[AArch64] Implement intrinsics for SVE FAMIN/FAMAX #99042

momchil-velikov · 2024-07-16T13:35:33Z

This patch implements the following intrinsics:

Floating-point absolute maximum (predicated)

svfloat16_t svamax[_f16]_m(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svamax[_f16]_x(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svamax[_f16]_z(svbool_t, svfloat16_t, svfloat16_t);

svfloat16_t svamax[_n_f16]_m(svbool_t, svfloat16_t, float16_t); svfloat16_t svamax[_n_f16]_x(svbool_t, svfloat16_t, float16_t); svfloat16_t svamax[_n_f16]_z(svbool_t, svfloat16_t, float16_t);
Floating-point absolute minimum (predicated)

svfloat16_t svmin[_f16]_m(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svmin[_f16]_x(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svmin[_f16]_z(svbool_t, svfloat16_t, svfloat16_t);

svfloat16_t svmin[_n_f16]_m(svbool_t, svfloat16_t, float16_t); svfloat16_t svmin[_n_f16]_x(svbool_t, svfloat16_t, float16_t); svfloat16_t svmin[_n_f16]_z(svbool_t, svfloat16_t, float16_t);

All the intrinsics have also variants for f32 and f64, and have the __arm_streaming attribute.

(cf. ARM-software/acle#324)

llvmbot · 2024-07-16T13:36:08Z

@llvm/pr-subscribers-backend-aarch64
@llvm/pr-subscribers-llvm-ir

@llvm/pr-subscribers-clang

Author: Momchil Velikov (momchil-velikov)

Changes

This patch implements the following intrinsics:

Floating-point absolute maximum (predicated)

svfloat16_t svamax[_f16]_m(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svamax[_f16]_x(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svamax[_f16]_z(svbool_t, svfloat16_t, svfloat16_t);

svfloat16_t svamax[_n_f16]_m(svbool_t, svfloat16_t, float16_t); svfloat16_t svamax[_n_f16]_x(svbool_t, svfloat16_t, float16_t); svfloat16_t svamax[_n_f16]_z(svbool_t, svfloat16_t, float16_t);
Floating-point absolute minimum (predicated)

svfloat16_t svmin[_f16]_m(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svmin[_f16]_x(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svmin[_f16]_z(svbool_t, svfloat16_t, svfloat16_t);

svfloat16_t svmin[_n_f16]_m(svbool_t, svfloat16_t, float16_t); svfloat16_t svmin[_n_f16]_x(svbool_t, svfloat16_t, float16_t); svfloat16_t svmin[_n_f16]_z(svbool_t, svfloat16_t, float16_t);

All the intrinsics have also variants for f32 and f64, and have the __arm_streaming attribute.

(cf. ARM-software/acle#324)

Patch is 85.95 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/99042.diff

7 Files Affected:

(modified) clang/include/clang/Basic/arm_sve.td (+5)
(added) clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_faminmax.c (+775)
(modified) llvm/include/llvm/IR/IntrinsicsAArch64.td (+7)
(modified) llvm/lib/Target/AArch64/AArch64ISelLowering.cpp (+8)
(modified) llvm/lib/Target/AArch64/AArch64ISelLowering.h (+2)
(modified) llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td (+12-2)
(added) llvm/test/CodeGen/AArch64/sve2-intrinsics-faminmax.ll (+266)

diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 94c093d891156..be27015ba42c3 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -2385,3 +2385,8 @@ let SVETargetGuard = "sve2p1", SMETargetGuard = "sme2" in {
   def SVBFMLSLB_LANE : SInst<"svbfmlslb_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslb_lane", [IsOverloadNone, VerifyRuntimeMode], [ImmCheck<3, ImmCheck0_7>]>;
   def SVBFMLSLT_LANE : SInst<"svbfmlslt_lane[_{d}]", "dd$$i", "f", MergeNone, "aarch64_sve_bfmlslt_lane", [IsOverloadNone, VerifyRuntimeMode], [ImmCheck<3, ImmCheck0_7>]>;
 }
+
+let SVETargetGuard = "sve2,faminmax", SMETargetGuard = "sme2,faminmax" in {
+  defm SVAMIN : SInstZPZZ<"svamin", "hfd", "aarch64_sve_famin", "aarch64_sve_famin_u">;
+  defm SVAMAX : SInstZPZZ<"svamax", "hfd", "aarch64_sve_famax", "aarch64_sve_famax_u">;
+}
diff --git a/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_faminmax.c b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_faminmax.c
new file mode 100644
index 0000000000000..3cf7d99d606f3
--- /dev/null
+++ b/clang/test/CodeGen/aarch64-sve2-intrinsics/acle_sve2_faminmax.c
@@ -0,0 +1,775 @@
+// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py UTC_ARGS: --version 5
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -target-feature +faminmax -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme2 -target-feature +faminmax -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+// RUN: %clang_cc1 -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -target-feature +faminmax -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
+
+// RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -target-feature +faminmax -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CPP
+// RUN: %clang_cc1 -x c++ -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme2 -target-feature +faminmax -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CPP
+// RUN: %clang_cc1 -x c++ -DSVE_OVERLOADED_FORMS -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -target-feature +faminmax -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s -check-prefix CHECK-CPP
+
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sve -target-feature +sve2 -target-feature +faminmax -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: %clang_cc1 -triple aarch64-none-linux-gnu -target-feature +sme -target-feature +sme2 -target-feature +faminmax -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+
+// REQUIRES: aarch64-registered-target
+
+#ifdef __ARM_FEATURE_SME
+#include "arm_sme.h"
+#else
+#include "arm_sve.h"
+#endif
+
+#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
+
+#ifdef __ARM_FEATURE_SME
+#define STREAMING __arm_streaming
+#else
+#define STREAMING
+#endif
+
+// CHECK-LABEL: define dso_local <vscale x 8 x half> @test_famin_f16_m(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], <vscale x 8 x half> [[B:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[B]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 8 x half> @_Z16test_famin_f16_mu10__SVBool_tu13__SVFloat16_tS0_(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], <vscale x 8 x half> [[B:%.*]]) local_unnamed_addr #[[ATTR0:[0-9]+]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[B]])
+// CHECK-CPP-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+svfloat16_t test_famin_f16_m(svbool_t pg, svfloat16_t a, svfloat16_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _f16, _m)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x half> @test_famin_f16_x(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], <vscale x 8 x half> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.u.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[B]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 8 x half> @_Z16test_famin_f16_xu10__SVBool_tu13__SVFloat16_tS0_(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], <vscale x 8 x half> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.u.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[B]])
+// CHECK-CPP-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+svfloat16_t test_famin_f16_x(svbool_t pg, svfloat16_t a, svfloat16_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _f16, _x)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x half> @test_famin_f16_z(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], <vscale x 8 x half> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[B]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 8 x half> @_Z16test_famin_f16_zu10__SVBool_tu13__SVFloat16_tS0_(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], <vscale x 8 x half> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> zeroinitializer
+// CHECK-CPP-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[B]])
+// CHECK-CPP-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+//
+svfloat16_t test_famin_f16_z(svbool_t pg, svfloat16_t a, svfloat16_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _f16, _z)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x half> @test_famin_n_f16_m(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], half noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x half> poison, half [[B]], i64 0
+// CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x half> [[DOTSPLATINSERT]], <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[DOTSPLAT]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 8 x half> @_Z18test_famin_n_f16_mu10__SVBool_tu13__SVFloat16_tDh(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], half noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x half> poison, half [[B]], i64 0
+// CHECK-CPP-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x half> [[DOTSPLATINSERT]], <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[DOTSPLAT]])
+// CHECK-CPP-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+svfloat16_t test_famin_n_f16_m(svbool_t pg, svfloat16_t a, float16_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _n_f16, _m)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x half> @test_famin_n_f16_x(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], half noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x half> poison, half [[B]], i64 0
+// CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x half> [[DOTSPLATINSERT]], <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.u.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[DOTSPLAT]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 8 x half> @_Z18test_famin_n_f16_xu10__SVBool_tu13__SVFloat16_tDh(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], half noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x half> poison, half [[B]], i64 0
+// CHECK-CPP-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x half> [[DOTSPLATINSERT]], <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.u.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> [[DOTSPLAT]])
+// CHECK-CPP-NEXT:    ret <vscale x 8 x half> [[TMP1]]
+//
+svfloat16_t test_famin_n_f16_x(svbool_t pg, svfloat16_t a, float16_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _n_f16, _x)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 8 x half> @test_famin_n_f16_z(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], half noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-NEXT:  [[ENTRY:.*:]]
+// CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x half> poison, half [[B]], i64 0
+// CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x half> [[DOTSPLATINSERT]], <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[DOTSPLAT]])
+// CHECK-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 8 x half> @_Z18test_famin_n_f16_zu10__SVBool_tu13__SVFloat16_tDh(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 8 x half> [[A:%.*]], half noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv8i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 8 x half> poison, half [[B]], i64 0
+// CHECK-CPP-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 8 x half> [[DOTSPLATINSERT]], <vscale x 8 x half> poison, <vscale x 8 x i32> zeroinitializer
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = select <vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[A]], <vscale x 8 x half> zeroinitializer
+// CHECK-CPP-NEXT:    [[TMP2:%.*]] = tail call <vscale x 8 x half> @llvm.aarch64.sve.famin.nxv8f16(<vscale x 8 x i1> [[TMP0]], <vscale x 8 x half> [[TMP1]], <vscale x 8 x half> [[DOTSPLAT]])
+// CHECK-CPP-NEXT:    ret <vscale x 8 x half> [[TMP2]]
+//
+svfloat16_t test_famin_n_f16_z(svbool_t pg, svfloat16_t a, float16_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _n_f16, _z)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 4 x float> @test_famin_f32_m(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// 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> [[PG]])
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> [[B]])
+// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP1]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 4 x float> @_Z16test_famin_f32_mu10__SVBool_tu13__SVFloat32_tS0_(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> [[B]])
+// CHECK-CPP-NEXT:    ret <vscale x 4 x float> [[TMP1]]
+//
+svfloat32_t test_famin_f32_m(svbool_t pg, svfloat32_t a, svfloat32_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _f32, _m)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 4 x float> @test_famin_f32_x(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// 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> [[PG]])
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.u.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> [[B]])
+// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP1]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 4 x float> @_Z16test_famin_f32_xu10__SVBool_tu13__SVFloat32_tS0_(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.u.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> [[B]])
+// CHECK-CPP-NEXT:    ret <vscale x 4 x float> [[TMP1]]
+//
+svfloat32_t test_famin_f32_x(svbool_t pg, svfloat32_t a, svfloat32_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _f32, _x)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 4 x float> @test_famin_f32_z(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// 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> [[PG]])
+// CHECK-NEXT:    [[TMP1:%.*]] = select <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> zeroinitializer
+// CHECK-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[B]])
+// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP2]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 4 x float> @_Z16test_famin_f32_zu10__SVBool_tu13__SVFloat32_tS0_(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], <vscale x 4 x float> [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = select <vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> zeroinitializer
+// CHECK-CPP-NEXT:    [[TMP2:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[TMP1]], <vscale x 4 x float> [[B]])
+// CHECK-CPP-NEXT:    ret <vscale x 4 x float> [[TMP2]]
+//
+svfloat32_t test_famin_f32_z(svbool_t pg, svfloat32_t a, svfloat32_t b) STREAMING {
+  return SVE_ACLE_FUNC(svamin, _f32, _z)(pg, a, b);
+}
+
+// CHECK-LABEL: define dso_local <vscale x 4 x float> @test_famin_n_f32_m(
+// CHECK-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// 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> [[PG]])
+// CHECK-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 4 x float> poison, float [[B]], i64 0
+// CHECK-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 4 x float> [[DOTSPLATINSERT]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
+// CHECK-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> [[DOTSPLAT]])
+// CHECK-NEXT:    ret <vscale x 4 x float> [[TMP1]]
+//
+// CHECK-CPP-LABEL: define dso_local <vscale x 4 x float> @_Z18test_famin_n_f32_mu10__SVBool_tu13__SVFloat32_tf(
+// CHECK-CPP-SAME: <vscale x 16 x i1> [[PG:%.*]], <vscale x 4 x float> [[A:%.*]], float noundef [[B:%.*]]) local_unnamed_addr #[[ATTR0]] {
+// CHECK-CPP-NEXT:  [[ENTRY:.*:]]
+// CHECK-CPP-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i1> @llvm.aarch64.sve.convert.from.svbool.nxv4i1(<vscale x 16 x i1> [[PG]])
+// CHECK-CPP-NEXT:    [[DOTSPLATINSERT:%.*]] = insertelement <vscale x 4 x float> poison, float [[B]], i64 0
+// CHECK-CPP-NEXT:    [[DOTSPLAT:%.*]] = shufflevector <vscale x 4 x float> [[DOTSPLATINSERT]], <vscale x 4 x float> poison, <vscale x 4 x i32> zeroinitializer
+// CHECK-CPP-NEXT:    [[TMP1:%.*]] = tail call <vscale x 4 x float> @llvm.aarch64.sve.famin.nxv4f32(<vscale x 4 x i1> [[TMP0]], <vscale x 4 x float> [[A]], <vscale x 4 x float> [[DOTSPLAT]])
+// CHECK-CPP-NEXT:    ret <vscale x 4 x float> [[TMP1]]
+//
+svfloat32_t test_famin_n_f32_m(svbool_t pg, svfloat32_t a, float32_t b) STREAMING {
+  return SVE_A...
[truncated]

clang/include/clang/Basic/arm_sve.td

llvm/lib/Target/AArch64/AArch64ISelLowering.h

paulwalker-arm

A couple of suggestions but otherwise this looks good to me.

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td

llvm/test/CodeGen/AArch64/sve2-intrinsics-faminmax.ll

momchil-velikov · 2024-09-04T10:02:53Z

Rebased.

This patch implements the following intrinsics: * Floating-point absolute maximum (predicated) svfloat16_t svamax[_f16]_m(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svamax[_f16]_x(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svamax[_f16]_z(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svamax[_n_f16]_m(svbool_t, svfloat16_t, float16_t); svfloat16_t svamax[_n_f16]_x(svbool_t, svfloat16_t, float16_t); svfloat16_t svamax[_n_f16]_z(svbool_t, svfloat16_t, float16_t); * Floating-point absolute minimum (predicated) svfloat16_t svmin[_f16]_m(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svmin[_f16]_x(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svmin[_f16]_z(svbool_t, svfloat16_t, svfloat16_t); svfloat16_t svmin[_n_f16]_m(svbool_t, svfloat16_t, float16_t); svfloat16_t svmin[_n_f16]_x(svbool_t, svfloat16_t, float16_t); svfloat16_t svmin[_n_f16]_z(svbool_t, svfloat16_t, float16_t); All the intrinsics have also variants for `f32` and `f64`, and have the `__arm_streaming` attribute. (cf. ARM-software/acle#324)

momchil-velikov requested review from Lukacma and CarolineConcatto July 16, 2024 13:35

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 Jul 16, 2024

CarolineConcatto approved these changes Jul 25, 2024

View reviewed changes

CarolineConcatto reviewed Jul 25, 2024

View reviewed changes

clang/include/clang/Basic/arm_sve.td Show resolved Hide resolved

paulwalker-arm reviewed Aug 13, 2024

View reviewed changes

llvm/lib/Target/AArch64/AArch64ISelLowering.h Outdated Show resolved Hide resolved

momchil-velikov force-pushed the faminmax-sve branch from 02d790e to 7dcd5a7 Compare August 29, 2024 09:45

paulwalker-arm approved these changes Aug 29, 2024

View reviewed changes

llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td Outdated Show resolved Hide resolved

llvm/test/CodeGen/AArch64/sve2-intrinsics-faminmax.ll Outdated Show resolved Hide resolved

momchil-velikov force-pushed the faminmax-sve branch from 261e75c to 628d5d4 Compare September 4, 2024 10:00

momchil-velikov added 3 commits September 4, 2024 11:22

[fixup] Remove AArch64ISD::FA{MIN,MAX}_PRED

7525e9d

[fixup] Stuff (NFC)

562512c

momchil-velikov force-pushed the faminmax-sve branch from 628d5d4 to 562512c Compare September 4, 2024 10:22

momchil-velikov merged commit bb1b368 into llvm:main Sep 4, 2024
5 of 8 checks passed

momchil-velikov deleted the faminmax-sve branch November 13, 2024 09:33

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

[AArch64] Implement intrinsics for SVE FAMIN/FAMAX #99042

[AArch64] Implement intrinsics for SVE FAMIN/FAMAX #99042

Uh oh!

momchil-velikov commented Jul 16, 2024

Uh oh!

llvmbot commented Jul 16, 2024 •

edited

Loading

Uh oh!

Uh oh!

Uh oh!

paulwalker-arm left a comment

Uh oh!

Uh oh!

Uh oh!

momchil-velikov commented Sep 4, 2024

Uh oh!

Uh oh!

Uh oh!

[AArch64] Implement intrinsics for SVE FAMIN/FAMAX #99042

[AArch64] Implement intrinsics for SVE FAMIN/FAMAX #99042

Uh oh!

Conversation

momchil-velikov commented Jul 16, 2024

Uh oh!

llvmbot commented Jul 16, 2024 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

paulwalker-arm left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

momchil-velikov commented Sep 4, 2024

Uh oh!

Uh oh!

Uh oh!

llvmbot commented Jul 16, 2024 •

edited

Loading