i32 3, i32 3>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <4 x i32> [[DUP]] to <8 x half>
-// CHECK: [[RES:%.*]] = tail call <8 x half> @llvm.aarch64.neon.vcmla.rot270.v8f16(<8 x half> %acc, <8 x half> %lhs, <8 x half> [[DUP_FLT]])
-// CHECK: ret <8 x half> [[RES]]
+// CHECK-LABEL: define dso_local <8 x half> @test_vcmlaq_rot270_laneq_f16(
+// CHECK-SAME: <8 x half> noundef [[ACC:%.*]], <8 x half> noundef [[LHS:%.*]], <8 x half> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_172:%.*]] = alloca <8 x half>, align 16
+// CHECK-NEXT: [[__REINT1_172:%.*]] = alloca <4 x i32>, align 16
+// CHECK-NEXT: store <8 x half> [[RHS]], ptr [[__REINT_172]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <4 x i32>, ptr [[__REINT_172]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <4 x i32> [[TMP0]], i32 3
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <4 x i32> poison, i32 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <4 x i32>, ptr [[__REINT_172]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <4 x i32> [[TMP1]], i32 3
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <4 x i32> [[VECINIT]], i32 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x i32>, ptr [[__REINT_172]], align 16
+// CHECK-NEXT: [[VGETQ_LANE8:%.*]] = extractelement <4 x i32> [[TMP2]], i32 3
+// CHECK-NEXT: [[VECINIT10:%.*]] = insertelement <4 x i32> [[VECINIT5]], i32 [[VGETQ_LANE8]], i32 2
+// CHECK-NEXT: [[TMP3:%.*]] = load <4 x i32>, ptr [[__REINT_172]], align 16
+// CHECK-NEXT: [[VGETQ_LANE13:%.*]] = extractelement <4 x i32> [[TMP3]], i32 3
+// CHECK-NEXT: [[VECINIT15:%.*]] = insertelement <4 x i32> [[VECINIT10]], i32 [[VGETQ_LANE13]], i32 3
+// CHECK-NEXT: store <4 x i32> [[VECINIT15]], ptr [[__REINT1_172]], align 16
+// CHECK-NEXT: [[TMP4:%.*]] = load <8 x half>, ptr [[__REINT1_172]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT270_F163_I:%.*]] = call <8 x half> @llvm.aarch64.neon.vcmla.rot270.v8f16(<8 x half> [[ACC]], <8 x half> [[LHS]], <8 x half> [[TMP4]])
+// CHECK-NEXT: ret <8 x half> [[VCMLAQ_ROT270_F163_I]]
+//
float16x8_t test_vcmlaq_rot270_laneq_f16(float16x8_t acc, float16x8_t lhs, float16x8_t rhs) {
return vcmlaq_rot270_laneq_f16(acc, lhs, rhs, 3);
}
-// CHECK-LABEL: @test_vcmla_rot270_lane_f32(
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot270.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> %rhs)
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot270_lane_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_198:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_198:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_198]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_198]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_198]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_198]], align 8
+// CHECK-NEXT: [[VCMLA_ROT270_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot270.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT270_F323_I]]
+//
float32x2_t test_vcmla_rot270_lane_f32(float32x2_t acc, float32x2_t lhs, float32x2_t rhs) {
return vcmla_rot270_lane_f32(acc, lhs, rhs, 0);
}
// ACLE says this exists, but it won't map to a single instruction if lane > 1.
-// CHECK-LABEL: @test_vcmla_rot270_laneq_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <4 x float> %rhs to <2 x i64>
-// CHECK: [[DUP:%.*]] = shufflevector <2 x i64> [[CPLX]], <2 x i64> poison, <1 x i32> <i32 1>
-// CHECK: [[DUP_FLT:%.*]] = bitcast <1 x i64> [[DUP]] to <2 x float>
-// CHECK: [[RES:%.*]] = tail call <2 x float> @llvm.aarch64.neon.vcmla.rot270.v2f32(<2 x float> %acc, <2 x float> %lhs, <2 x float> [[DUP_FLT]])
-// CHECK: ret <2 x float> [[RES]]
+// CHECK-LABEL: define dso_local <2 x float> @test_vcmla_rot270_laneq_f32(
+// CHECK-SAME: <2 x float> noundef [[ACC:%.*]], <2 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_202:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_202:%.*]] = alloca <1 x i64>, align 8
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_202]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_202]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <1 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: store <1 x i64> [[VECINIT]], ptr [[__REINT1_202]], align 8
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x float>, ptr [[__REINT1_202]], align 8
+// CHECK-NEXT: [[VCMLA_ROT270_F323_I:%.*]] = call <2 x float> @llvm.aarch64.neon.vcmla.rot270.v2f32(<2 x float> [[ACC]], <2 x float> [[LHS]], <2 x float> [[TMP1]])
+// CHECK-NEXT: ret <2 x float> [[VCMLA_ROT270_F323_I]]
+//
float32x2_t test_vcmla_rot270_laneq_f32(float32x2_t acc, float32x2_t lhs, float32x4_t rhs) {
return vcmla_rot270_laneq_f32(acc, lhs, rhs, 1);
}
-// CHECK-LABEL: @test_vcmlaq_rot270_lane_f32(
-// CHECK: [[CPLX:%.*]] = bitcast <2 x float> %rhs to i64
-// CHECK: [[CPLX_VEC:%.*]] = insertelement <2 x i64> poison, i64 [[CPLX]], i64 0
-// CHECK: [[CPLX2:%.*]] = bitcast <2 x i64> [[DUP]] to <4 x float>
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> [[CPLX2]], <4 x float> poison, <4 x i32> <i32 0, i32 1, i32 0, i32 1>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot270.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot270_lane_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <2 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_200:%.*]] = alloca <2 x float>, align 8
+// CHECK-NEXT: [[__REINT1_200:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <2 x float> [[RHS]], ptr [[__REINT_200]], align 8
+// CHECK-NEXT: [[TMP0:%.*]] = load <1 x i64>, ptr [[__REINT_200]], align 8
+// CHECK-NEXT: [[VGET_LANE:%.*]] = extractelement <1 x i64> [[TMP0]], i32 0
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGET_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <1 x i64>, ptr [[__REINT_200]], align 8
+// CHECK-NEXT: [[VGET_LANE3:%.*]] = extractelement <1 x i64> [[TMP1]], i32 0
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGET_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_200]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_200]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT270_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot270.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT270_F323_I]]
+//
float32x4_t test_vcmlaq_rot270_lane_f32(float32x4_t acc, float32x4_t lhs, float32x2_t rhs) {
return vcmlaq_rot270_lane_f32(acc, lhs, rhs, 0);
}
-// CHECK-LABEL: @test_vcmlaq_rot270_laneq_f32(
-// CHECK: [[DUP:%.*]] = shufflevector <4 x float> %rhs, <4 x float> poison, <4 x i32> <i32 2, i32 3, i32 2, i32 3>
-// CHECK: [[RES:%.*]] = tail call <4 x float> @llvm.aarch64.neon.vcmla.rot270.v4f32(<4 x float> %acc, <4 x float> %lhs, <4 x float> [[DUP]])
-// CHECK: ret <4 x float> [[RES]]
+// CHECK-LABEL: define dso_local <4 x float> @test_vcmlaq_rot270_laneq_f32(
+// CHECK-SAME: <4 x float> noundef [[ACC:%.*]], <4 x float> noundef [[LHS:%.*]], <4 x float> noundef [[RHS:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: [[__REINT_204:%.*]] = alloca <4 x float>, align 16
+// CHECK-NEXT: [[__REINT1_204:%.*]] = alloca <2 x i64>, align 16
+// CHECK-NEXT: store <4 x float> [[RHS]], ptr [[__REINT_204]], align 16
+// CHECK-NEXT: [[TMP0:%.*]] = load <2 x i64>, ptr [[__REINT_204]], align 16
+// CHECK-NEXT: [[VGETQ_LANE:%.*]] = extractelement <2 x i64> [[TMP0]], i32 1
+// CHECK-NEXT: [[VECINIT:%.*]] = insertelement <2 x i64> poison, i64 [[VGETQ_LANE]], i32 0
+// CHECK-NEXT: [[TMP1:%.*]] = load <2 x i64>, ptr [[__REINT_204]], align 16
+// CHECK-NEXT: [[VGETQ_LANE3:%.*]] = extractelement <2 x i64> [[TMP1]], i32 1
+// CHECK-NEXT: [[VECINIT5:%.*]] = insertelement <2 x i64> [[VECINIT]], i64 [[VGETQ_LANE3]], i32 1
+// CHECK-NEXT: store <2 x i64> [[VECINIT5]], ptr [[__REINT1_204]], align 16
+// CHECK-NEXT: [[TMP2:%.*]] = load <4 x float>, ptr [[__REINT1_204]], align 16
+// CHECK-NEXT: [[VCMLAQ_ROT270_F323_I:%.*]] = call <4 x float> @llvm.aarch64.neon.vcmla.rot270.v4f32(<4 x float> [[ACC]], <4 x float> [[LHS]], <4 x float> [[TMP2]])
+// CHECK-NEXT: ret <4 x float> [[VCMLAQ_ROT270_F323_I]]
+//
float32x4_t test_vcmlaq_rot270_laneq_f32(float32x4_t acc, float32x4_t lhs, float32x4_t rhs) {
return vcmlaq_rot270_laneq_f32(acc, lhs, rhs, 1);
}
More information about the cfe-commits
mailing list