[clang] [MSVC, ARM64] Add _Copy* and _Count* intrinsics (PR #66554)
Amy Huang via cfe-commits
cfe-commits at lists.llvm.org
Fri Sep 15 15:33:49 PDT 2023
https://github.com/amykhuang created https://github.com/llvm/llvm-project/pull/66554
Implement the _Count* and _Copy* Windows ARM intrinsics:
```
double _CopyDoubleFromInt64(__int64)
float _CopyFloatFromInt32(__int32)
__int32 _CopyInt32FromFloat(float)
__int64 _CopyInt64FromDouble(double)
unsigned int _CountLeadingOnes(unsigned long)
unsigned int _CountLeadingOnes64(unsigned __int64)
unsigned int _CountLeadingSigns(long)
unsigned int _CountLeadingSigns64(__int64)
unsigned int _CountLeadingZeros(unsigned long)
unsigned int _CountLeadingZeros64(unsigned __int64)
unsigned int _CountOneBits(unsigned long)
unsigned int _CountOneBits64(unsigned __int64)
```
Full list of intrinsics here: [https://learn.microsoft.com/en-us/cpp/intrinsics/arm64-intrinsics](https://learn.microsoft.com/en-us/cpp/intrinsics/arm64-intrinsics)
Bug: [65405](https://github.com/llvm/llvm-project/issues/65405)
>From bbd538659e363e7e7e57fd5e585bc90c2bc0d88b Mon Sep 17 00:00:00 2001
From: Amy Huang <akhuang at google.com>
Date: Thu, 7 Sep 2023 11:32:17 -0700
Subject: [PATCH] [MSVC, ARM64] Add _Copy* and _Count* intrinsics
---
clang/include/clang/Basic/BuiltinsAArch64.def | 15 +-
clang/lib/CodeGen/CGBuiltin.cpp | 62 ++++++++
clang/lib/Headers/intrin.h | 14 ++
.../test/CodeGen/arm64-microsoft-intrinsics.c | 140 ++++++++++++++++++
4 files changed, 230 insertions(+), 1 deletion(-)
diff --git a/clang/include/clang/Basic/BuiltinsAArch64.def b/clang/include/clang/Basic/BuiltinsAArch64.def
index eaae6c9ad846868..c4d34f8e15e8b8d 100644
--- a/clang/include/clang/Basic/BuiltinsAArch64.def
+++ b/clang/include/clang/Basic/BuiltinsAArch64.def
@@ -259,7 +259,6 @@ TARGET_HEADER_BUILTIN(__umulh, "ULLiULLiULLi", "nh", INTRIN_H, ALL_MS_LANGUAGES,
TARGET_HEADER_BUILTIN(__break, "vi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
-
TARGET_HEADER_BUILTIN(__writex18byte, "vUNiUc", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
TARGET_HEADER_BUILTIN(__writex18word, "vUNiUs", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
TARGET_HEADER_BUILTIN(__writex18dword, "vUNiUNi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
@@ -270,6 +269,20 @@ TARGET_HEADER_BUILTIN(__readx18word, "UsUNi", "nh", INTRIN_H, ALL_MS_LANGUAGES,
TARGET_HEADER_BUILTIN(__readx18dword, "UNiUNi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
TARGET_HEADER_BUILTIN(__readx18qword, "ULLiUNi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CopyDoubleFromInt64, "dSLLi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CopyFloatFromInt32, "fSi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CopyInt32FromFloat, "Sif", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CopyInt64FromDouble, "SLLid", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+
+TARGET_HEADER_BUILTIN(_CountLeadingOnes, "UiULi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CountLeadingOnes64, "UiULLi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CountLeadingSigns, "UiSLi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CountLeadingSigns64, "UiSLLi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CountLeadingZeros, "UiULi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CountLeadingZeros64, "UiULLi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CountOneBits, "UiULi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+TARGET_HEADER_BUILTIN(_CountOneBits64, "UiULLi", "nh", INTRIN_H, ALL_MS_LANGUAGES, "")
+
#undef BUILTIN
#undef LANGBUILTIN
#undef TARGET_BUILTIN
diff --git a/clang/lib/CodeGen/CGBuiltin.cpp b/clang/lib/CodeGen/CGBuiltin.cpp
index 037a2f9f7b15322..08d09c1c3d8e9de 100644
--- a/clang/lib/CodeGen/CGBuiltin.cpp
+++ b/clang/lib/CodeGen/CGBuiltin.cpp
@@ -10752,6 +10752,68 @@ Value *CodeGenFunction::EmitAArch64BuiltinExpr(unsigned BuiltinID,
return Load;
}
+ if (BuiltinID == AArch64::BI_CopyDoubleFromInt64 ||
+ BuiltinID == AArch64::BI_CopyFloatFromInt32 ||
+ BuiltinID == AArch64::BI_CopyInt32FromFloat ||
+ BuiltinID == AArch64::BI_CopyInt64FromDouble) {
+ return EmitScalarExpr(E->getArg(0));
+ }
+
+ if (BuiltinID == AArch64::BI_CountLeadingOnes ||
+ BuiltinID == AArch64::BI_CountLeadingOnes64 ||
+ BuiltinID == AArch64::BI_CountLeadingZeros ||
+ BuiltinID == AArch64::BI_CountLeadingZeros64) {
+ Value *Arg = EmitScalarExpr(E->getArg(0));
+ llvm::Type *ArgType = Arg->getType();
+
+ if (BuiltinID == AArch64::BI_CountLeadingOnes ||
+ BuiltinID == AArch64::BI_CountLeadingOnes64)
+ Arg = Builder.CreateXor(Arg, Constant::getAllOnesValue(ArgType));
+
+ Function *F = CGM.getIntrinsic(Intrinsic::ctlz, ArgType);
+ Value *Result = Builder.CreateCall(F, {Arg, Builder.getInt1(false)});
+
+ if (BuiltinID == AArch64::BI_CountLeadingOnes64 ||
+ BuiltinID == AArch64::BI_CountLeadingZeros64)
+ Result = Builder.CreateTrunc(Result, Builder.getInt32Ty());
+ return Result;
+ }
+
+ if (BuiltinID == AArch64::BI_CountLeadingSigns ||
+ BuiltinID == AArch64::BI_CountLeadingSigns64) {
+ Value *Arg = EmitScalarExpr(E->getArg(0));
+ llvm::Type *ArgType = Arg->getType();
+
+ Function *F;
+ if (BuiltinID == AArch64::BI_CountLeadingSigns) {
+ F = CGM.getIntrinsic(Intrinsic::aarch64_cls);
+ if (ArgType != Builder.getInt32Ty())
+ Arg =
+ Builder.CreateIntCast(Arg, Builder.getInt32Ty(), /*isSigned*/ true);
+ } else {
+ F = CGM.getIntrinsic(Intrinsic::aarch64_cls64);
+ if (ArgType != Builder.getInt64Ty())
+ Arg =
+ Builder.CreateIntCast(Arg, Builder.getInt64Ty(), /*isSigned*/ true);
+ }
+ Value *Result = Builder.CreateCall(F, Arg, "cls");
+ if (BuiltinID == AArch64::BI_CountLeadingSigns64)
+ Result = Builder.CreateTrunc(Result, Builder.getInt32Ty());
+ return Result;
+ }
+
+ if (BuiltinID == AArch64::BI_CountOneBits ||
+ BuiltinID == AArch64::BI_CountOneBits64) {
+ Value *ArgValue = EmitScalarExpr(E->getArg(0));
+ llvm::Type *ArgType = ArgValue->getType();
+ Function *F = CGM.getIntrinsic(Intrinsic::ctpop, ArgType);
+
+ Value *Result = Builder.CreateCall(F, ArgValue);
+ if (BuiltinID == AArch64::BI_CountOneBits64)
+ Result = Builder.CreateTrunc(Result, Builder.getInt32Ty());
+ return Result;
+ }
+
// Handle MSVC intrinsics before argument evaluation to prevent double
// evaluation.
if (std::optional<MSVCIntrin> MsvcIntId =
diff --git a/clang/lib/Headers/intrin.h b/clang/lib/Headers/intrin.h
index de68b07491c6c54..4678c527bfaab5a 100644
--- a/clang/lib/Headers/intrin.h
+++ b/clang/lib/Headers/intrin.h
@@ -572,6 +572,20 @@ unsigned char __readx18byte(unsigned long offset);
unsigned short __readx18word(unsigned long offset);
unsigned long __readx18dword(unsigned long offset);
unsigned __int64 __readx18qword(unsigned long offset);
+
+double _CopyDoubleFromInt64(__int64);
+float _CopyFloatFromInt32(__int32);
+__int32 _CopyInt32FromFloat(float);
+__int64 _CopyInt64FromDouble(double);
+
+unsigned int _CountLeadingOnes(unsigned long);
+unsigned int _CountLeadingOnes64(unsigned __int64);
+unsigned int _CountLeadingSigns(long);
+unsigned int _CountLeadingSigns64(__int64);
+unsigned int _CountLeadingZeros(unsigned long);
+unsigned int _CountLeadingZeros64(unsigned _int64);
+unsigned int _CountOneBits(unsigned long);
+unsigned int _CountOneBits64(unsigned __int64);
#endif
/*----------------------------------------------------------------------------*\
diff --git a/clang/test/CodeGen/arm64-microsoft-intrinsics.c b/clang/test/CodeGen/arm64-microsoft-intrinsics.c
index bd8e4cb27e52680..e6693f3a4a676f4 100644
--- a/clang/test/CodeGen/arm64-microsoft-intrinsics.c
+++ b/clang/test/CodeGen/arm64-microsoft-intrinsics.c
@@ -265,5 +265,145 @@ unsigned __int64 check__readx18qword(unsigned LONG offset) {
// CHECK-MSCOMPAT: %[[RETVAL:.*]] = load i64, ptr %[[PTR]], align 1
// CHECK-MSCOMPAT: ret i64 %[[RETVAL]]
+double check__CopyDoubleFromInt64(__int64 arg1) {
+ return _CopyDoubleFromInt64(arg1);
+}
+
+// CHECK-MSCOMPAT: %[[RETVAL:.*]] = alloca double, align 8
+// CHECK-MSCOMPAT: %[[ARG:.*]].addr = alloca i64, align 8
+// CHECK-MSCOMPAT: store i64 %[[ARG]], ptr %[[ARG]].addr, align 8
+// CHECK-MSCOMPAT: %[[VAR0:.*]] = load i64, ptr %[[ARG]].addr, align 8
+// CHECK-MSCOMPAT: store i64 %[[VAR0]], ptr %[[RETVAL]], align 8
+// CHECK-MSCOMPAT: %[[VAR1:.*]] = load double, ptr %[[RETVAL]], align 8
+// CHECK-MSCOMPAT: ret double %[[VAR1]]
+// CHECK-LINUX: error: call to undeclared function '_CopyDoubleFromInt64'
+
+float check__CopyFloatFromInt32(__int32 arg1) {
+ return _CopyFloatFromInt32(arg1);
+}
+
+// CHECK-MSCOMPAT: %[[RETVAL:.*]] = alloca float, align 4
+// CHECK-MSCOMPAT: %[[ARG:.*]].addr = alloca i32, align 4
+// CHECK-MSCOMPAT: store i32 %[[ARG]], ptr %[[ARG]].addr, align 4
+// CHECK-MSCOMPAT: %[[VAR0:.*]] = load i32, ptr %[[ARG]].addr, align 4
+// CHECK-MSCOMPAT: store i32 %[[VAR0]], ptr %[[RETVAL]], align 4
+// CHECK-MSCOMPAT: %[[VAR1:.*]] = load float, ptr %[[RETVAL]], align 4
+// CHECK-MSCOMPAT: ret float %[[VAR1]]
+// CHECK-LINUX: error: call to undeclared function '_CopyFloatFromInt32'
+
+__int32 check__CopyInt32FromFloat(float arg1) {
+ return _CopyInt32FromFloat(arg1);
+}
+
+// CHECK-MSCOMPAT: %[[RETVAL:.*]] = alloca i32, align 4
+// CHECK-MSCOMPAT: %[[ARG:.*]].addr = alloca float, align 4
+// CHECK-MSCOMPAT: store float %[[ARG]], ptr %[[ARG]].addr, align 4
+// CHECK-MSCOMPAT: %[[VAR0:.*]] = load float, ptr %[[ARG]].addr, align 4
+// CHECK-MSCOMPAT: store float %[[VAR0]], ptr %[[RETVAL]], align 4
+// CHECK-MSCOMPAT: %[[VAR1:.*]] = load i32, ptr %[[RETVAL]], align 4
+// CHECK-MSCOMPAT: ret i32 %[[VAR1]]
+// CHECK-LINUX: error: call to undeclared function '_CopyInt32FromFloat'
+
+__int64 check__CopyInt64FromDouble(double arg1) {
+ return _CopyInt64FromDouble(arg1);
+}
+
+// CHECK-MSCOMPAT: %[[RETVAL:.*]] = alloca i64, align 8
+// CHECK-MSCOMPAT: %[[ARG:.*]].addr = alloca double, align 8
+// CHECK-MSCOMPAT: store double %[[ARG]], ptr %[[ARG]].addr, align 8
+// CHECK-MSCOMPAT: %[[VAR0:.*]] = load double, ptr %[[ARG]].addr, align 8
+// CHECK-MSCOMPAT: store double %[[VAR0]], ptr %[[RETVAL]], align 8
+// CHECK-MSCOMPAT: %[[VAR1:.*]] = load i64, ptr %[[RETVAL]], align 8
+// CHECK-MSCOMPAT: ret i64 %[[VAR1]]
+// CHECK-LINUX: error: call to undeclared function '_CopyInt64FromDouble'
+
+unsigned int check__CountLeadingOnes(unsigned LONG arg1) {
+ return _CountLeadingOnes(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i32, align 4
+// CHECK-MSVC: store i32 %[[ARG1]], ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[VAR0:.*]] = load i32, ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[VAR1:.*]] = xor i32 %[[VAR0]], -1
+// CHECK-MSVC: %[[VAR2:.*]] = call i32 @llvm.ctlz.i32(i32 %1, i1 false)
+// CHECK-MSVC: ret i32 %[[VAR2]]
+
+unsigned int check__CountLeadingOnes64(unsigned __int64 arg1) {
+ return _CountLeadingOnes64(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i64, align 8
+// CHECK-MSVC: store i64 %[[ARG1]], ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[VAR0:.*]] = load i64, ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[VAR1:.*]] = xor i64 %[[VAR0]], -1
+// CHECK-MSVC: %[[VAR2:.*]] = call i64 @llvm.ctlz.i64(i64 %1, i1 false)
+// CHECK-MSVC: %[[VAR3:.*]] = trunc i64 %2 to i32
+// CHECK-MSVC: ret i32 %[[VAR3]]
+
+unsigned int check__CountLeadingSigns(__int32 arg1) {
+ return _CountLeadingSigns(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i32, align 4
+// CHECK-MSVC: store i32 %[[ARG1]], ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[VAR0:.*]] = load i32, ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[CLS:.*]] = call i32 @llvm.aarch64.cls(i32 %[[VAR0]])
+// CHECK-MSVC: ret i32 %[[CLS]]
+
+unsigned int check__CountLeadingSigns64(__int64 arg1) {
+ return _CountLeadingSigns64(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i64, align 8
+// CHECK-MSVC: store i64 %[[ARG1]], ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[VAR0:.*]] = load i64, ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[CLS:.*]] = call i32 @llvm.aarch64.cls64(i64 %[[VAR0]])
+// CHECK-MSVC: ret i32 %[[CLS]]
+
+unsigned int check__CountLeadingZeros(__int32 arg1) {
+ return _CountLeadingZeros(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i32, align 4
+// CHECK-MSVC: store i32 %[[ARG1]], ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[VAR0:.*]] = load i32, ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[VAR1:.*]] = call i32 @llvm.ctlz.i32(i32 %[[VAR0]], i1 false)
+// CHECK-MSVC: ret i32 %[[VAR1]]
+
+unsigned int check__CountLeadingZeros64(__int64 arg1) {
+ return _CountLeadingZeros64(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i64, align 8
+// CHECK-MSVC: store i64 %[[ARG1]], ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[VAR0:.*]] = load i64, ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[VAR1:.*]] = call i64 @llvm.ctlz.i64(i64 %[[VAR0]], i1 false)
+// CHECK-MSVC: %[[VAR2:.*]] = trunc i64 %[[VAR1]] to i32
+// CHECK-MSVC: ret i32 %[[VAR2]]
+
+unsigned int check_CountOneBits(unsigned LONG arg1) {
+ return _CountOneBits(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i32, align 4
+// CHECK-MSVC: store i32 %[[ARG1]], ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[VAR0:.*]] = load i32, ptr %[[ARG1]].addr, align 4
+// CHECK-MSVC: %[[VAR1:.*]] = call i32 @llvm.ctpop.i32(i32 %0)
+// CHECK-MSVC: ret i32 %[[VAR1]]
+
+unsigned int check_CountOneBits64(unsigned __int64 arg1) {
+ return _CountOneBits64(arg1);
+}
+
+// CHECK-MSVC: %[[ARG1:.*]].addr = alloca i64, align 8
+// CHECK-MSVC: store i64 %[[ARG1]], ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[VAR0:.*]] = load i64, ptr %[[ARG1]].addr, align 8
+// CHECK-MSVC: %[[VAR1:.*]] = call i64 @llvm.ctpop.i64(i64 %0)
+// CHECK-MSVC: %[[VAR2:.*]] = trunc i64 %1 to i32
+// CHECK-MSVC: ret i32 %[[VAR2]]
+
+
// CHECK-MSCOMPAT: ![[MD2]] = !{!"x18"}
// CHECK-MSCOMPAT: ![[MD3]] = !{!"sp"}
+
+
More information about the cfe-commits
mailing list