161 lines · c
1// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 52// REQUIRES: aarch64-registered-target3// RUN: %clang_cc1 -triple aarch64 -target-feature +sme-f8f32 -target-feature +sme-f8f16 -target-feature +sme-mop4 -target-feature +sme -target-feature +sme2 -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s4// RUN: %clang_cc1 -triple aarch64 -target-feature +sme-f8f32 -target-feature +sme-f8f16 -target-feature +sme-mop4 -target-feature +sme -target-feature +sme2 -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK5// RUN: %clang_cc1 -DSME_OVERLOADED_FORMS -triple aarch64 -target-feature +sme-f8f32 -target-feature +sme-f8f16 -target-feature +sme-mop4 -target-feature +sme -target-feature +sme2 -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s6// RUN: %clang_cc1 -DSME_OVERLOADED_FORMS -triple aarch64 -target-feature +sme-f8f32 -target-feature +sme-f8f16 -target-feature +sme-mop4 -target-feature +sme -target-feature +sme2 -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK7// RUN: %clang_cc1 -triple aarch64 -target-feature +sme-f8f32 -target-feature +sme-f8f16 -target-feature +sme-mop4 -target-feature +sme -target-feature +sme2 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s8 9 10#include <arm_sme.h>11 12#ifdef SME_OVERLOADED_FORMS13#define SME_ACLE_FUNC(A1,A2_UNUSED,A3,A4_UNUSED,A5) A1##A3##A514#else15#define SME_ACLE_FUNC(A1,A2,A3,A4,A5) A1##A2##A3##A4##A516#endif17 18// CHECK-LABEL: define dso_local void @test_svmop4a_1x1_za16_mf8_mf8_fpm(19// CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0:[0-9]+]] {20// CHECK-NEXT: [[ENTRY:.*:]]21// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])22// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.1x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])23// CHECK-NEXT: ret void24//25// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_1x1_za16_mf8_mf8_fpmu13__SVMfloat8_tS_m(26// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0:[0-9]+]] {27// CPP-CHECK-NEXT: [[ENTRY:.*:]]28// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])29// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.1x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])30// CPP-CHECK-NEXT: ret void31//32void test_svmop4a_1x1_za16_mf8_mf8_fpm(svmfloat8_t zn, svmfloat8_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {33 SME_ACLE_FUNC(svmop4a,_1x1,_za16,_mf8_mf8,_fpm)(1, zn, zm, fpmr);34}35 36// CHECK-LABEL: define dso_local void @test_svmop4a_1x2_za16_mf8_mf8_fpm(37// CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {38// CHECK-NEXT: [[ENTRY:.*:]]39// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])40// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.1x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])41// CHECK-NEXT: ret void42//43// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_1x2_za16_mf8_mf8_fpmu13__SVMfloat8_t13svmfloat8x2_tm(44// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {45// CPP-CHECK-NEXT: [[ENTRY:.*:]]46// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])47// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.1x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])48// CPP-CHECK-NEXT: ret void49//50void test_svmop4a_1x2_za16_mf8_mf8_fpm(svmfloat8_t zn, svmfloat8x2_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {51 SME_ACLE_FUNC(svmop4a,_1x2,_za16,_mf8_mf8,_fpm)(1, zn, zm, fpmr);52}53 54// CHECK-LABEL: define dso_local void @test_svmop4a_2x1_za16_mf8_mf8_fpm(55// CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {56// CHECK-NEXT: [[ENTRY:.*:]]57// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])58// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.2x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])59// CHECK-NEXT: ret void60//61// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_2x1_za16_mf8_mf8_fpm13svmfloat8x2_tu13__SVMfloat8_tm(62// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {63// CPP-CHECK-NEXT: [[ENTRY:.*:]]64// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])65// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.2x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])66// CPP-CHECK-NEXT: ret void67//68void test_svmop4a_2x1_za16_mf8_mf8_fpm(svmfloat8x2_t zn, svmfloat8_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {69 SME_ACLE_FUNC(svmop4a,_2x1,_za16,_mf8_mf8,_fpm)(1, zn, zm, fpmr);70}71 72// CHECK-LABEL: define dso_local void @test_svmop4a_2x2_za16_mf8_mf8_fpm(73// CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {74// CHECK-NEXT: [[ENTRY:.*:]]75// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])76// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.2x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])77// CHECK-NEXT: ret void78//79// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_2x2_za16_mf8_mf8_fpm13svmfloat8x2_tS_m(80// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {81// CPP-CHECK-NEXT: [[ENTRY:.*:]]82// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])83// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za16.2x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])84// CPP-CHECK-NEXT: ret void85//86void test_svmop4a_2x2_za16_mf8_mf8_fpm(svmfloat8x2_t zn, svmfloat8x2_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {87 SME_ACLE_FUNC(svmop4a,_2x2,_za16,_mf8_mf8,_fpm)(1, zn, zm, fpmr);88}89 90// CHECK-LABEL: define dso_local void @test_svmop4a_1x1_za32_mf8_mf8_fpm(91// CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {92// CHECK-NEXT: [[ENTRY:.*:]]93// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])94// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.1x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])95// CHECK-NEXT: ret void96//97// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_1x1_za32_mf8_mf8_fpmu13__SVMfloat8_tS_m(98// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {99// CPP-CHECK-NEXT: [[ENTRY:.*:]]100// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])101// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.1x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])102// CPP-CHECK-NEXT: ret void103//104void test_svmop4a_1x1_za32_mf8_mf8_fpm(svmfloat8_t zn, svmfloat8_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {105 SME_ACLE_FUNC(svmop4a,_1x1,_za32,_mf8_mf8,_fpm)(1, zn, zm, fpmr);106}107 108// CHECK-LABEL: define dso_local void @test_svmop4a_1x2_za32_mf8_mf8_fpm(109// CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {110// CHECK-NEXT: [[ENTRY:.*:]]111// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])112// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.1x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])113// CHECK-NEXT: ret void114//115// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_1x2_za32_mf8_mf8_fpmu13__SVMfloat8_t13svmfloat8x2_tm(116// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {117// CPP-CHECK-NEXT: [[ENTRY:.*:]]118// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])119// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.1x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])120// CPP-CHECK-NEXT: ret void121//122void test_svmop4a_1x2_za32_mf8_mf8_fpm(svmfloat8_t zn, svmfloat8x2_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {123 SME_ACLE_FUNC(svmop4a,_1x2,_za32,_mf8_mf8,_fpm)(1, zn, zm, fpmr);124}125 126// CHECK-LABEL: define dso_local void @test_svmop4a_2x1_za32_mf8_mf8_fpm(127// CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {128// CHECK-NEXT: [[ENTRY:.*:]]129// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])130// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.2x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])131// CHECK-NEXT: ret void132//133// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_2x1_za32_mf8_mf8_fpm13svmfloat8x2_tu13__SVMfloat8_tm(134// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {135// CPP-CHECK-NEXT: [[ENTRY:.*:]]136// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])137// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.2x1.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])138// CPP-CHECK-NEXT: ret void139//140void test_svmop4a_2x1_za32_mf8_mf8_fpm(svmfloat8x2_t zn, svmfloat8_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {141 SME_ACLE_FUNC(svmop4a,_2x1,_za32,_mf8_mf8,_fpm)(1, zn, zm, fpmr);142}143 144// CHECK-LABEL: define dso_local void @test_svmop4a_2x2_za32_mf8_mf8_fpm(145// CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {146// CHECK-NEXT: [[ENTRY:.*:]]147// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])148// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.2x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])149// CHECK-NEXT: ret void150//151// CPP-CHECK-LABEL: define dso_local void @_Z33test_svmop4a_2x2_za32_mf8_mf8_fpm13svmfloat8x2_tS_m(152// CPP-CHECK-SAME: <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM_COERCE0:%.*]], <vscale x 16 x i8> [[ZM_COERCE1:%.*]], i64 noundef [[FPMR:%.*]]) #[[ATTR0]] {153// CPP-CHECK-NEXT: [[ENTRY:.*:]]154// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPMR]])155// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmop4a.za32.2x2.nxv16i8(i32 1, <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM_COERCE0]], <vscale x 16 x i8> [[ZM_COERCE1]])156// CPP-CHECK-NEXT: ret void157//158void test_svmop4a_2x2_za32_mf8_mf8_fpm(svmfloat8x2_t zn, svmfloat8x2_t zm, fpm_t fpmr) __arm_streaming __arm_inout("za") {159 SME_ACLE_FUNC(svmop4a,_2x2,_za32,_mf8_mf8,_fpm)(1, zn, zm, fpmr);160}161