[clang] [llvm] [AArch64] Implement intrinsics for FMLAL/FMLALL (single) (PR #119568)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Dec 11 07:27:37 PST 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-aarch64
Author: None (SpencerAbson)
<details>
<summary>Changes</summary>
Multi-vector 8-bit floating-point multiply-add long (single)
```c
// Only if __ARM_FEATURE_SME_F8F16 != 0
void svmla[_single]_za16[_mf8]_vg2x1_fpm(uint32_t slice, svmfloat8_t zn,
svmfloat8_t zm, fpm_t fpm)
__arm_streaming __arm_inout("za");
void svmla[_single]_za16[_mf8]_vg2x2_fpm(uint32_t slice, svmfloat8x2_t zn,
svmfloat8_t zm, fpm_t fpm)
__arm_streaming __arm_inout("za");
void svmla[_single]_za16[_mf8]_vg2x4_fpm(uint32_t slice, svmfloat8x4_t zn,
svmfloat8_t zm, fpm_t fpm)
__arm_streaming __arm_inout("za");
// Only if __ARM_FEATURE_SME_F8F32 != 0
void svmla[_single]_za32[_mf8]_vg4x1_fpm(uint32_t slice, svmfloat8_t zn,
svmfloat8_t zm, fpm_t fpm)
__arm_streaming __arm_inout("za");
void svmla[_single]_za32[_mf8]_vg4x2_fpm(uint32_t slice, svmfloat8x2_t zn,
svmfloat8_t zm, fpm_t fpm)
__arm_streaming __arm_inout("za");
void svmla[_single]_za32[_mf8]_vg4x4_fpm(uint32_t slice, svmfloat8x4_t zn,
svmfloat8_t zm, fpm_t fpm)
__arm_streaming __arm_inout("za");
```
In accordance with https://github.com/ARM-software/acle/pull/323.
Co-authored-by: Momchil Velikov momchil.velikov@<!-- -->arm.com
---
Patch is 45.64 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/119568.diff
7 Files Affected:
- (modified) clang/include/clang/Basic/arm_sme.td (+14)
- (added) clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sme2_fp8_mla.c (+128)
- (added) clang/test/Sema/aarch64-fp8-intrinsics/acle_sme_fp8_mla.c (+26)
- (modified) llvm/include/llvm/IR/IntrinsicsAArch64.td (+31)
- (modified) llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td (+30-26)
- (modified) llvm/lib/Target/AArch64/SMEInstrFormats.td (+34-23)
- (added) llvm/test/CodeGen/AArch64/sme2-fp8-intrinsics-mla.ll (+101)
``````````diff
diff --git a/clang/include/clang/Basic/arm_sme.td b/clang/include/clang/Basic/arm_sme.td
index 71b2c7cdd04f93..8e5786f196a288 100644
--- a/clang/include/clang/Basic/arm_sme.td
+++ b/clang/include/clang/Basic/arm_sme.td
@@ -827,11 +827,25 @@ let SMETargetGuard = "sme-lutv2" in {
let SMETargetGuard = "sme-f8f32" in {
def SVMOPA_FP8_ZA32 : Inst<"svmopa_za32[_mf8]_m_fpm", "viPPdd>", "m", MergeNone, "aarch64_sme_fp8_fmopa_za32",
[IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], [ImmCheck<0, ImmCheck0_3>]>;
+ // FMLAL (single)
+ def SVMLA_FP8_SINGLE_ZA32_VG4x1 : Inst<"svmla[_single]_za32[_mf8]_vg4x1_fpm", "vmdd>", "m", MergeNone, "aarch64_sme_fp8_fmlall_single_za32_vg4x1",
+ [IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], []>;
+ def SVMLA_FP8_SINGLE_ZA32_VG4x2 : Inst<"svmla[_single]_za32[_mf8]_vg4x2_fpm", "vm2d>", "m", MergeNone, "aarch64_sme_fp8_fmlall_single_za32_vg4x2",
+ [IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], []>;
+ def SVMLA_FP8_SINGLE_ZA32_VG4x4 : Inst<"svmla[_single]_za32[_mf8]_vg4x4_fpm", "vm4d>", "m", MergeNone, "aarch64_sme_fp8_fmlall_single_za32_vg4x4",
+ [IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], []>;
}
let SMETargetGuard = "sme-f8f16" in {
def SVMOPA_FP8_ZA16 : Inst<"svmopa_za16[_mf8]_m_fpm", "viPPdd>", "m", MergeNone, "aarch64_sme_fp8_fmopa_za16",
[IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], [ImmCheck<0, ImmCheck0_1>]>;
+ // FMLAL (single)
+ def SVMLA_FP8_SINGLE_ZA16_VG2x1 : Inst<"svmla[_single]_za16[_mf8]_vg2x1_fpm", "vmdd>", "m", MergeNone, "aarch64_sme_fp8_fmlal_single_za16_vg2x1",
+ [IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], []>;
+ def SVMLA_FP8_SINGLE_ZA16_VG2x2 : Inst<"svmla[_single]_za16[_mf8]_vg2x2_fpm", "vm2d>", "m", MergeNone, "aarch64_sme_fp8_fmlal_single_za16_vg2x2",
+ [IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], []>;
+ def SVMLA_FP8_SINGLE_ZA16_VG2x4 : Inst<"svmla[_single]_za16[_mf8]_vg2x4_fpm", "vm4d>", "m", MergeNone, "aarch64_sme_fp8_fmlal_single_za16_vg2x4",
+ [IsStreaming, IsInOutZA, SetsFPMR, IsOverloadNone], []>;
}
} // let SVETargetGuard = InvalidMode
diff --git a/clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sme2_fp8_mla.c b/clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sme2_fp8_mla.c
new file mode 100644
index 00000000000000..0f6d45474c363f
--- /dev/null
+++ b/clang/test/CodeGen/AArch64/fp8-intrinsics/acle_sme2_fp8_mla.c
@@ -0,0 +1,128 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+// REQUIRES: aarch64-registered-target
+
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme-f8f16 -target-feature +sme-f8f32 -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme-f8f16 -target-feature +sme-f8f32 -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -DSME_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme-f8f16 -target-feature +sme-f8f32 -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -DSME_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme-f8f16 -target-feature +sme-f8f32 -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -passes=mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme-f8f16 -target-feature +sme-f8f32 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+
+#include <arm_sme.h>
+
+#ifdef SME_OVERLOADED_FORMS
+#define SME_ACLE_FUNC(A1, A2_UNUSED, A3, A4_UNUSED, A5) A1##A3##A5
+#else
+#define SME_ACLE_FUNC(A1, A2, A3, A4, A5) A1##A2##A3##A4##A5
+#endif
+
+// FMLAL (single)
+
+// CHECK-LABEL: define dso_local void @test_svmla_single_za16_vg2x1(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlal.single.za16.vg2x1(i32 [[SLICE]], <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: define dso_local void @_Z28test_svmla_single_za16_vg2x1ju13__SVMfloat8_tS_m(
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0:[0-9]+]] {
+// CPP-CHECK-NEXT: [[ENTRY:.*:]]
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlal.single.za16.vg2x1(i32 [[SLICE]], <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svmla_single_za16_vg2x1(uint32_t slice, svmfloat8_t zn, svmfloat8_t zm, fpm_t fpm) __arm_streaming __arm_inout("za") {
+ SME_ACLE_FUNC(svmla,_single,_za16,_mf8,_vg2x1_fpm)(slice, zn, zm, fpm);
+}
+
+// CHECK-LABEL: define dso_local void @test_svmla_single_za16_vg2x2(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlal.single.za16.vg2x2(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: define dso_local void @_Z28test_svmla_single_za16_vg2x2j13svmfloat8x2_tu13__SVMfloat8_tm(
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: [[ENTRY:.*:]]
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlal.single.za16.vg2x2(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svmla_single_za16_vg2x2(uint32_t slice, svmfloat8x2_t zn, svmfloat8_t zm, fpm_t fpm) __arm_streaming __arm_inout("za") {
+ SME_ACLE_FUNC(svmla,_single,_za16,_mf8,_vg2x2_fpm)(slice, zn, zm, fpm);
+}
+
+// CHECK-LABEL: define dso_local void @test_svmla_single_za16_vg2x4(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlal.single.za16.vg2x4(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZN_COERCE2]], <vscale x 16 x i8> [[ZN_COERCE3]], <vscale x 16 x i8> [[ZM]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: define dso_local void @_Z28test_svmla_single_za16_vg2x4j13svmfloat8x4_tu13__SVMfloat8_tm(
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: [[ENTRY:.*:]]
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlal.single.za16.vg2x4(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZN_COERCE2]], <vscale x 16 x i8> [[ZN_COERCE3]], <vscale x 16 x i8> [[ZM]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svmla_single_za16_vg2x4(uint32_t slice, svmfloat8x4_t zn, svmfloat8_t zm, fpm_t fpm) __arm_streaming __arm_inout("za") {
+ SME_ACLE_FUNC(svmla,_single,_za16,_mf8,_vg2x4_fpm)(slice, zn, zm, fpm);
+}
+
+// FMLALL (single)
+
+// CHECK-LABEL: define dso_local void @test_svmla_single_za32_vg4x1(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlall.single.za32.vg4x1(i32 [[SLICE]], <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: define dso_local void @_Z28test_svmla_single_za32_vg4x1ju13__SVMfloat8_tS_m(
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: [[ENTRY:.*:]]
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlall.single.za32.vg4x1(i32 [[SLICE]], <vscale x 16 x i8> [[ZN]], <vscale x 16 x i8> [[ZM]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svmla_single_za32_vg4x1(uint32_t slice, svmfloat8_t zn, svmfloat8_t zm, fpm_t fpm) __arm_streaming __arm_inout("za") {
+ SME_ACLE_FUNC(svmla,_single,_za32,_mf8,_vg4x1_fpm)(slice, zn, zm, fpm);
+}
+
+// CHECK-LABEL: define dso_local void @test_svmla_single_za32_vg4x2(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlall.single.za32.vg4x2(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: define dso_local void @_Z28test_svmla_single_za32_vg4x2j13svmfloat8x2_tu13__SVMfloat8_tm(
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: [[ENTRY:.*:]]
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlall.single.za32.vg4x2(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZM]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svmla_single_za32_vg4x2(uint32_t slice, svmfloat8x2_t zn, svmfloat8_t zm, fpm_t fpm) __arm_streaming __arm_inout("za") {
+ SME_ACLE_FUNC(svmla,_single,_za32,_mf8,_vg4x2_fpm)(slice, zn, zm, fpm);
+}
+
+// CHECK-LABEL: define dso_local void @test_svmla_single_za32_vg4x4(
+// CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CHECK-NEXT: [[ENTRY:.*:]]
+// CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlall.single.za32.vg4x4(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZN_COERCE2]], <vscale x 16 x i8> [[ZN_COERCE3]], <vscale x 16 x i8> [[ZM]])
+// CHECK-NEXT: ret void
+//
+// CPP-CHECK-LABEL: define dso_local void @_Z28test_svmla_single_za32_vg4x4j13svmfloat8x4_tu13__SVMfloat8_tm(
+// CPP-CHECK-SAME: i32 noundef [[SLICE:%.*]], <vscale x 16 x i8> [[ZN_COERCE0:%.*]], <vscale x 16 x i8> [[ZN_COERCE1:%.*]], <vscale x 16 x i8> [[ZN_COERCE2:%.*]], <vscale x 16 x i8> [[ZN_COERCE3:%.*]], <vscale x 16 x i8> [[ZM:%.*]], i64 noundef [[FPM:%.*]]) #[[ATTR0]] {
+// CPP-CHECK-NEXT: [[ENTRY:.*:]]
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.set.fpmr(i64 [[FPM]])
+// CPP-CHECK-NEXT: tail call void @llvm.aarch64.sme.fp8.fmlall.single.za32.vg4x4(i32 [[SLICE]], <vscale x 16 x i8> [[ZN_COERCE0]], <vscale x 16 x i8> [[ZN_COERCE1]], <vscale x 16 x i8> [[ZN_COERCE2]], <vscale x 16 x i8> [[ZN_COERCE3]], <vscale x 16 x i8> [[ZM]])
+// CPP-CHECK-NEXT: ret void
+//
+void test_svmla_single_za32_vg4x4(uint32_t slice, svmfloat8x4_t zn, svmfloat8_t zm, fpm_t fpm) __arm_streaming __arm_inout("za") {
+ SME_ACLE_FUNC(svmla,_single,_za32,_mf8,_vg4x4_fpm)(slice, zn, zm, fpm);
+}
diff --git a/clang/test/Sema/aarch64-fp8-intrinsics/acle_sme_fp8_mla.c b/clang/test/Sema/aarch64-fp8-intrinsics/acle_sme_fp8_mla.c
new file mode 100644
index 00000000000000..b299490e264868
--- /dev/null
+++ b/clang/test/Sema/aarch64-fp8-intrinsics/acle_sme_fp8_mla.c
@@ -0,0 +1,26 @@
+// RUN: %clang_cc1 -triple aarch64 -target-feature +sme -verify -emit-llvm-only %s
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_sme.h>
+
+void test_svmla(uint32_t slice, svmfloat8_t zn, svmfloat8x2_t znx2, svmfloat8x4_t znx4,
+ fpm_t fpmr) __arm_streaming __arm_inout("za") {
+ // expected-error at +1 {{'svmla_single_za16_mf8_vg2x1_fpm' needs target feature sme,sme-f8f16}}
+ svmla_single_za16_mf8_vg2x1_fpm(slice, zn, zn, fpmr);
+
+ // expected-error at +1 {{'svmla_single_za16_mf8_vg2x2_fpm' needs target feature sme,sme-f8f16}}
+ svmla_single_za16_mf8_vg2x2_fpm(slice, znx2, zn, fpmr);
+
+ // expected-error at +1 {{'svmla_single_za16_mf8_vg2x4_fpm' needs target feature sme,sme-f8f16}}
+ svmla_single_za16_mf8_vg2x4_fpm(slice, znx4, zn, fpmr);
+
+ // expected-error at +1 {{'svmla_single_za32_mf8_vg4x1_fpm' needs target feature sme,sme-f8f32}}
+ svmla_single_za32_mf8_vg4x1_fpm(slice, zn, zn, fpmr);
+
+ // expected-error at +1 {{'svmla_single_za32_mf8_vg4x2_fpm' needs target feature sme,sme-f8f32}}
+ svmla_single_za32_mf8_vg4x2_fpm(slice, znx2, zn, fpmr);
+
+ // expected-error at +1 {{'svmla_single_za32_mf8_vg4x4_fpm' needs target feature sme,sme-f8f32}}
+ svmla_single_za32_mf8_vg4x4_fpm(slice, znx4, zn, fpmr);
+}
\ No newline at end of file
diff --git a/llvm/include/llvm/IR/IntrinsicsAArch64.td b/llvm/include/llvm/IR/IntrinsicsAArch64.td
index b2f0aa2f7e4d90..f893cd437144c2 100644
--- a/llvm/include/llvm/IR/IntrinsicsAArch64.td
+++ b/llvm/include/llvm/IR/IntrinsicsAArch64.td
@@ -3888,6 +3888,24 @@ let TargetPrefix = "aarch64" in {
llvm_nxv16i1_ty, llvm_nxv16i1_ty,
llvm_nxv16i8_ty, llvm_nxv16i8_ty],
[ImmArg<ArgIndex<0>>, IntrInaccessibleMemOnly, IntrHasSideEffects]>;
+
+ class SME_FP8_ZA_SINGLE_VGx1_Intrinsic
+ : DefaultAttrsIntrinsic<[], [llvm_i32_ty,
+ llvm_nxv16i8_ty,
+ llvm_nxv16i8_ty],
+ [IntrInaccessibleMemOnly, IntrHasSideEffects]>;
+
+ class SME_FP8_ZA_SINGLE_VGx2_Intrinsic
+ : DefaultAttrsIntrinsic<[], [llvm_i32_ty,
+ llvm_nxv16i8_ty, llvm_nxv16i8_ty,
+ llvm_nxv16i8_ty],
+ [IntrInaccessibleMemOnly, IntrHasSideEffects]>;
+
+ class SME_FP8_ZA_SINGLE_VGx4_Intrinsic
+ : DefaultAttrsIntrinsic<[], [llvm_i32_ty,
+ llvm_nxv16i8_ty, llvm_nxv16i8_ty, llvm_nxv16i8_ty, llvm_nxv16i8_ty,
+ llvm_nxv16i8_ty],
+ [IntrInaccessibleMemOnly, IntrHasSideEffects]>;
//
// CVT from FP8 to half-precision/BFloat16 multi-vector
//
@@ -3914,4 +3932,17 @@ let TargetPrefix = "aarch64" in {
// FP8 outer product
def int_aarch64_sme_fp8_fmopa_za16 : SME_FP8_OuterProduct_Intrinsic;
def int_aarch64_sme_fp8_fmopa_za32 : SME_FP8_OuterProduct_Intrinsic;
+
+ //
+ // ZA multiply-add
+ //
+ // Double-vector groups (F8F16)
+ def int_aarch64_sme_fp8_fmlal_single_za16_vg2x1 : SME_FP8_ZA_SINGLE_VGx1_Intrinsic;
+ def int_aarch64_sme_fp8_fmlal_single_za16_vg2x2 : SME_FP8_ZA_SINGLE_VGx2_Intrinsic;
+ def int_aarch64_sme_fp8_fmlal_single_za16_vg2x4 : SME_FP8_ZA_SINGLE_VGx4_Intrinsic;
+
+ // Quad-vector groups (F8F32)
+ def int_aarch64_sme_fp8_fmlall_single_za32_vg4x1 : SME_FP8_ZA_SINGLE_VGx1_Intrinsic;
+ def int_aarch64_sme_fp8_fmlall_single_za32_vg4x2 : SME_FP8_ZA_SINGLE_VGx2_Intrinsic;
+ def int_aarch64_sme_fp8_fmlall_single_za32_vg4x4 : SME_FP8_ZA_SINGLE_VGx4_Intrinsic;
}
diff --git a/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td
index bd36e21d1be46c..1fd496f6ea4afd 100644
--- a/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SMEInstrInfo.td
@@ -564,8 +564,8 @@ defm SMLALL_MZZI_BtoS : sme2_mla_ll_array_index_32b<"smlall", 0b00, 0b000,
defm SMLALL_VG2_M2ZZI_BtoS : sme2_mla_ll_array_vg2_index_32b<"smlall", 0b00, 0b000, int_aarch64_sme_smla_za32_lane_vg4x2>;
defm SMLALL_VG4_M4ZZI_BtoS : sme2_mla_ll_array_vg4_index_32b<"smlall", 0b00, 0b0000, int_aarch64_sme_smla_za32_lane_vg4x4>;
defm SMLALL_MZZ_BtoS : sme2_mla_ll_array_single<"smlall", 0b00000, MatrixOp32, ZPR8, ZPR4b8, nxv16i8, int_aarch64_sme_smla_za32_single_vg4x1>;
-defm SMLALL_VG2_M2ZZ_BtoS : sme2_mla_ll_array_vg2_single<"smlall", 0b00000, MatrixOp32, ZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_smla_za32_single_vg4x2>;
-defm SMLALL_VG4_M4ZZ_BtoS : sme2_mla_ll_array_vg4_single<"smlall", 0b01000, MatrixOp32, ZZZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_smla_za32_single_vg4x4>;
+defm SMLALL_VG2_M2ZZ_BtoS : sme2_mla_ll_array_vg2_single<"smlall", 0b000000, MatrixOp32, ZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_smla_za32_single_vg4x2>;
+defm SMLALL_VG4_M4ZZ_BtoS : sme2_mla_ll_array_vg4_single<"smlall", 0b010000, MatrixOp32, ZZZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_smla_za32_single_vg4x4>;
defm SMLALL_VG2_M2Z2Z_BtoS : sme2_mla_ll_array_vg2_multi<"smlall", 0b00000, MatrixOp32, ZZ_b_mul_r, nxv16i8, int_aarch64_sme_smla_za32_vg4x2>;
defm SMLALL_VG4_M4Z4Z_BtoS : sme2_mla_ll_array_vg4_multi<"smlall", 0b00000, MatrixOp32, ZZZZ_b_mul_r, nxv16i8, int_aarch64_sme_smla_za32_vg4x4>;
@@ -573,8 +573,8 @@ defm USMLALL_MZZI_BtoS : sme2_mla_ll_array_index_32b<"usmlall", 0b00, 0b001
defm USMLALL_VG2_M2ZZI_BtoS : sme2_mla_ll_array_vg2_index_32b<"usmlall", 0b00, 0b100, int_aarch64_sme_usmla_za32_lane_vg4x2>;
defm USMLALL_VG4_M4ZZI_BtoS : sme2_mla_ll_array_vg4_index_32b<"usmlall", 0b00, 0b0100, int_aarch64_sme_usmla_za32_lane_vg4x4>;
defm USMLALL_MZZ_BtoS : sme2_mla_ll_array_single<"usmlall", 0b00001, MatrixOp32, ZPR8, ZPR4b8, nxv16i8, int_aarch64_sme_usmla_za32_single_vg4x1>;
-defm USMLALL_VG2_M2ZZ_BtoS : sme2_mla_ll_array_vg2_single<"usmlall", 0b00001, MatrixOp32, ZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_usmla_za32_single_vg4x2>;
-defm USMLALL_VG4_M4ZZ_BtoS : sme2_mla_ll_array_vg4_single<"usmlall", 0b01001, MatrixOp32, ZZZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_usmla_za32_single_vg4x4>;
+defm USMLALL_VG2_M2ZZ_BtoS : sme2_mla_ll_array_vg2_single<"usmlall", 0b000010, MatrixOp32, ZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_usmla_za32_single_vg4x2>;
+defm USMLALL_VG4_M4ZZ_BtoS : sme2_mla_ll_array_vg4_single<"usmlall", 0b010010, MatrixOp32, ZZZZ_b, ZPR4b8, nxv16i8, int_aarch64_sme_usmla_za32_single_vg4x4>;
defm USMLALL_VG2_M2Z2Z_BtoS : sme2_mla_ll_array_vg2_multi<"usmlall", 0b00001, MatrixOp32, ZZ_b_mul_r, nxv16i8, int_aarch64_sme_usmla_za32_vg4x2>;
defm USMLALL_VG4_M4Z4Z_BtoS : sme2_mla_ll_array_vg4_multi<"usmlall", 0b00001, MatrixOp32, ZZZZ_b_mul_r, nxv16i8, int_aarch64_sme_usmla_za32_vg4x4>;
@@ -582,8 +582,8 @@ defm SMLSLL_MZZI_BtoS : sme2_mla_ll_array_index_32b<"smlsll", 0b00, 0b010,
defm SMLSLL_VG2_M2ZZI_BtoS : sme2_mla_ll_array_vg2_index_32b<"smlsll", 0b00, 0b001, int_aarch64_sme_smls_za32_lane_vg4x2>;
defm SMLSLL_VG4_M4ZZI_BtoS : sme2_mla_ll_array_vg4_index...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/119568
More information about the llvm-commits
mailing list