[clang] [AArch64][SME2] Enable bfm builtins for sme2 (PR #71927)
Sam Tebbs via cfe-commits
cfe-commits at lists.llvm.org
Wed Nov 15 06:45:25 PST 2023
https://github.com/SamTebbs33 updated https://github.com/llvm/llvm-project/pull/71927
>From ff33739706ce758f416153aa3cdb7c632e068da3 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Thu, 9 Nov 2023 17:39:26 +0000
Subject: [PATCH 1/4] [AArch64][SME2] Enable bfm builtings for sme2
This patch enables the following builtins for SME2
svbfmlslb_f32
svbfmlslb_lane_f32
svbfmlslt_f32
svbfmlslt_lane_f32
Patch by: Kerry McLaughlin <kerry.mclaughlin at arm.com>
---
clang/include/clang/Basic/arm_sve.td | 15 +++++++++------
.../acle_sve2p1_bfmlsl.c | 5 +++++
2 files changed, 14 insertions(+), 6 deletions(-)
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 3d4c2129565903d..1094c424e68abd5 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -1959,12 +1959,6 @@ def SVDOT_X2_F : SInst<"svdot[_{d}_{2}_{3}]", "ddhh", "f", MergeNone, "aarch64_
def SVDOT_LANE_X2_S : SInst<"svdot_lane[_{d}_{2}_{3}]", "ddhhi", "i", MergeNone, "aarch64_sve_sdot_lane_x2", [], [ImmCheck<3, ImmCheck0_3>]>;
def SVDOT_LANE_X2_U : SInst<"svdot_lane[_{d}_{2}_{3}]", "ddhhi", "Ui", MergeNone, "aarch64_sve_udot_lane_x2", [], [ImmCheck<3, ImmCheck0_3>]>;
def SVDOT_LANE_X2_F : SInst<"svdot_lane[_{d}_{2}_{3}]", "ddhhi", "f", MergeNone, "aarch64_sve_fdot_lane_x2", [], [ImmCheck<3, ImmCheck0_3>]>;
-
-def SVBFMLSLB : SInst<"svbfmlslb[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslb", [IsOverloadNone], []>;
-def SVBFMLSLT : SInst<"svbfmlslt[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslt", [IsOverloadNone], []>;
-
-def SVBFMLSLB_LANE : SInst<"svbfmlslb_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslb_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
-def SVBFMLSLT_LANE : SInst<"svbfmlslt_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslt_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
}
let TargetGuard = "sve2p1" in {
@@ -1992,3 +1986,12 @@ let TargetGuard = "sme2" in {
def SVADD_SINGLE_X2 : SInst<"svadd[_single_{d}_x2]", "22d", "cUcsUsiUilUl", MergeNone, "aarch64_sve_add_single_x2", [IsStreaming], []>;
def SVADD_SINGLE_X4 : SInst<"svadd[_single_{d}_x4]", "44d", "cUcsUsiUilUl", MergeNone, "aarch64_sve_add_single_x4", [IsStreaming], []>;
}
+
+let TargetGuard = "sve2p1|sme" in {
+// == BFloat16 multiply-subtract ==
+ def SVBFMLSLB : SInst<"svbfmlslb[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslb", [IsOverloadNone], []>;
+ def SVBFMLSLT : SInst<"svbfmlslt[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslt", [IsOverloadNone], []>;
+
+ def SVBFMLSLB_LANE : SInst<"svbfmlslb_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslb_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
+ def SVBFMLSLT_LANE : SInst<"svbfmlslt_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslt_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
+}
diff --git a/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c b/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c
index 7e1763a63ec4e1d..519bf019ff35ba5 100644
--- a/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c
+++ b/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c
@@ -2,10 +2,15 @@
// REQUIRES: aarch64-registered-target
// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s
// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve2p1 -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sve -S -disable-O0-optnone -Werror -Wall -emit-llvm -o - -x c++ %s | opt -S -p mem2reg,instcombine,tailcallelim | FileCheck %s -check-prefix=CPP-CHECK
// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sve2p1 -target-feature -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -target-feature +sve -target-feature -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
#include <arm_sve.h>
>From a87cbf3ac40a29fb858550a846df300ad14ce13d Mon Sep 17 00:00:00 2001
From: Sam Tebbs <sam at tebbs.me>
Date: Fri, 10 Nov 2023 17:13:14 +0000
Subject: [PATCH 2/4] fixup! fix sme targetguard
Co-authored-by: Matthew Devereau <matthew.devereau at arm.com>
---
clang/include/clang/Basic/arm_sve.td | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 1094c424e68abd5..25a65b12001654c 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -1987,7 +1987,7 @@ let TargetGuard = "sme2" in {
def SVADD_SINGLE_X4 : SInst<"svadd[_single_{d}_x4]", "44d", "cUcsUsiUilUl", MergeNone, "aarch64_sve_add_single_x4", [IsStreaming], []>;
}
-let TargetGuard = "sve2p1|sme" in {
+let TargetGuard = "sve2p1|sme2" in {
// == BFloat16 multiply-subtract ==
def SVBFMLSLB : SInst<"svbfmlslb[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslb", [IsOverloadNone], []>;
def SVBFMLSLT : SInst<"svbfmlslt[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslt", [IsOverloadNone], []>;
>From 4358addaa230ff660789fda593eb03cfff29cc20 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Tue, 14 Nov 2023 14:53:53 +0000
Subject: [PATCH 3/4] add IsStreaming to builtins
---
clang/include/clang/Basic/arm_sve.td | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 25a65b12001654c..be65dc6e1ddb7aa 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -1989,9 +1989,9 @@ let TargetGuard = "sme2" in {
let TargetGuard = "sve2p1|sme2" in {
// == BFloat16 multiply-subtract ==
- def SVBFMLSLB : SInst<"svbfmlslb[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslb", [IsOverloadNone], []>;
- def SVBFMLSLT : SInst<"svbfmlslt[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslt", [IsOverloadNone], []>;
+ def SVBFMLSLB : SInst<"svbfmlslb[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslb", [IsOverloadNone, IsStreaming], []>;
+ def SVBFMLSLT : SInst<"svbfmlslt[_{d}]", "dd$$", "f", MergeNone, "aarch64_sve_bfmlslt", [IsOverloadNone, IsStreaming], []>;
- def SVBFMLSLB_LANE : SInst<"svbfmlslb_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslb_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
- def SVBFMLSLT_LANE : SInst<"svbfmlslt_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslt_lane", [IsOverloadNone], [ImmCheck<3, ImmCheck0_7>]>;
+ def SVBFMLSLB_LANE : SInst<"svbfmlslb_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslb_lane", [IsOverloadNone, IsStreaming], [ImmCheck<3, ImmCheck0_7>]>;
+ def SVBFMLSLT_LANE : SInst<"svbfmlslt_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslt_lane", [IsOverloadNone, IsStreaming], [ImmCheck<3, ImmCheck0_7>]>;
}
>From 7ebad6fe1a8fc105e9c02757daf743a4c07e4fb5 Mon Sep 17 00:00:00 2001
From: Samuel Tebbs <samuel.tebbs at arm.com>
Date: Tue, 14 Nov 2023 16:23:50 +0000
Subject: [PATCH 4/4] add __arm_streaming_compatible attribute to tests
---
.../aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c | 8 ++++----
.../Sema/aarch64-sme2-intrinsics/acle_sme2_imm.cpp | 11 +++++++++++
2 files changed, 15 insertions(+), 4 deletions(-)
create mode 100644 clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_imm.cpp
diff --git a/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c b/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c
index 519bf019ff35ba5..d0fd07ec43d3cd5 100644
--- a/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c
+++ b/clang/test/CodeGen/aarch64-sve2p1-intrinsics/acle_sve2p1_bfmlsl.c
@@ -34,7 +34,7 @@
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlslb(<vscale x 4 x float> [[ZDA:%.*]], <vscale x 8 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
// CPP-CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
//
-svfloat32_t test_bfmlslb(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm)
+svfloat32_t test_bfmlslb(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm) __arm_streaming_compatible
{
return SVE_ACLE_FUNC(svbfmlslb,_f32,,)(zda, zn, zm);
}
@@ -50,7 +50,7 @@ svfloat32_t test_bfmlslb(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm)
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlslb.lane(<vscale x 4 x float> [[ZDA:%.*]], <vscale x 8 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
// CPP-CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
//
-svfloat32_t test_bfmlslb_lane(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm)
+svfloat32_t test_bfmlslb_lane(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm) __arm_streaming_compatible
{
return SVE_ACLE_FUNC(svbfmlslb_lane,_f32,,)(zda, zn, zm, 7);
}
@@ -68,7 +68,7 @@ svfloat32_t test_bfmlslb_lane(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm)
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlslt(<vscale x 4 x float> [[ZDA:%.*]], <vscale x 8 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]])
// CPP-CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
//
-svfloat32_t test_bfmlslt(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm)
+svfloat32_t test_bfmlslt(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm) __arm_streaming_compatible
{
return SVE_ACLE_FUNC(svbfmlslt,_f32,,)(zda, zn, zm);
}
@@ -84,7 +84,7 @@ svfloat32_t test_bfmlslt(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm)
// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.bfmlslt.lane(<vscale x 4 x float> [[ZDA:%.*]], <vscale x 8 x bfloat> [[ZN:%.*]], <vscale x 8 x bfloat> [[ZM:%.*]], i32 7)
// CPP-CHECK-NEXT: ret <vscale x 4 x float> [[TMP0]]
//
-svfloat32_t test_bfmlslt_lane(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm)
+svfloat32_t test_bfmlslt_lane(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm) __arm_streaming_compatible
{
return SVE_ACLE_FUNC(svbfmlslt_lane,_f32,,)(zda, zn, zm, 7);
}
diff --git a/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_imm.cpp b/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_imm.cpp
new file mode 100644
index 000000000000000..1e5851e7337479d
--- /dev/null
+++ b/clang/test/Sema/aarch64-sme2-intrinsics/acle_sme2_imm.cpp
@@ -0,0 +1,11 @@
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu \
+// RUN: -target-feature +sve2 -target-feature +sme2 -target-feature +sme-i16i64 -target-feature +sme-f64f64 -fsyntax-only -verify %s
+
+// REQUIRES: aarch64-registered-target
+
+#include <arm_sme_draft_spec_subject_to_change.h>
+
+void test_bfmlslb_bad_lane(svfloat32_t zda, svbfloat16_t zn, svbfloat16_t zm) __arm_streaming_compatible {
+ svbfmlslb_lane_f32(zda, zn, zm, 8); // expected-error {{argument value 8 is outside the valid range [0, 7]}}
+ svbfmlslt_lane_f32(zda, zn, zm, 8); // expected-error {{argument value 8 is outside the valid range [0, 7]}}
+}
More information about the cfe-commits
mailing list