[Clang][AArch64] make bitperm intrinsics available in streaming mode (#129700)

Lukacma · web-flow · commit e4d9be3d1205 · 2025-05-02T11:52:09.000+01:00
Based on recent changes in armv9.6 BitPerm instructions and thus intrinsics are available in streaming mode when [FEAT_SSVE_BitPerm](https://developer.arm.com/documentation/109697/2024_12/Feature-descriptions/The-Armv9-6-architecture-extension?lang=en#md463-the-armv96-architecture-extension__feat_FEAT_SSVE_BitPerm) is available. This patch reflects this change and is based on [ACLE proposal](ARM-software/acle#385).
diff --git a/clang/include/clang/Basic/arm_sve.td b/clang/include/clang/Basic/arm_sve.td
@@ -1990,13 +1990,13 @@ def SVSM4E    : SInst<"svsm4e[_{d}]",    "ddd", "Ui", MergeNone, "aarch64_sve_sm
 def SVSM4EKEY : SInst<"svsm4ekey[_{d}]", "ddd", "Ui", MergeNone, "aarch64_sve_sm4ekey", [IsOverloadNone]>;
 }
 
-let SVETargetGuard = "sve2,sve-bitperm", SMETargetGuard = InvalidMode in {
-def SVBDEP   : SInst<"svbdep[_{d}]",   "ddd", "UcUsUiUl", MergeNone, "aarch64_sve_bdep_x">;
-def SVBDEP_N : SInst<"svbdep[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bdep_x">;
-def SVBEXT   : SInst<"svbext[_{d}]",   "ddd", "UcUsUiUl", MergeNone, "aarch64_sve_bext_x">;
-def SVBEXT_N : SInst<"svbext[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bext_x">;
-def SVBGRP   : SInst<"svbgrp[_{d}]",   "ddd", "UcUsUiUl", MergeNone, "aarch64_sve_bgrp_x">;
-def SVBGRP_N : SInst<"svbgrp[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bgrp_x">;
+let SVETargetGuard = "sve2,sve-bitperm", SMETargetGuard = "sme,ssve-bitperm" in {
+def SVBDEP   : SInst<"svbdep[_{d}]",   "ddd", "UcUsUiUl", MergeNone, "aarch64_sve_bdep_x", [VerifyRuntimeMode]>;
+def SVBDEP_N : SInst<"svbdep[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bdep_x", [VerifyRuntimeMode]>;
+def SVBEXT   : SInst<"svbext[_{d}]",   "ddd", "UcUsUiUl", MergeNone, "aarch64_sve_bext_x", [VerifyRuntimeMode]>;
+def SVBEXT_N : SInst<"svbext[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bext_x", [VerifyRuntimeMode]>;
+def SVBGRP   : SInst<"svbgrp[_{d}]",   "ddd", "UcUsUiUl", MergeNone, "aarch64_sve_bgrp_x", [VerifyRuntimeMode]>;
+def SVBGRP_N : SInst<"svbgrp[_n_{d}]", "dda", "UcUsUiUl", MergeNone, "aarch64_sve_bgrp_x", [VerifyRuntimeMode]>;
 }
 
 let SVETargetGuard = "sve2p1", SMETargetGuard = "sme" in {
diff --git a/clang/test/CodeGen/AArch64/sve2-intrinsics/acle_sve2_bdep.c b/clang/test/CodeGen/AArch64/sve2-intrinsics/acle_sve2_bdep.c
@@ -5,6 +5,8 @@
 // RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
 // RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
 // RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64 -target-feature +sme -target-feature +ssve-bitperm -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: not %clang_cc1 -triple aarch64 -target-feature +sme -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
 
 #include <arm_sve.h>
 
@@ -15,6 +17,12 @@
 #define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
 #endif
 
+#ifdef __ARM_FEATURE_SME
+#define STREAMING __arm_streaming
+#else
+#define STREAMING
+#endif
+
 // CHECK-LABEL: @test_svbdep_u8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bdep.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[OP2:%.*]])
@@ -25,7 +33,7 @@
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bdep.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
-svuint8_t test_svbdep_u8(svuint8_t op1, svuint8_t op2)
+svuint8_t test_svbdep_u8(svuint8_t op1, svuint8_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_u8,,)(op1, op2);
 }
@@ -40,7 +48,7 @@ svuint8_t test_svbdep_u8(svuint8_t op1, svuint8_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bdep.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
-svuint16_t test_svbdep_u16(svuint16_t op1, svuint16_t op2)
+svuint16_t test_svbdep_u16(svuint16_t op1, svuint16_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_u16,,)(op1, op2);
 }
@@ -55,7 +63,7 @@ svuint16_t test_svbdep_u16(svuint16_t op1, svuint16_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bdep.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 4 x i32> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
 //
-svuint32_t test_svbdep_u32(svuint32_t op1, svuint32_t op2)
+svuint32_t test_svbdep_u32(svuint32_t op1, svuint32_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_u32,,)(op1, op2);
 }
@@ -70,7 +78,7 @@ svuint32_t test_svbdep_u32(svuint32_t op1, svuint32_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bdep.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.*]], <vscale x 2 x i64> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
 //
-svuint64_t test_svbdep_u64(svuint64_t op1, svuint64_t op2)
+svuint64_t test_svbdep_u64(svuint64_t op1, svuint64_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_u64,,)(op1, op2);
 }
@@ -89,7 +97,7 @@ svuint64_t test_svbdep_u64(svuint64_t op1, svuint64_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bdep.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
-svuint8_t test_svbdep_n_u8(svuint8_t op1, uint8_t op2)
+svuint8_t test_svbdep_n_u8(svuint8_t op1, uint8_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_n_u8,,)(op1, op2);
 }
@@ -108,7 +116,7 @@ svuint8_t test_svbdep_n_u8(svuint8_t op1, uint8_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bdep.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.*]], <vscale x 8 x i16> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
-svuint16_t test_svbdep_n_u16(svuint16_t op1, uint16_t op2)
+svuint16_t test_svbdep_n_u16(svuint16_t op1, uint16_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_n_u16,,)(op1, op2);
 }
@@ -127,7 +135,7 @@ svuint16_t test_svbdep_n_u16(svuint16_t op1, uint16_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bdep.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 4 x i32> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
 //
-svuint32_t test_svbdep_n_u32(svuint32_t op1, uint32_t op2)
+svuint32_t test_svbdep_n_u32(svuint32_t op1, uint32_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_n_u32,,)(op1, op2);
 }
@@ -146,7 +154,7 @@ svuint32_t test_svbdep_n_u32(svuint32_t op1, uint32_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bdep.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.*]], <vscale x 2 x i64> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
 //
-svuint64_t test_svbdep_n_u64(svuint64_t op1, uint64_t op2)
+svuint64_t test_svbdep_n_u64(svuint64_t op1, uint64_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbdep,_n_u64,,)(op1, op2);
 }
diff --git a/clang/test/CodeGen/AArch64/sve2-intrinsics/acle_sve2_bext.c b/clang/test/CodeGen/AArch64/sve2-intrinsics/acle_sve2_bext.c
@@ -5,6 +5,8 @@
 // RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
 // RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
 // RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64 -target-feature +sme -target-feature +ssve-bitperm -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: not %clang_cc1 -triple aarch64 -target-feature +sme -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
 
 #include <arm_sve.h>
 
@@ -15,6 +17,12 @@
 #define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
 #endif
 
+#ifdef __ARM_FEATURE_SME
+#define STREAMING __arm_streaming
+#else
+#define STREAMING
+#endif
+
 // CHECK-LABEL: @test_svbext_u8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bext.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[OP2:%.*]])
@@ -25,7 +33,7 @@
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bext.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
-svuint8_t test_svbext_u8(svuint8_t op1, svuint8_t op2)
+svuint8_t test_svbext_u8(svuint8_t op1, svuint8_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_u8,,)(op1, op2);
 }
@@ -40,7 +48,7 @@ svuint8_t test_svbext_u8(svuint8_t op1, svuint8_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bext.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
-svuint16_t test_svbext_u16(svuint16_t op1, svuint16_t op2)
+svuint16_t test_svbext_u16(svuint16_t op1, svuint16_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_u16,,)(op1, op2);
 }
@@ -55,7 +63,7 @@ svuint16_t test_svbext_u16(svuint16_t op1, svuint16_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bext.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 4 x i32> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
 //
-svuint32_t test_svbext_u32(svuint32_t op1, svuint32_t op2)
+svuint32_t test_svbext_u32(svuint32_t op1, svuint32_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_u32,,)(op1, op2);
 }
@@ -70,7 +78,7 @@ svuint32_t test_svbext_u32(svuint32_t op1, svuint32_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bext.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.*]], <vscale x 2 x i64> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
 //
-svuint64_t test_svbext_u64(svuint64_t op1, svuint64_t op2)
+svuint64_t test_svbext_u64(svuint64_t op1, svuint64_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_u64,,)(op1, op2);
 }
@@ -89,7 +97,7 @@ svuint64_t test_svbext_u64(svuint64_t op1, svuint64_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bext.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
-svuint8_t test_svbext_n_u8(svuint8_t op1, uint8_t op2)
+svuint8_t test_svbext_n_u8(svuint8_t op1, uint8_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_n_u8,,)(op1, op2);
 }
@@ -108,7 +116,7 @@ svuint8_t test_svbext_n_u8(svuint8_t op1, uint8_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bext.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.*]], <vscale x 8 x i16> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
-svuint16_t test_svbext_n_u16(svuint16_t op1, uint16_t op2)
+svuint16_t test_svbext_n_u16(svuint16_t op1, uint16_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_n_u16,,)(op1, op2);
 }
@@ -127,7 +135,7 @@ svuint16_t test_svbext_n_u16(svuint16_t op1, uint16_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bext.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 4 x i32> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
 //
-svuint32_t test_svbext_n_u32(svuint32_t op1, uint32_t op2)
+svuint32_t test_svbext_n_u32(svuint32_t op1, uint32_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_n_u32,,)(op1, op2);
 }
@@ -146,7 +154,7 @@ svuint32_t test_svbext_n_u32(svuint32_t op1, uint32_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bext.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.*]], <vscale x 2 x i64> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
 //
-svuint64_t test_svbext_n_u64(svuint64_t op1, uint64_t op2)
+svuint64_t test_svbext_n_u64(svuint64_t op1, uint64_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbext,_n_u64,,)(op1, op2);
 }
diff --git a/clang/test/CodeGen/AArch64/sve2-intrinsics/acle_sve2_bgrp.c b/clang/test/CodeGen/AArch64/sve2-intrinsics/acle_sve2_bgrp.c
@@ -5,6 +5,8 @@
 // RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
 // RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - %s | FileCheck %s
 // RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s | FileCheck %s -check-prefix=CPP-CHECK
+// RUN: %clang_cc1 -triple aarch64 -target-feature +sme -target-feature +ssve-bitperm -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
+// RUN: not %clang_cc1 -triple aarch64 -target-feature +sme -S -disable-O0-optnone -Werror -Wall -o /dev/null %s
 
 #include <arm_sve.h>
 
@@ -15,6 +17,12 @@
 #define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4
 #endif
 
+#ifdef __ARM_FEATURE_SME
+#define STREAMING __arm_streaming
+#else
+#define STREAMING
+#endif
+
 // CHECK-LABEL: @test_svbgrp_u8(
 // CHECK-NEXT:  entry:
 // CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bgrp.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[OP2:%.*]])
@@ -25,7 +33,7 @@
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bgrp.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
-svuint8_t test_svbgrp_u8(svuint8_t op1, svuint8_t op2)
+svuint8_t test_svbgrp_u8(svuint8_t op1, svuint8_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_u8,,)(op1, op2);
 }
@@ -40,7 +48,7 @@ svuint8_t test_svbgrp_u8(svuint8_t op1, svuint8_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bgrp.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.*]], <vscale x 8 x i16> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
-svuint16_t test_svbgrp_u16(svuint16_t op1, svuint16_t op2)
+svuint16_t test_svbgrp_u16(svuint16_t op1, svuint16_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_u16,,)(op1, op2);
 }
@@ -55,7 +63,7 @@ svuint16_t test_svbgrp_u16(svuint16_t op1, svuint16_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bgrp.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 4 x i32> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
 //
-svuint32_t test_svbgrp_u32(svuint32_t op1, svuint32_t op2)
+svuint32_t test_svbgrp_u32(svuint32_t op1, svuint32_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_u32,,)(op1, op2);
 }
@@ -70,7 +78,7 @@ svuint32_t test_svbgrp_u32(svuint32_t op1, svuint32_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bgrp.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.*]], <vscale x 2 x i64> [[OP2:%.*]])
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
 //
-svuint64_t test_svbgrp_u64(svuint64_t op1, svuint64_t op2)
+svuint64_t test_svbgrp_u64(svuint64_t op1, svuint64_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_u64,,)(op1, op2);
 }
@@ -89,7 +97,7 @@ svuint64_t test_svbgrp_u64(svuint64_t op1, svuint64_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bgrp.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.*]], <vscale x 16 x i8> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 16 x i8> [[TMP0]]
 //
-svuint8_t test_svbgrp_n_u8(svuint8_t op1, uint8_t op2)
+svuint8_t test_svbgrp_n_u8(svuint8_t op1, uint8_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_n_u8,,)(op1, op2);
 }
@@ -108,7 +116,7 @@ svuint8_t test_svbgrp_n_u8(svuint8_t op1, uint8_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bgrp.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.*]], <vscale x 8 x i16> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 8 x i16> [[TMP0]]
 //
-svuint16_t test_svbgrp_n_u16(svuint16_t op1, uint16_t op2)
+svuint16_t test_svbgrp_n_u16(svuint16_t op1, uint16_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_n_u16,,)(op1, op2);
 }
@@ -127,7 +135,7 @@ svuint16_t test_svbgrp_n_u16(svuint16_t op1, uint16_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bgrp.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.*]], <vscale x 4 x i32> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 4 x i32> [[TMP0]]
 //
-svuint32_t test_svbgrp_n_u32(svuint32_t op1, uint32_t op2)
+svuint32_t test_svbgrp_n_u32(svuint32_t op1, uint32_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_n_u32,,)(op1, op2);
 }
@@ -146,7 +154,7 @@ svuint32_t test_svbgrp_n_u32(svuint32_t op1, uint32_t op2)
 // CPP-CHECK-NEXT:    [[TMP0:%.*]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bgrp.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.*]], <vscale x 2 x i64> [[DOTSPLAT]])
 // CPP-CHECK-NEXT:    ret <vscale x 2 x i64> [[TMP0]]
 //
-svuint64_t test_svbgrp_n_u64(svuint64_t op1, uint64_t op2)
+svuint64_t test_svbgrp_n_u64(svuint64_t op1, uint64_t op2) STREAMING
 {
   return SVE_ACLE_FUNC(svbgrp,_n_u64,,)(op1, op2);
 }
diff --git a/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_aes_bitperm_sha3_sm4.cpp b/clang/test/Sema/aarch64-sve2-intrinsics/acle_sve2_aes_bitperm_sha3_sm4.cpp

Original file line number	Diff line number	Diff line change
`@@ -5,6 +5,8 @@`
`5`	`5`	`// RUN: %clang_cc1 -fclang-abi-compat=latest -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s \| FileCheck %s -check-prefix=CPP-CHECK`
`6`	`6`	`// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - %s \| FileCheck %s`
`7`	`7`	`// RUN: %clang_cc1 -fclang-abi-compat=latest -DSVE_OVERLOADED_FORMS -triple aarch64 -target-feature +sve -target-feature +sve2 -target-feature +sve-bitperm -O1 -Werror -Wall -emit-llvm -o - -x c++ %s \| FileCheck %s -check-prefix=CPP-CHECK`
	`8`	`+// RUN: %clang_cc1 -triple aarch64 -target-feature +sme -target-feature +ssve-bitperm -S -disable-O0-optnone -Werror -Wall -o /dev/null %s`
	`9`	`+// RUN: not %clang_cc1 -triple aarch64 -target-feature +sme -S -disable-O0-optnone -Werror -Wall -o /dev/null %s`
`8`	`10`
`9`	`11`	`#include <arm_sve.h>`
`10`	`12`
`@@ -15,6 +17,12 @@`
`15`	`17`	`#define SVE_ACLE_FUNC(A1,A2,A3,A4) A1##A2##A3##A4`
`16`	`18`	`#endif`
`17`	`19`
	`20`	`+#ifdef __ARM_FEATURE_SME`
	`21`	`+#define STREAMING __arm_streaming`
	`22`	`+#else`
	`23`	`+#define STREAMING`
	`24`	`+#endif`
	`25`	`+`
`18`	`26`	`// CHECK-LABEL: @test_svbdep_u8(`
`19`	`27`	`// CHECK-NEXT: entry:`
`20`	`28`	`// CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bdep.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.]], <vscale x 16 x i8> [[OP2:%.*]])`
`@@ -25,7 +33,7 @@`
`25`	`33`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bdep.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.]], <vscale x 16 x i8> [[OP2:%.*]])`
`26`	`34`	`// CPP-CHECK-NEXT: ret <vscale x 16 x i8> [[TMP0]]`
`27`	`35`	`//`
`28`		`-svuint8_t test_svbdep_u8(svuint8_t op1, svuint8_t op2)`
	`36`	`+svuint8_t test_svbdep_u8(svuint8_t op1, svuint8_t op2) STREAMING`
`29`	`37`	`{`
`30`	`38`	`return SVE_ACLE_FUNC(svbdep,_u8,,)(op1, op2);`
`31`	`39`	`}`
`@@ -40,7 +48,7 @@ svuint8_t test_svbdep_u8(svuint8_t op1, svuint8_t op2)`
`40`	`48`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bdep.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.]], <vscale x 8 x i16> [[OP2:%.*]])`
`41`	`49`	`// CPP-CHECK-NEXT: ret <vscale x 8 x i16> [[TMP0]]`
`42`	`50`	`//`
`43`		`-svuint16_t test_svbdep_u16(svuint16_t op1, svuint16_t op2)`
	`51`	`+svuint16_t test_svbdep_u16(svuint16_t op1, svuint16_t op2) STREAMING`
`44`	`52`	`{`
`45`	`53`	`return SVE_ACLE_FUNC(svbdep,_u16,,)(op1, op2);`
`46`	`54`	`}`
`@@ -55,7 +63,7 @@ svuint16_t test_svbdep_u16(svuint16_t op1, svuint16_t op2)`
`55`	`63`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bdep.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.]], <vscale x 4 x i32> [[OP2:%.*]])`
`56`	`64`	`// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]`
`57`	`65`	`//`
`58`		`-svuint32_t test_svbdep_u32(svuint32_t op1, svuint32_t op2)`
	`66`	`+svuint32_t test_svbdep_u32(svuint32_t op1, svuint32_t op2) STREAMING`
`59`	`67`	`{`
`60`	`68`	`return SVE_ACLE_FUNC(svbdep,_u32,,)(op1, op2);`
`61`	`69`	`}`
`@@ -70,7 +78,7 @@ svuint32_t test_svbdep_u32(svuint32_t op1, svuint32_t op2)`
`70`	`78`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bdep.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.]], <vscale x 2 x i64> [[OP2:%.*]])`
`71`	`79`	`// CPP-CHECK-NEXT: ret <vscale x 2 x i64> [[TMP0]]`
`72`	`80`	`//`
`73`		`-svuint64_t test_svbdep_u64(svuint64_t op1, svuint64_t op2)`
	`81`	`+svuint64_t test_svbdep_u64(svuint64_t op1, svuint64_t op2) STREAMING`
`74`	`82`	`{`
`75`	`83`	`return SVE_ACLE_FUNC(svbdep,_u64,,)(op1, op2);`
`76`	`84`	`}`
`@@ -89,7 +97,7 @@ svuint64_t test_svbdep_u64(svuint64_t op1, svuint64_t op2)`
`89`	`97`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 16 x i8> @llvm.aarch64.sve.bdep.x.nxv16i8(<vscale x 16 x i8> [[OP1:%.]], <vscale x 16 x i8> [[DOTSPLAT]])`
`90`	`98`	`// CPP-CHECK-NEXT: ret <vscale x 16 x i8> [[TMP0]]`
`91`	`99`	`//`
`92`		`-svuint8_t test_svbdep_n_u8(svuint8_t op1, uint8_t op2)`
	`100`	`+svuint8_t test_svbdep_n_u8(svuint8_t op1, uint8_t op2) STREAMING`
`93`	`101`	`{`
`94`	`102`	`return SVE_ACLE_FUNC(svbdep,_n_u8,,)(op1, op2);`
`95`	`103`	`}`
`@@ -108,7 +116,7 @@ svuint8_t test_svbdep_n_u8(svuint8_t op1, uint8_t op2)`
`108`	`116`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 8 x i16> @llvm.aarch64.sve.bdep.x.nxv8i16(<vscale x 8 x i16> [[OP1:%.]], <vscale x 8 x i16> [[DOTSPLAT]])`
`109`	`117`	`// CPP-CHECK-NEXT: ret <vscale x 8 x i16> [[TMP0]]`
`110`	`118`	`//`
`111`		`-svuint16_t test_svbdep_n_u16(svuint16_t op1, uint16_t op2)`
	`119`	`+svuint16_t test_svbdep_n_u16(svuint16_t op1, uint16_t op2) STREAMING`
`112`	`120`	`{`
`113`	`121`	`return SVE_ACLE_FUNC(svbdep,_n_u16,,)(op1, op2);`
`114`	`122`	`}`
`@@ -127,7 +135,7 @@ svuint16_t test_svbdep_n_u16(svuint16_t op1, uint16_t op2)`
`127`	`135`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 4 x i32> @llvm.aarch64.sve.bdep.x.nxv4i32(<vscale x 4 x i32> [[OP1:%.]], <vscale x 4 x i32> [[DOTSPLAT]])`
`128`	`136`	`// CPP-CHECK-NEXT: ret <vscale x 4 x i32> [[TMP0]]`
`129`	`137`	`//`
`130`		`-svuint32_t test_svbdep_n_u32(svuint32_t op1, uint32_t op2)`
	`138`	`+svuint32_t test_svbdep_n_u32(svuint32_t op1, uint32_t op2) STREAMING`
`131`	`139`	`{`
`132`	`140`	`return SVE_ACLE_FUNC(svbdep,_n_u32,,)(op1, op2);`
`133`	`141`	`}`
`@@ -146,7 +154,7 @@ svuint32_t test_svbdep_n_u32(svuint32_t op1, uint32_t op2)`
`146`	`154`	`// CPP-CHECK-NEXT: [[TMP0:%.]] = tail call <vscale x 2 x i64> @llvm.aarch64.sve.bdep.x.nxv2i64(<vscale x 2 x i64> [[OP1:%.]], <vscale x 2 x i64> [[DOTSPLAT]])`
`147`	`155`	`// CPP-CHECK-NEXT: ret <vscale x 2 x i64> [[TMP0]]`
`148`	`156`	`//`
`149`		`-svuint64_t test_svbdep_n_u64(svuint64_t op1, uint64_t op2)`
	`157`	`+svuint64_t test_svbdep_n_u64(svuint64_t op1, uint64_t op2) STREAMING`
`150`	`158`	`{`
`151`	`159`	`return SVE_ACLE_FUNC(svbdep,_n_u64,,)(op1, op2);`
`152`	`160`	`}`