[clang] [llvm] [AArch64] Add intrinsics for bflaot16 min/max/minnm/maxnm (PR #90105)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Apr 25 12:16:39 PDT 2024
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-llvm-ir
@llvm/pr-subscribers-clang
Author: Hassnaa Hamdi (hassnaaHamdi)
<details>
<summary>Changes</summary>
According to specifications in [ARM-software/acle/pull/309](https://github.com/ARM-software/acle/pull/309)
Add following intrinsics:
```
// svmax single,multi
svbfloat16x2_t svmax_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm)
svbfloat16x4_t svmax_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm)
svbfloat16x2_t svmax_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm)
svbfloat16x4_t svmax_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm)
```
```
// svmin single,multi
svbfloat16x2_t svmin_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm)
svbfloat16x4_t svmin_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm)
svbfloat16x2_t svmin_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm)
svbfloat16x4_t svmin_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm)
```
```
// svmaxnm single,multi
svbfloat16x2_t svmaxnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm)
svbfloat16x4_t svmaxnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm)
svbfloat16x2_t svmaxnm_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm)
svbfloat16x4_t svmaxnm_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm)
```
```
// svminnm single,multi
svbfloat16x2_t svminnm_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm)
svbfloat16x4_t svminnm_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm)
svbfloat16x2_t svminnm_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm)
svbfloat16x4_t svminnm_bf16_x4(svbfloat16x4_t zdn, svbfloat16x4_t zm)
```
- Variations other than bfloat16 had been already supported.
---
Patch is 124.44 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/90105.diff
9 Files Affected:
- (modified) clang/include/clang/Basic/arm_sve.td (+28)
- (modified) clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c (+145-5)
- (modified) clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_maxnm.c (+145-5)
- (modified) clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_min.c (+145-5)
- (modified) clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_minnm.c (+145-5)
- (modified) llvm/include/llvm/IR/IntrinsicsAArch64.td (+2-2)
- (modified) llvm/lib/Target/AArch64/AArch64ISelDAGToDAG.cpp (+113-1)
- (modified) llvm/test/CodeGen/AArch64/sme2-intrinsics-max.ll (+125-1)
- (modified) llvm/test/CodeGen/AArch64/sme2-intrinsics-min.ll (+126-1)
``````````diff
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
index 6cc249837d3f3d..884ca6a3363421 100644
--- a/clang/include/clang/Basic/arm_sve.td
+++ b/clang/include/clang/Basic/arm_sve.td
@@ -2111,6 +2111,20 @@ let TargetGuard = "sme2" in {
defm MIN_MULTI_X4 : MinMaxIntr<"min", "", "x4", "444">;
}
+multiclass BFMinMaxIntr<string name> {
+ def SVBF # NAME # _SINGLE_X2 : SInst<"sv" # name # "[_single_{d}_x2]", "22d", "b", MergeNone, "aarch64_sve_bf" # name # "_single_x2", [IsStreaming], []>;
+ def SVBF # NAME # _SINGLE_X4 : SInst<"sv" # name # "[_single_{d}_x4]", "44d", "b", MergeNone, "aarch64_sve_bf" # name # "_single_x4", [IsStreaming], []>;
+
+ def SVBF # NAME # _MULTI_X2 : SInst<"sv" # name # "[_{d}_x2]", "222", "b", MergeNone, "aarch64_sve_bf" # name # "_x2", [IsStreaming], []>;
+ def SVBF # NAME # _MULTI_X4 : SInst<"sv" # name # "[_{d}_x4]", "444", "b", MergeNone, "aarch64_sve_bf" # name # "_x4", [IsStreaming], []>;
+}
+
+let TargetGuard = "sme2,b16b16" in {
+// == BFMIN / BFMAX ==
+ defm SVMIN : BFMinMaxIntr<"min">;
+ defm SVMAX : BFMinMaxIntr<"max">;
+}
+
multiclass SInstMinMaxByVector<string name> {
def NAME # _SINGLE_X2 : SInst<"sv" # name # "nm[_single_{d}_x2]", "22d", "hfd", MergeNone, "aarch64_sve_f" # name # "nm_single_x2", [IsStreaming], []>;
def NAME # _SINGLE_X4 : SInst<"sv" # name # "nm[_single_{d}_x4]", "44d", "hfd", MergeNone, "aarch64_sve_f" # name # "nm_single_x4", [IsStreaming], []>;
@@ -2125,6 +2139,20 @@ let TargetGuard = "sme2" in {
defm SVMAXNM : SInstMinMaxByVector<"max">;
}
+multiclass SInstBFMinMaxByVector<string name> {
+ def SVBF # NAME # _SINGLE_X2 : SInst<"sv" # name # "[_single_{d}_x2]", "22d", "b", MergeNone, "aarch64_sve_bf" # name # "_single_x2", [IsStreaming], []>;
+ def SVBF # NAME # _SINGLE_X4 : SInst<"sv" # name # "[_single_{d}_x4]", "44d", "b", MergeNone, "aarch64_sve_bf" # name # "_single_x4", [IsStreaming], []>;
+
+ def SVBF # NAME # _MULTI_X2 : SInst<"sv" # name # "[_{d}_x2]", "222", "b", MergeNone, "aarch64_sve_bf" # name # "_x2", [IsStreaming], []>;
+ def SVBF # NAME # _MULTI_X4 : SInst<"sv" # name # "[_{d}_x4]", "444", "b", MergeNone, "aarch64_sve_bf" # name # "_x4", [IsStreaming], []>;
+}
+
+let TargetGuard = "sme2,b16b16" in {
+// == BFMINNM / BFMAXNM ==
+ defm SVMINNM : SInstBFMinMaxByVector<"minnm">;
+ defm SVMAXNM : SInstBFMinMaxByVector<"maxnm">;
+}
+
let TargetGuard = "sme2" in {
// FRINTA / FRINTM / FRINTN / FRINTP
def SVRINTA_X2 : SInst<"svrinta[_{d}_x2]", "22", "f", MergeNone, "aarch64_sve_frinta_x2", [IsStreaming], []>;
diff --git a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
index a4e2616784efa7..2f09ad8074e39e 100644
--- a/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
+++ b/clang/test/CodeGen/aarch64-sme2-intrinsics/acle_sme2_max.c
@@ -1,9 +1,9 @@
// NOTE: Assertions have been autogenerated by utils/update_cc_test_checks.py
-// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64-none-linux-gnu -target-feature +sme2 -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 -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 -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 -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 -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 +b16b16 -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 +b16b16 -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 +b16b16 -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 +b16b16 -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 +b16b16 -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
// REQUIRES: aarch64-registered-target
#include <arm_sme.h>
@@ -224,6 +224,32 @@ svuint64x2_t test_svmax_single_u64_x2(svuint64x2_t zdn, svuint64_t zm) __arm_str
return SVE_ACLE_FUNC(svmax,_single_u64_x2)(zdn, zm);
}
+// CHECK-LABEL: @test_svmax_single_bf16_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CHECK-NEXT: ret <vscale x 16 x bfloat> [[TMP6]]
+//
+// CPP-CHECK-LABEL: @_Z25test_svmax_single_bf16_x214svbfloat16x2_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfmax.single.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 0
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP3]], i64 0)
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP2]], 1
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], i64 8)
+// CPP-CHECK-NEXT: ret <vscale x 16 x bfloat> [[TMP6]]
+//
+svbfloat16x2_t test_svmax_single_bf16_x2(svbfloat16x2_t zdn, svbfloat16_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svmax,_single_bf16_x2)(zdn, zm);
+}
+
// CHECK-LABEL: @test_svmax_single_f16_x2(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -608,6 +634,44 @@ svuint64x4_t test_svmax_single_u64_x4(svuint64x4_t zdn, svuint64_t zm) __arm_str
return SVE_ACLE_FUNC(svmax,_single_u64_x4)(zdn, zm);
}
+// CHECK-LABEL: @test_svmax_single_bf16_x4(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CHECK-NEXT: ret <vscale x 32 x bfloat> [[TMP12]]
+//
+// CPP-CHECK-LABEL: @_Z25test_svmax_single_bf16_x414svbfloat16x4_tu14__SVBfloat16_t(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfmax.single.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[ZM:%.*]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 2
+// CPP-CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP8]], <vscale x 8 x bfloat> [[TMP9]], i64 16)
+// CPP-CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 3
+// CPP-CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 24)
+// CPP-CHECK-NEXT: ret <vscale x 32 x bfloat> [[TMP12]]
+//
+svbfloat16x4_t test_svmax_single_bf16_x4(svbfloat16x4_t zdn, svbfloat16_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svmax,_single_bf16_x4)(zdn, zm);
+}
+
// CHECK-LABEL: @test_svmax_single_f16_x4(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv32f16(<vscale x 32 x half> [[ZDN:%.*]], i64 0)
@@ -964,6 +1028,36 @@ svuint64x2_t test_svmax_u64_x2(svuint64x2_t zdn, svuint64x2_t zm) __arm_streamin
return SVE_ACLE_FUNC(svmax,_u64_x2)(zdn, zm);
}
+// CHECK-LABEL: @test_svmax_bf16_x2(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfmax.x2.nxv8bf16(<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: [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CHECK-NEXT: ret <vscale x 16 x bfloat> [[TMP8]]
+//
+// CPP-CHECK-LABEL: @_Z18test_svmax_bf16_x214svbfloat16x2_tS_(
+// CPP-CHECK-NEXT: entry:
+// CPP-CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZDN]], i64 8)
+// CPP-CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM:%.*]], i64 0)
+// CPP-CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv16bf16(<vscale x 16 x bfloat> [[ZM]], i64 8)
+// CPP-CHECK-NEXT: [[TMP4:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfmax.x2.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]])
+// CPP-CHECK-NEXT: [[TMP5:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 0
+// CPP-CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> poison, <vscale x 8 x bfloat> [[TMP5]], i64 0)
+// CPP-CHECK-NEXT: [[TMP7:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP4]], 1
+// CPP-CHECK-NEXT: [[TMP8:%.*]] = tail call <vscale x 16 x bfloat> @llvm.vector.insert.nxv16bf16.nxv8bf16(<vscale x 16 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]], i64 8)
+// CPP-CHECK-NEXT: ret <vscale x 16 x bfloat> [[TMP8]]
+//
+svbfloat16x2_t test_svmax_bf16_x2(svbfloat16x2_t zdn, svbfloat16x2_t zm) __arm_streaming {
+ return SVE_ACLE_FUNC(svmax,_bf16_x2)(zdn, zm);
+}
+
// CHECK-LABEL: @test_svmax_f16_x2(
// CHECK-NEXT: entry:
// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x half> @llvm.vector.extract.nxv8f16.nxv16f16(<vscale x 16 x half> [[ZDN:%.*]], i64 0)
@@ -1424,6 +1518,52 @@ svuint64x4_t test_svmax_u64_x4(svuint64x4_t zdn, svuint64x4_t zm) __arm_streamin
return SVE_ACLE_FUNC(svmax,_u64_x4)(zdn, zm);
}
+// CHECK-LABEL: @test_svmax_bf16_x4(
+// CHECK-NEXT: entry:
+// CHECK-NEXT: [[TMP0:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN:%.*]], i64 0)
+// CHECK-NEXT: [[TMP1:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 8)
+// CHECK-NEXT: [[TMP2:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 16)
+// CHECK-NEXT: [[TMP3:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZDN]], i64 24)
+// CHECK-NEXT: [[TMP4:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM:%.*]], i64 0)
+// CHECK-NEXT: [[TMP5:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 8)
+// CHECK-NEXT: [[TMP6:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 16)
+// CHECK-NEXT: [[TMP7:%.*]] = tail call <vscale x 8 x bfloat> @llvm.vector.extract.nxv8bf16.nxv32bf16(<vscale x 32 x bfloat> [[ZM]], i64 24)
+// CHECK-NEXT: [[TMP8:%.*]] = tail call { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } @llvm.aarch64.sve.bfmax.x4.nxv8bf16(<vscale x 8 x bfloat> [[TMP0]], <vscale x 8 x bfloat> [[TMP1]], <vscale x 8 x bfloat> [[TMP2]], <vscale x 8 x bfloat> [[TMP3]], <vscale x 8 x bfloat> [[TMP4]], <vscale x 8 x bfloat> [[TMP5]], <vscale x 8 x bfloat> [[TMP6]], <vscale x 8 x bfloat> [[TMP7]])
+// CHECK-NEXT: [[TMP9:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 0
+// CHECK-NEXT: [[TMP10:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> poison, <vscale x 8 x bfloat> [[TMP9]], i64 0)
+// CHECK-NEXT: [[TMP11:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 1
+// CHECK-NEXT: [[TMP12:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP10]], <vscale x 8 x bfloat> [[TMP11]], i64 8)
+// CHECK-NEXT: [[TMP13:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x 8 x bfloat> } [[TMP8]], 2
+// CHECK-NEXT: [[TMP14:%.*]] = tail call <vscale x 32 x bfloat> @llvm.vector.insert.nxv32bf16.nxv8bf16(<vscale x 32 x bfloat> [[TMP12]], <vscale x 8 x bfloat> [[TMP13]], i64 16)
+// CHECK-NEXT: [[TMP15:%.*]] = extractvalue { <vscale x 8 x bfloat>, <vscale x 8 x bfloat>, <vscale x...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/90105
More information about the llvm-commits
mailing list