Skip to content

[AArch64] Add intrinsics for multi-vector to ZA array vector accumulators #88266

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom
Merged
merged 1 commit into from
May 9, 2024
Merged

[AArch64] Add intrinsics for multi-vector to ZA array vector accumulators #88266

merged 1 commit into from
May 9, 2024

Conversation

momchil-velikov
Copy link
Collaborator

According to the specification in
ARM-software/acle#309 this adds the intrinsics

void_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za");
void_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za");
void_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za");
void_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za");

as well as the corresponding bf16 variants.

@llvmbot llvmbot added clang Clang issues not falling into any other category backend:AArch64 clang:frontend Language frontend issues, e.g. anything involving "Sema" llvm:ir labels Apr 10, 2024
@llvmbot
Copy link
Member

llvmbot commented Apr 10, 2024 •
edited
Loading

@llvm/pr-subscribers-clang

@llvm/pr-subscribers-backend-aarch64

Author: Momchil Velikov (momchil-velikov)

Changes

According to the specification in
ARM-software/acle#309 this adds the intrinsics

void_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za");
void_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za");
void_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za");
void_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za");

as well as the corresponding bf16 variants.

Patch is 31.84 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/88266.diff

6 Files Affected:

  • (modified) clang/include/clang/Basic/arm_sme.td (+11)
  • (added) clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c (+191)
  • (added) clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c (+26)
  • (modified) llvm/include/llvm/IR/IntrinsicsAArch64.td (+2-2)
  • (modified) llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td (+8-8)
  • (added) llvm/test/CodeGen/AArch64/sme2-intrinsics-add-sub-za16.ll (+146)
diff --git a/clang/include/clang/Basic/arm_sme.td b/clang/include/clang/Basic/arm_sme.td
index 1ac6d5170ea283..dcfaefa7a3e266 100644
--- a/clang/include/clang/Basic/arm_sme.td
+++ b/clang/include/clang/Basic/arm_sme.td
@@ -298,6 +298,17 @@ multiclass ZAAddSub<string n_suffix> {
     def NAME # _ZA64_VG1X2_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x2", "vm2", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x2", [IsStreaming, IsInOutZA], []>;
     def NAME # _ZA64_VG1X4_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x4", "vm4", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x4", [IsStreaming, IsInOutZA], []>;
   }
+
+  let TargetGuard = "sme2p1,sme-f16f16" in {
+    def NAME # _ZA16_VG1X2_F16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x2", "vm2", "h", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x2", [IsStreaming, IsInOutZA], []>;
+    def NAME # _ZA16_VG1X4_F16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x4", "vm4", "h", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x4", [IsStreaming, IsInOutZA], []>;
+  }
+
+  let TargetGuard = "sme2p1,b16b16" in {
+    def NAME # _ZA16_VG1X2_BF16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x2", "vm2", "b", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x2", [IsStreaming, IsInOutZA], []>;
+    def NAME # _ZA16_VG1X4_BF16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x4", "vm4", "b", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x4", [IsStreaming, IsInOutZA], []>;
+  }
+
 }
 
 defm SVADD : ZAAddSub<"add">;
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
new file mode 100644
index 00000000000000..bdf07f86b9c93d
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
@@ -0,0 +1,191 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 4
+// RUN: %clang_cc1                               -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1                        -x c++ -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CXX
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS        -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -x c++ -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CXX
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_sme.h>
+
+#ifdef SVE_OVERLOADED_FORMS
+#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3) A1##A3
+#else
+#define SVE_ACLE_FUNC(A1,A2,A3) A1##A2##A3
+#endif
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x2_f16j13svfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_f16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x4_f16j13svfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_f16,_vg1x4)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x2_f16j13svfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_f16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x4_f16j13svfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_f16,_vg1x4)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x2_bf16j14svbfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_bf16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x4_bf16j14svbfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_bf16,_vg1x4)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x2_bf16j14svbfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_bf16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x4_bf16j14svbfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_bf16,_vg1x4)(slice, zn);
+}
diff --git a/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c b/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
new file mode 100644
index 00000000000000..963c6d03115761
--- /dev/null
+++ b/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
@@ -0,0 +1,26 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -fsyntax-only -verify -emit-llvm %s
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_sme.h>
+
+void test_features(uint32_t slice, svfloat16x2_t zn2, svfloat16x4_t zn4,
+                   svbfloat16x2_t bzn2, svbfloat16x4_t bzn4) __arm_streaming __arm_inout("za") {
+  // expected-error@+1 {{'svadd_za16_f16_vg1x2' needs target feature sme2p1,sme-f16f16}}
+  svadd_za16_f16_vg1x2(slice, zn2);
+  // expected-error@+1 {{'svadd_za16_f16_vg1x4' needs target feature sme2p1,sme-f16f16}}
+  svadd_za16_f16_vg1x4(slice, zn4);
+  // expected-error@+1 {{'svsub_za16_f16_vg1x2' needs target feature sme2p1,sme-f16f16}}
+  svsub_za16_f16_vg1x2(slice, zn2);
+  // expected-error@+1 {{'svsub_za16_f16_vg1x4' needs target feature sme2p1,sme-f16f16}}
+  svsub_za16_f16_vg1x4(slice, zn4);
+
+  // expected-error@+1 {{'svadd_za16_bf16_vg1x2' needs target feature sme2p1,b16b16}}
+  svadd_za16_bf16_vg1x2(slice, bzn2);
+  // expected-error@+1 {{'svadd_za16_bf16_vg1x4' needs target feature sme2p1,b16b16}}
+  svadd_za16_bf16_vg1x4(slice, bzn4);
+  // expected-error@+1 {{'svsub_za16_bf16_vg1x2' needs target feature sme2p1,b16b16}}
+  svsub_za16_bf16_vg1x2(slice, bzn2);
+  // expected-error@+1 {{'svsub_za16_bf16_vg1x4' needs target feature sme2p1,b16b16}}
+  svsub_za16_bf16_vg1x4(slice, bzn4);
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsAArch64.td b/llvm/include/llvm/IR/IntrinsicsAArch64.td
index bcaa37de74b630..8abf082cc312cc 100644
--- a/llvm/include/llvm/IR/IntrinsicsAArch64.td
+++ b/llvm/include/llvm/IR/IntrinsicsAArch64.td
@@ -3479,7 +3479,7 @@ let TargetPrefix = "aarch64" in {
   // Multi-vector add/sub and accumulate into ZA
   //
   foreach intr = ["add", "sub"] in {
-    foreach za = ["za32", "za64"] in {
+    foreach za = ["za16","za32", "za64"] in {
       def int_aarch64_sme_ # intr # _ # za # _vg1x2 : SME2_ZA_Write_...
[truncated]

@llvmbot
Copy link
Member

llvmbot commented Apr 10, 2024

@llvm/pr-subscribers-llvm-ir

Author: Momchil Velikov (momchil-velikov)

Changes

According to the specification in
ARM-software/acle#309 this adds the intrinsics

void_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za");
void_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za");
void_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za");
void_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za");

as well as the corresponding bf16 variants.

Patch is 31.84 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/88266.diff

6 Files Affected:

  • (modified) clang/include/clang/Basic/arm_sme.td (+11)
  • (added) clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c (+191)
  • (added) clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c (+26)
  • (modified) llvm/include/llvm/IR/IntrinsicsAArch64.td (+2-2)
  • (modified) llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td (+8-8)
  • (added) llvm/test/CodeGen/AArch64/sme2-intrinsics-add-sub-za16.ll (+146)
diff --git a/clang/include/clang/Basic/arm_sme.td b/clang/include/clang/Basic/arm_sme.td
index 1ac6d5170ea283..dcfaefa7a3e266 100644
--- a/clang/include/clang/Basic/arm_sme.td
+++ b/clang/include/clang/Basic/arm_sme.td
@@ -298,6 +298,17 @@ multiclass ZAAddSub<string n_suffix> {
     def NAME # _ZA64_VG1X2_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x2", "vm2", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x2", [IsStreaming, IsInOutZA], []>;
     def NAME # _ZA64_VG1X4_F64 : Inst<"sv" # n_suffix # "_za64[_{d}]_vg1x4", "vm4", "d", MergeNone, "aarch64_sme_" # n_suffix # "_za64_vg1x4", [IsStreaming, IsInOutZA], []>;
   }
+
+  let TargetGuard = "sme2p1,sme-f16f16" in {
+    def NAME # _ZA16_VG1X2_F16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x2", "vm2", "h", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x2", [IsStreaming, IsInOutZA], []>;
+    def NAME # _ZA16_VG1X4_F16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x4", "vm4", "h", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x4", [IsStreaming, IsInOutZA], []>;
+  }
+
+  let TargetGuard = "sme2p1,b16b16" in {
+    def NAME # _ZA16_VG1X2_BF16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x2", "vm2", "b", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x2", [IsStreaming, IsInOutZA], []>;
+    def NAME # _ZA16_VG1X4_BF16 : Inst<"sv" # n_suffix # "_za16[_{d}]_vg1x4", "vm4", "b", MergeNone, "aarch64_sme_" # n_suffix # "_za16_vg1x4", [IsStreaming, IsInOutZA], []>;
+  }
+
 }
 
 defm SVADD : ZAAddSub<"add">;
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
new file mode 100644
index 00000000000000..bdf07f86b9c93d
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
@@ -0,0 +1,191 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 4
+// RUN: %clang_cc1                               -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1                        -x c++ -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CXX
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS        -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -x c++ -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2p1  -target-feature +sme-f16f16 -target-feature +b16b16 -O2 -S -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CXX
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_sme.h>
+
+#ifdef SVE_OVERLOADED_FORMS
+#define SVE_ACLE_FUNC(A1,A2_UNUSED,A3) A1##A3
+#else
+#define SVE_ACLE_FUNC(A1,A2,A3) A1##A2##A3
+#endif
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x2_f16j13svfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_f16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svadd_za16_vg1x4_f16j13svfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_f16,_vg1x4)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x2_f16j13svfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_f16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_f16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z25test_svsub_za16_vg1x4_f16j13svfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x half> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8f16(i32 [[SLICE]], <vscale x 8 x half> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[TMP2]], <vscale x 8 x half> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_f16,_vg1x4)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x2_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x2_bf16j14svbfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_bf16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svadd_za16_vg1x4_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svadd_za16_vg1x4_bf16j14svbfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.add.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svadd_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svadd_za16,_bf16,_vg1x4)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x2_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x2_bf16j14svbfloat16x2_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x2.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x2_bf16(uint32_t slice, svbfloat16x2_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_bf16,_vg1x2)(slice, zn);
+}
+
+// CHECK-LABEL: define dso_local void @test_svsub_za16_vg1x4_bf16(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  entry:
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-NEXT:    ret void
+//
+// CHECK-CXX-LABEL: define dso_local void @_Z26test_svsub_za16_vg1x4_bf16j14svbfloat16x4_t(
+// CHECK-CXX-SAME: i32 noundef [[SLICE:%.*]], <vscale x 32 x bfloat> [[ZN:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CXX-NEXT:  entry:
+// CHECK-CXX-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 0)
+// CHECK-CXX-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 8)
+// CHECK-CXX-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 16)
+// CHECK-CXX-NEXT:    [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZN]], i64 24)
+// CHECK-CXX-NEXT:    tail call void @llvm.aarch64.sme.sub.za16.vg1x4.nxv8bf16(i32 [[SLICE]], <vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CHECK-CXX-NEXT:    ret void
+//
+void test_svsub_za16_vg1x4_bf16(uint32_t slice, svbfloat16x4_t zn) __arm_streaming __arm_inout("za") {
+  SVE_ACLE_FUNC(svsub_za16,_bf16,_vg1x4)(slice, zn);
+}
diff --git a/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c b/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
new file mode 100644
index 00000000000000..963c6d03115761
--- /dev/null
+++ b/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_add_sub_za16.c
@@ -0,0 +1,26 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -fsyntax-only -verify -emit-llvm %s
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_sme.h>
+
+void test_features(uint32_t slice, svfloat16x2_t zn2, svfloat16x4_t zn4,
+                   svbfloat16x2_t bzn2, svbfloat16x4_t bzn4) __arm_streaming __arm_inout("za") {
+  // expected-error@+1 {{'svadd_za16_f16_vg1x2' needs target feature sme2p1,sme-f16f16}}
+  svadd_za16_f16_vg1x2(slice, zn2);
+  // expected-error@+1 {{'svadd_za16_f16_vg1x4' needs target feature sme2p1,sme-f16f16}}
+  svadd_za16_f16_vg1x4(slice, zn4);
+  // expected-error@+1 {{'svsub_za16_f16_vg1x2' needs target feature sme2p1,sme-f16f16}}
+  svsub_za16_f16_vg1x2(slice, zn2);
+  // expected-error@+1 {{'svsub_za16_f16_vg1x4' needs target feature sme2p1,sme-f16f16}}
+  svsub_za16_f16_vg1x4(slice, zn4);
+
+  // expected-error@+1 {{'svadd_za16_bf16_vg1x2' needs target feature sme2p1,b16b16}}
+  svadd_za16_bf16_vg1x2(slice, bzn2);
+  // expected-error@+1 {{'svadd_za16_bf16_vg1x4' needs target feature sme2p1,b16b16}}
+  svadd_za16_bf16_vg1x4(slice, bzn4);
+  // expected-error@+1 {{'svsub_za16_bf16_vg1x2' needs target feature sme2p1,b16b16}}
+  svsub_za16_bf16_vg1x2(slice, bzn2);
+  // expected-error@+1 {{'svsub_za16_bf16_vg1x4' needs target feature sme2p1,b16b16}}
+  svsub_za16_bf16_vg1x4(slice, bzn4);
+}
diff --git a/llvm/include/llvm/IR/IntrinsicsAArch64.td b/llvm/include/llvm/IR/IntrinsicsAArch64.td
index bcaa37de74b630..8abf082cc312cc 100644
--- a/llvm/include/llvm/IR/IntrinsicsAArch64.td
+++ b/llvm/include/llvm/IR/IntrinsicsAArch64.td
@@ -3479,7 +3479,7 @@ let TargetPrefix = "aarch64" in {
   // Multi-vector add/sub and accumulate into ZA
   //
   foreach intr = ["add", "sub"] in {
-    foreach za = ["za32", "za64"] in {
+    foreach za = ["za16","za32", "za64"] in {
       def int_aarch64_sme_ # intr # _ # za # _vg1x2 : SME2_ZA_Write_...
[truncated]

@momchil-velikov momchil-velikov merged commit e88ba6d into llvm:main May 9, 2024
4 checks passed
@momchil-velikov momchil-velikov mentioned this pull request May 9, 2024
Merged
momchil-velikov added a commit that referenced this pull request May 9, 2024
@momchil-velikov
Revert "[AArch64] Add intrinsics for multi-vector to ZA array vector …
139e0aa 
…accumulators" (#91597)

Reverts #88266  due to test failures

error: 'expected-error' diagnostics seen but not expected: 
(frontend): '-fsyntax-only' action ignored; '-emit-llvm' action
specified previously
@nico
Copy link
Contributor

nico commented May 9, 2024

Thanks for the quick revert!

Is the failure due to a conflict with another commit that landed?

@momchil-velikov
Copy link
Collaborator Author

Thanks for the quick revert!

Is the failure due to a conflict with another commit that landed?

Perhaps, e.g. #91140

momchil-velikov added a commit to momchil-velikov/llvm-project that referenced this pull request May 9, 2024
@momchil-velikov
[AArch64] Add intrinsics for multi-vector to ZA array vector accumula…
fc45f19 
…tors (llvm#88266)

According to the specification in
ARM-software/acle#309 this adds the intrinsics

void_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn)
__arm_streaming __arm_inout("za");
void_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn)
__arm_streaming __arm_inout("za");
void_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn)
__arm_streaming __arm_inout("za");
void_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn)
__arm_streaming __arm_inout("za");

as well as the corresponding `bf16` variants.
momchil-velikov added a commit to momchil-velikov/llvm-project that referenced this pull request May 10, 2024
@momchil-velikov
[AArch64] Add intrinsics for multi-vector to ZA array vector accumula…
d3e381a 
…tors (llvm#88266)

According to the specification in
ARM-software/acle#309 this adds the intrinsics

void_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn)
__arm_streaming __arm_inout("za");
void_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn)
__arm_streaming __arm_inout("za");
void_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn)
__arm_streaming __arm_inout("za");
void_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn)
__arm_streaming __arm_inout("za");

as well as the corresponding `bf16` variants.
momchil-velikov added a commit to momchil-velikov/llvm-project that referenced this pull request May 10, 2024
@momchil-velikov
[AArch64] Add intrinsics for multi-vector to ZA array vector accumula…
43fb20b 
…tors (llvm#88266)

According to the specification in
ARM-software/acle#309 this adds the intrinsics

void_svadd_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn)
__arm_streaming __arm_inout("za");
void_svadd_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn)
__arm_streaming __arm_inout("za");
void_svsub_za16_vg1x2_f16(uint32_t slice, svfloat16x2_t zn)
__arm_streaming __arm_inout("za");
void_svsub_za16_vg1x4_f16(uint32_t slice, svfloat16x4_t zn)
__arm_streaming __arm_inout("za");

as well as the corresponding `bf16` variants.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@CarolineConcatto CarolineConcatto CarolineConcatto approved these changes

@hassnaaHamdi hassnaaHamdi Awaiting requested review from hassnaaHamdi

@Lukacma Lukacma Awaiting requested review from Lukacma

Assignees
No one assigned
Labels
backend:AArch64 clang:frontend Language frontend issues, e.g. anything involving "Sema" clang Clang issues not falling into any other category llvm:ir
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.
4 participants
@momchil-velikov @llvmbot @nico @CarolineConcatto