sagarkulkarni19 created this revision.
sagarkulkarni19 added reviewers: sdesmalen, david-arm, rsandifo-arm.
sagarkulkarni19 added a project: clang.
Herald added a subscriber: kristof.beyls.
Herald added a project: All.
sagarkulkarni19 requested review of this revision.
Herald added a subscriber: cfe-commits.
This patch adds support for the following SME ACLE intrinsics:
- svaddha_za32[_u32] // Also for s32
- svaddva_za32[_u32] // Also for s32
- svaddha_za64[_u64] // Also for s64
- svaddva_za64[_u64] // Also for s64
Repository:
rG LLVM Github Monorepo
https://reviews.llvm.org/D134680
Files:
clang/include/clang/Basic/TargetBuiltins.h
clang/include/clang/Basic/arm_sve.td
clang/lib/Basic/Targets/AArch64.cpp
clang/lib/Basic/Targets/AArch64.h
clang/lib/CodeGen/CGBuiltin.cpp
clang/lib/CodeGen/CodeGenFunction.h
clang/test/CodeGen/aarch64-sme-intrinsics/acle_sme_add-i32.c
clang/test/CodeGen/aarch64-sme-intrinsics/acle_sme_add-i64.c
clang/utils/TableGen/SveEmitter.cpp
Index: clang/utils/TableGen/SveEmitter.cpp
===================================================================
--- clang/utils/TableGen/SveEmitter.cpp
+++ clang/utils/TableGen/SveEmitter.cpp
@@ -879,6 +879,8 @@
this->SMEAttributes = "arm_streaming_compatible, arm_shared_za, arm_preserves_za";
else if (this->Flags & Emitter.getEnumValueForFlag("IsSMECnt"))
this->SMEAttributes = "arm_streaming_compatible, arm_preserves_za";
+ else if (this->Flags & Emitter.getEnumValueForFlag("IsSMEAdd"))
+ this->SMEAttributes = "arm_streaming, arm_shared_za";
else
llvm_unreachable("Unknown SME instruction");
} else
Index: clang/test/CodeGen/aarch64-sme-intrinsics/acle_sme_add-i64.c
===================================================================
--- /dev/null
+++ clang/test/CodeGen/aarch64-sme-intrinsics/acle_sme_add-i64.c
@@ -0,0 +1,159 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+// REQUIRES: aarch64-registered-target
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -triple aarch64-none-linux-gnu -target-feature +sme-i64 -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -triple aarch64-none-linux-gnu -target-feature +sme-i64 -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -DSME_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme-i64 -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -DSME_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme-i64 -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -triple aarch64-none-linux-gnu -target-feature +sme-i64 -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -o /dev/null %s
+
+#include <arm_sme.h>
+
+#ifdef SME_OVERLOADED_FORMS
+#define SME_ACLE_FUNC(A1,A2_UNUSED,A3) A1##A3
+#else
+#define SME_ACLE_FUNC(A1,A2,A3) A1##A2##A3
+#endif
+
+// CHECK-LABEL: @test_svaddha_za64_u64(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddha_za64_u64u10__SVBool_tu10__SVBool_tu12__SVUint64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za64_u64(svbool_t pn, svbool_t pm, svuint64_t zn) {
+ SME_ACLE_FUNC(svaddha_za64, _u64,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddha_za64_u64_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddha_za64_u64_1u10__SVBool_tu10__SVBool_tu12__SVUint64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za64_u64_1(svbool_t pn, svbool_t pm, svuint64_t zn) {
+ SME_ACLE_FUNC(svaddha_za64, _u64,)(7, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddha_za64_s64(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddha_za64_s64u10__SVBool_tu10__SVBool_tu11__SVInt64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za64_s64(svbool_t pn, svbool_t pm, svint64_t zn) {
+ SME_ACLE_FUNC(svaddha_za64, _s64,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddha_za64_s64_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddha_za64_s64_1u10__SVBool_tu10__SVBool_tu11__SVInt64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za64_s64_1(svbool_t pn, svbool_t pm, svint64_t zn) {
+ SME_ACLE_FUNC(svaddha_za64, _s64,)(7, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za64_u64(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddva_za64_u64u10__SVBool_tu10__SVBool_tu12__SVUint64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za64_u64(svbool_t pn, svbool_t pm, svuint64_t zn) {
+ SME_ACLE_FUNC(svaddva_za64, _u64,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za64_u64_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddva_za64_u64_1u10__SVBool_tu10__SVBool_tu12__SVUint64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za64_u64_1(svbool_t pn, svbool_t pm, svuint64_t zn) {
+ SME_ACLE_FUNC(svaddva_za64, _u64,)(7, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za64_s64(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddva_za64_s64u10__SVBool_tu10__SVBool_tu11__SVInt64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 0, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za64_s64(svbool_t pn, svbool_t pm, svint64_t zn) {
+ SME_ACLE_FUNC(svaddva_za64, _s64,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za64_s64_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddva_za64_s64_1u10__SVBool_tu10__SVBool_tu11__SVInt64_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 2 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv2i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv2i64(i64 7, <vscale x 2 x i1> [[TMP0]], <vscale x 2 x i1> [[TMP1]], <vscale x 2 x i64> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za64_s64_1(svbool_t pn, svbool_t pm, svint64_t zn) {
+ SME_ACLE_FUNC(svaddva_za64, _s64,)(7, pn, pm, zn);
+}
Index: clang/test/CodeGen/aarch64-sme-intrinsics/acle_sme_add-i32.c
===================================================================
--- /dev/null
+++ clang/test/CodeGen/aarch64-sme-intrinsics/acle_sme_add-i32.c
@@ -0,0 +1,159 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
+// REQUIRES: aarch64-registered-target
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -DSME_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -DSME_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -D__ARM_FEATURE_SME -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sve -fallow-half-arguments-and-returns -S -O1 -Werror -o /dev/null %s
+
+#include <arm_sme.h>
+
+#ifdef SME_OVERLOADED_FORMS
+#define SME_ACLE_FUNC(A1,A2_UNUSED,A3) A1##A3
+#else
+#define SME_ACLE_FUNC(A1,A2,A3) A1##A2##A3
+#endif
+
+// CHECK-LABEL: @test_svaddha_za32_u32(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddha_za32_u32u10__SVBool_tu10__SVBool_tu12__SVUint32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za32_u32(svbool_t pn, svbool_t pm, svuint32_t zn) {
+ SME_ACLE_FUNC(svaddha_za32, _u32,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddha_za32_u32_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddha_za32_u32_1u10__SVBool_tu10__SVBool_tu12__SVUint32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za32_u32_1(svbool_t pn, svbool_t pm, svuint32_t zn) {
+ SME_ACLE_FUNC(svaddha_za32, _u32,)(3, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddha_za32_s32(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddha_za32_s32u10__SVBool_tu10__SVBool_tu11__SVInt32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za32_s32(svbool_t pn, svbool_t pm, svint32_t zn) {
+ SME_ACLE_FUNC(svaddha_za32, _s32,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddha_za32_s32_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddha_za32_s32_1u10__SVBool_tu10__SVBool_tu11__SVInt32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addha.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddha_za32_s32_1(svbool_t pn, svbool_t pm, svint32_t zn) {
+ SME_ACLE_FUNC(svaddha_za32, _s32,)(3, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za32_u32(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddva_za32_u32u10__SVBool_tu10__SVBool_tu12__SVUint32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za32_u32(svbool_t pn, svbool_t pm, svuint32_t zn) {
+ SME_ACLE_FUNC(svaddva_za32, _u32,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za32_u32_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddva_za32_u32_1u10__SVBool_tu10__SVBool_tu12__SVUint32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za32_u32_1(svbool_t pn, svbool_t pm, svuint32_t zn) {
+ SME_ACLE_FUNC(svaddva_za32, _u32,)(3, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za32_s32(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z21test_svaddva_za32_s32u10__SVBool_tu10__SVBool_tu11__SVInt32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 0, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za32_s32(svbool_t pn, svbool_t pm, svint32_t zn) {
+ SME_ACLE_FUNC(svaddva_za32, _s32,)(0, pn, pm, zn);
+}
+
+// CHECK-LABEL: @test_svaddva_za32_s32_1(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: @_Z23test_svaddva_za32_s32_1u10__SVBool_tu10__SVBool_tu11__SVInt32_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PN:%.*]])
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PM:%.*]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.addva.nxv4i32(i64 3, <vscale x 4 x i1> [[TMP0]], <vscale x 4 x i1> [[TMP1]], <vscale x 4 x i32> [[ZN:%.*]])
+// CPP-CHECK-NEXT: ret void
+//
+__attribute__((arm_streaming)) void test_svaddva_za32_s32_1(svbool_t pn, svbool_t pm, svint32_t zn) {
+ SME_ACLE_FUNC(svaddva_za32, _s32,)(3, pn, pm, zn);
+}
Index: clang/lib/CodeGen/CodeGenFunction.h
===================================================================
--- clang/lib/CodeGen/CodeGenFunction.h
+++ clang/lib/CodeGen/CodeGenFunction.h
@@ -4246,6 +4246,9 @@
llvm::Value *EmitSMELdrStr(SVETypeFlags TypeFlags,
llvm::SmallVectorImpl<llvm::Value *> &Ops,
unsigned IntID);
+ llvm::Value *EmitSMEAdd(SVETypeFlags TypeFlags,
+ llvm::SmallVectorImpl<llvm::Value *> &Ops,
+ unsigned IntID);
llvm::Value *EmitSVEPrefetchLoad(const SVETypeFlags &TypeFlags,
SmallVectorImpl<llvm::Value *> &Ops,
unsigned BuiltinID);
Index: clang/lib/CodeGen/CGBuiltin.cpp
===================================================================
--- clang/lib/CodeGen/CGBuiltin.cpp
+++ clang/lib/CodeGen/CGBuiltin.cpp
@@ -9149,6 +9149,17 @@
return Builder.CreateCall(F, Ops);
}
+Value *CodeGenFunction::EmitSMEAdd(SVETypeFlags TypeFlags,
+ SmallVectorImpl<Value *> &Ops,
+ unsigned IntID) {
+ auto *DefaultType = getSVEType(TypeFlags);
+ Ops[0] = Builder.CreateIntCast(Ops[0], Int64Ty, false);
+ Ops[1] = EmitSVEPredicateCast(Ops[1], DefaultType);
+ Ops[2] = EmitSVEPredicateCast(Ops[2], DefaultType);
+ Function *F = CGM.getIntrinsic(IntID, {DefaultType});
+ return Builder.CreateCall(F, Ops);
+}
+
// Limit the usage of scalable llvm IR generated by the ACLE by using the
// sve dup.x intrinsic instead of IRBuilder::CreateVectorSplat.
Value *CodeGenFunction::EmitSVEDupX(Value *Scalar, llvm::Type *Ty) {
@@ -9289,6 +9300,8 @@
return EmitSMEZero(TypeFlags, Ops, Builtin->LLVMIntrinsic);
else if (TypeFlags.isSMELdr() || TypeFlags.isSMEStr())
return EmitSMELdrStr(TypeFlags, Ops, Builtin->LLVMIntrinsic);
+ else if (TypeFlags.isSMEAdd())
+ return EmitSMEAdd(TypeFlags, Ops, Builtin->LLVMIntrinsic);
else if (TypeFlags.isGatherLoad())
return EmitSVEGatherLoad(TypeFlags, Ops, Builtin->LLVMIntrinsic);
else if (TypeFlags.isScatterStore())
Index: clang/lib/Basic/Targets/AArch64.h
===================================================================
--- clang/lib/Basic/Targets/AArch64.h
+++ clang/lib/Basic/Targets/AArch64.h
@@ -56,6 +56,7 @@
bool HasMOPS;
bool HasRCPC;
bool HasSME;
+ bool HasSMEI64;
llvm::AArch64::ArchKind ArchKind;
Index: clang/lib/Basic/Targets/AArch64.cpp
===================================================================
--- clang/lib/Basic/Targets/AArch64.cpp
+++ clang/lib/Basic/Targets/AArch64.cpp
@@ -411,6 +411,9 @@
if ((FPU & NeonMode) && HasFP16FML)
Builder.defineMacro("__ARM_FEATURE_FP16_FML", "1");
+ if (HasSME && HasSMEI64)
+ Builder.defineMacro("__ARM_FEATURE_SME_I16I64", "1");
+
if (Opts.hasSignReturnAddress()) {
// Bitmask:
// 0: Protection using the A key
@@ -557,6 +560,7 @@
HasMOPS = false;
HasRCPC = false;
HasSME = false;
+ HasSMEI64 = false;
ArchKind = llvm::AArch64::ArchKind::INVALID;
@@ -568,6 +572,12 @@
HasBFloat16 = true;
HasFullFP16 = true;
}
+ if (Feature == "+sme-i64") {
+ HasSME = true;
+ HasSMEI64 = true;
+ HasBFloat16 = true;
+ HasFullFP16 = true;
+ }
if (Feature == "+sve") {
FPU |= SveMode;
HasFullFP16 = true;
Index: clang/include/clang/Basic/arm_sve.td
===================================================================
--- clang/include/clang/Basic/arm_sve.td
+++ clang/include/clang/Basic/arm_sve.td
@@ -215,6 +215,7 @@
def IsSMELdr : FlagType<0x20000000000>;
def IsSMEStr : FlagType<0x40000000000>;
def IsSMECnt : FlagType<0x80000000000>;
+def IsSMEAdd : FlagType<0x100000000000>;
// These must be kept in sync with the flags in include/clang/Basic/TargetBuiltins.h
class ImmCheckType<int val> {
@@ -2182,3 +2183,15 @@
def SVCNTSH : SInst<"svcntsh", "n", "", MergeNone, "aarch64_sme_cntsh", [IsOverloadNone, IsSME, IsSMECnt]>;
def SVCNTSW : SInst<"svcntsw", "n", "", MergeNone, "aarch64_sme_cntsw", [IsOverloadNone, IsSME, IsSMECnt]>;
def SVCNTSD : SInst<"svcntsd", "n", "", MergeNone, "aarch64_sme_cntsd", [IsOverloadNone, IsSME, IsSMECnt]>;
+
+////////////////////////////////////////////////////////////////////////////////
+// SME - ADDHA/ADDVA
+
+def SVADDHA_ZA32 : SInst<"svaddha_za32[_{d}]", "viPPd", "iUi", MergeNone, "aarch64_sme_addha", [IsSME, IsSMEAdd]>;
+let ArchGuard = "defined(__ARM_FEATURE_SME_I16I64)" in {
+ def SVADDHA_ZA64 : SInst<"svaddha_za64[_{d}]", "viPPd", "lUl", MergeNone, "aarch64_sme_addha", [IsSME, IsSMEAdd]>;
+}
+def SVADDVA_ZA32 : SInst<"svaddva_za32[_{d}]", "viPPd", "iUi", MergeNone, "aarch64_sme_addva", [IsSME, IsSMEAdd]>;
+let ArchGuard = "defined(__ARM_FEATURE_SME_I16I64)" in {
+ def SVADDVA_ZA64 : SInst<"svaddva_za64[_{d}]", "viPPd", "lUl", MergeNone, "aarch64_sme_addva", [IsSME, IsSMEAdd]>;
+}
Index: clang/include/clang/Basic/TargetBuiltins.h
===================================================================
--- clang/include/clang/Basic/TargetBuiltins.h
+++ clang/include/clang/Basic/TargetBuiltins.h
@@ -289,6 +289,7 @@
bool isSMEZero() const { return Flags & IsSMEZero; }
bool isSMELdr() const { return Flags & IsSMELdr; }
bool isSMEStr() const { return Flags & IsSMEStr; }
+ bool isSMEAdd() const { return Flags & IsSMEAdd; }
uint64_t getBits() const { return Flags; }
bool isFlagSet(uint64_t Flag) const { return Flags & Flag; }
_______________________________________________
cfe-commits mailing list
cfe-commits@lists.llvm.org
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
