163 changes: 162 additions & 1 deletion clang/test/CodeGen/builtins-arm-exclusive.c
Original file line number Diff line number Diff line change
@@ -1,5 +1,5 @@
// REQUIRES: arm-registered-target
// RUN: %clang_cc1 -Wall -Werror -triple thumbv7-linux-gnueabi -fno-signed-char -O3 -emit-llvm -o - %s | FileCheck %s
// RUN: %clang_cc1 -Wall -Werror -triple thumbv8-linux-gnueabi -fno-signed-char -O3 -emit-llvm -o - %s | FileCheck %s
// RUN: %clang_cc1 -Wall -Werror -triple arm64-apple-ios7.0 -O3 -emit-llvm -o - %s | FileCheck %s --check-prefix=CHECK-ARM64

// Make sure the canonical use works before going into smaller details:
Expand Down Expand Up @@ -116,6 +116,90 @@ int test_ldrex(char *addr, long long *addr64, float *addrfloat) {
return sum;
}

int test_ldaex(char *addr, long long *addr64, float *addrfloat) {
// CHECK-LABEL: @test_ldaex
// CHECK-ARM64-LABEL: @test_ldaex
int sum = 0;
sum += __builtin_arm_ldaex(addr);
// CHECK: [[INTRES:%.*]] = tail call i32 @llvm.arm.ldaex.p0i8(i8* %addr)
// CHECK: and i32 [[INTRES]], 255

// CHECK-ARM64: [[INTRES:%.*]] = tail call i64 @llvm.aarch64.ldaxr.p0i8(i8* %addr)
// CHECK-ARM64: [[TRUNCRES:%.*]] = trunc i64 [[INTRES]] to i32
// CHECK-ARM64: [[SEXTTMP:%.*]] = shl i32 [[TRUNCRES]], 24
// CHECK-ARM64: ashr exact i32 [[SEXTTMP]], 24

sum += __builtin_arm_ldaex((short *)addr);
// CHECK: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
// CHECK: [[INTRES:%.*]] = tail call i32 @llvm.arm.ldaex.p0i16(i16* [[ADDR16]])
// CHECK: [[TMPSEXT:%.*]] = shl i32 [[INTRES]], 16
// CHECK: ashr exact i32 [[TMPSEXT]], 16

// CHECK-ARM64: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
// CHECK-ARM64: [[INTRES:%.*]] = tail call i64 @llvm.aarch64.ldaxr.p0i16(i16* [[ADDR16]])
// CHECK-ARM64: [[TRUNCRES:%.*]] = trunc i64 [[INTRES]] to i32
// CHECK-ARM64: [[TMPSEXT:%.*]] = shl i32 [[TRUNCRES]], 16
// CHECK-ARM64: ashr exact i32 [[TMPSEXT]], 16

sum += __builtin_arm_ldaex((int *)addr);
// CHECK: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
// CHECK: call i32 @llvm.arm.ldaex.p0i32(i32* [[ADDR32]])

// CHECK-ARM64: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
// CHECK-ARM64: [[INTRES:%.*]] = tail call i64 @llvm.aarch64.ldaxr.p0i32(i32* [[ADDR32]])
// CHECK-ARM64: trunc i64 [[INTRES]] to i32

sum += __builtin_arm_ldaex((long long *)addr);
// CHECK: call { i32, i32 } @llvm.arm.ldaexd(i8* %addr)

// CHECK-ARM64: [[ADDR64:%.*]] = bitcast i8* %addr to i64*
// CHECK-ARM64: call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[ADDR64]])

sum += __builtin_arm_ldaex(addr64);
// CHECK: [[ADDR64_AS8:%.*]] = bitcast i64* %addr64 to i8*
// CHECK: call { i32, i32 } @llvm.arm.ldaexd(i8* [[ADDR64_AS8]])

// CHECK-ARM64: call i64 @llvm.aarch64.ldaxr.p0i64(i64* %addr64)

sum += __builtin_arm_ldaex(addrfloat);
// CHECK: [[INTADDR:%.*]] = bitcast float* %addrfloat to i32*
// CHECK: [[INTRES:%.*]] = tail call i32 @llvm.arm.ldaex.p0i32(i32* [[INTADDR]])
// CHECK: bitcast i32 [[INTRES]] to float

// CHECK-ARM64: [[INTADDR:%.*]] = bitcast float* %addrfloat to i32*
// CHECK-ARM64: [[INTRES:%.*]] = tail call i64 @llvm.aarch64.ldaxr.p0i32(i32* [[INTADDR]])
// CHECK-ARM64: [[TRUNCRES:%.*]] = trunc i64 [[INTRES]] to i32
// CHECK-ARM64: bitcast i32 [[TRUNCRES]] to float

sum += __builtin_arm_ldaex((double *)addr);
// CHECK: [[STRUCTRES:%.*]] = tail call { i32, i32 } @llvm.arm.ldaexd(i8* %addr)
// CHECK: [[RESHI:%.*]] = extractvalue { i32, i32 } [[STRUCTRES]], 1
// CHECK: [[RESLO:%.*]] = extractvalue { i32, i32 } [[STRUCTRES]], 0
// CHECK: [[RESHI64:%.*]] = zext i32 [[RESHI]] to i64
// CHECK: [[RESLO64:%.*]] = zext i32 [[RESLO]] to i64
// CHECK: [[RESHIHI:%.*]] = shl nuw i64 [[RESHI64]], 32
// CHECK: [[INTRES:%.*]] = or i64 [[RESHIHI]], [[RESLO64]]
// CHECK: bitcast i64 [[INTRES]] to double

// CHECK-ARM64: [[INTRES:%.*]] = tail call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[ADDR64]])
// CHECK-ARM64: bitcast i64 [[INTRES]] to double

sum += *__builtin_arm_ldaex((int **)addr);
// CHECK: [[INTRES:%.*]] = tail call i32 @llvm.arm.ldaex.p0i32(i32* [[ADDR32]])
// CHECK: inttoptr i32 [[INTRES]] to i32*

// CHECK-ARM64: [[INTRES:%.*]] = tail call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[ADDR64]])
// CHECK-ARM64: inttoptr i64 [[INTRES]] to i32*

sum += __builtin_arm_ldaex((struct Simple **)addr)->a;
// CHECK: [[INTRES:%.*]] = tail call i32 @llvm.arm.ldaex.p0i32(i32* [[ADDR32]])
// CHECK: inttoptr i32 [[INTRES]] to %struct.Simple*

// CHECK-ARM64: [[INTRES:%.*]] = tail call i64 @llvm.aarch64.ldaxr.p0i64(i64* [[ADDR64]])
// CHECK-ARM64: inttoptr i64 [[INTRES]] to %struct.Simple*
return sum;
}

int test_strex(char *addr) {
// CHECK-LABEL: @test_strex
// CHECK-ARM64-LABEL: @test_strex
Expand Down Expand Up @@ -166,6 +250,56 @@ int test_strex(char *addr) {
return res;
}

int test_stlex(char *addr) {
// CHECK-LABEL: @test_stlex
// CHECK-ARM64-LABEL: @test_stlex
int res = 0;
struct Simple var = {0};
res |= __builtin_arm_stlex(4, addr);
// CHECK: call i32 @llvm.arm.stlex.p0i8(i32 4, i8* %addr)

// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i8(i64 4, i8* %addr)

res |= __builtin_arm_stlex(42, (short *)addr);
// CHECK: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
// CHECK: call i32 @llvm.arm.stlex.p0i16(i32 42, i16* [[ADDR16]])

// CHECK-ARM64: [[ADDR16:%.*]] = bitcast i8* %addr to i16*
// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i16(i64 42, i16* [[ADDR16]])

res |= __builtin_arm_stlex(42, (int *)addr);
// CHECK: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
// CHECK: call i32 @llvm.arm.stlex.p0i32(i32 42, i32* [[ADDR32]])

// CHECK-ARM64: [[ADDR32:%.*]] = bitcast i8* %addr to i32*
// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i32(i64 42, i32* [[ADDR32]])

res |= __builtin_arm_stlex(42, (long long *)addr);
// CHECK: call i32 @llvm.arm.stlexd(i32 42, i32 0, i8* %addr)

// CHECK-ARM64: [[ADDR64:%.*]] = bitcast i8* %addr to i64*
// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 42, i64* [[ADDR64]])

res |= __builtin_arm_stlex(2.71828f, (float *)addr);
// CHECK: call i32 @llvm.arm.stlex.p0i32(i32 1076754509, i32* [[ADDR32]])

// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i32(i64 1076754509, i32* [[ADDR32]])

res |= __builtin_arm_stlex(3.14159, (double *)addr);
// CHECK: call i32 @llvm.arm.stlexd(i32 -266631570, i32 1074340345, i8* %addr)

// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 4614256650576692846, i64* [[ADDR64]])

res |= __builtin_arm_stlex(&var, (struct Simple **)addr);
// CHECK: [[INTVAL:%.*]] = ptrtoint i16* %var to i32
// CHECK: call i32 @llvm.arm.stlex.p0i32(i32 [[INTVAL]], i32* [[ADDR32]])

// CHECK-ARM64: [[INTVAL:%.*]] = ptrtoint i16* %var to i64
// CHECK-ARM64: call i32 @llvm.aarch64.stlxr.p0i64(i64 [[INTVAL]], i64* [[ADDR64]])

return res;
}

void test_clrex() {
// CHECK-LABEL: @test_clrex
// CHECK-ARM64-LABEL: @test_clrex
Expand Down Expand Up @@ -203,4 +337,31 @@ int test_strex_128(__int128 *addr, __int128 val) {
// CHECK-ARM64: [[ADDR8:%.*]] = bitcast i128* %addr to i8*
// CHECK-ARM64: [[RES:%.*]] = tail call i32 @llvm.aarch64.stxp(i64 [[VALLO]], i64 [[VALHI]], i8* [[ADDR8]])
}

__int128 test_ldaex_128(__int128 *addr) {
// CHECK-ARM64-LABEL: @test_ldaex_128

return __builtin_arm_ldaex(addr);
// CHECK-ARM64: [[ADDR8:%.*]] = bitcast i128* %addr to i8*
// CHECK-ARM64: [[STRUCTRES:%.*]] = tail call { i64, i64 } @llvm.aarch64.ldaxp(i8* [[ADDR8]])
// CHECK-ARM64: [[RESHI:%.*]] = extractvalue { i64, i64 } [[STRUCTRES]], 1
// CHECK-ARM64: [[RESLO:%.*]] = extractvalue { i64, i64 } [[STRUCTRES]], 0
// CHECK-ARM64: [[RESHI64:%.*]] = zext i64 [[RESHI]] to i128
// CHECK-ARM64: [[RESLO64:%.*]] = zext i64 [[RESLO]] to i128
// CHECK-ARM64: [[RESHIHI:%.*]] = shl nuw i128 [[RESHI64]], 64
// CHECK-ARM64: [[INTRES:%.*]] = or i128 [[RESHIHI]], [[RESLO64]]
// CHECK-ARM64: ret i128 [[INTRES]]
}

int test_stlex_128(__int128 *addr, __int128 val) {
// CHECK-ARM64-LABEL: @test_stlex_128

return __builtin_arm_stlex(val, addr);
// CHECK-ARM64: [[VALLO:%.*]] = trunc i128 %val to i64
// CHECK-ARM64: [[VALHI128:%.*]] = lshr i128 %val, 64
// CHECK-ARM64: [[VALHI:%.*]] = trunc i128 [[VALHI128]] to i64
// CHECK-ARM64: [[ADDR8:%.*]] = bitcast i128* %addr to i8*
// CHECK-ARM64: [[RES:%.*]] = tail call i32 @llvm.aarch64.stlxp(i64 [[VALLO]], i64 [[VALHI]], i8* [[ADDR8]])
}

#endif
51 changes: 51 additions & 0 deletions clang/test/Sema/builtins-arm-exclusive.c
Original file line number Diff line number Diff line change
Expand Up @@ -55,6 +55,57 @@ int test_strex(char *addr) {
return res;
}

int test_ldaex(char *addr) {
int sum = 0;
sum += __builtin_arm_ldaex(addr);
sum += __builtin_arm_ldaex((short *)addr);
sum += __builtin_arm_ldaex((int *)addr);
sum += __builtin_arm_ldaex((long long *)addr);
sum += __builtin_arm_ldaex((float *)addr);
sum += __builtin_arm_ldaex((double *)addr);
sum += *__builtin_arm_ldaex((int **)addr);
sum += __builtin_arm_ldaex((struct Simple **)addr)->a;
sum += __builtin_arm_ldaex((volatile char *)addr);
sum += __builtin_arm_ldaex((const volatile char *)addr);

// In principle this might be valid, but stick to ints and floats for scalar
// types at the moment.
sum += __builtin_arm_ldaex((struct Simple *)addr).a; // expected-error {{address argument to atomic builtin must be a pointer to}}

sum += __builtin_arm_ldaex((__int128 *)addr); // expected-error {{__int128 is not supported on this target}} expected-error {{address argument to load or store exclusive builtin must be a pointer to 1,2,4 or 8 byte type}}

__builtin_arm_ldaex(); // expected-error {{too few arguments to function call}}
__builtin_arm_ldaex(1, 2); // expected-error {{too many arguments to function call}}
return sum;
}

int test_stlex(char *addr) {
int res = 0;
struct Simple var = {0};
res |= __builtin_arm_stlex(4, addr);
res |= __builtin_arm_stlex(42, (short *)addr);
res |= __builtin_arm_stlex(42, (int *)addr);
res |= __builtin_arm_stlex(42, (long long *)addr);
res |= __builtin_arm_stlex(2.71828f, (float *)addr);
res |= __builtin_arm_stlex(3.14159, (double *)addr);
res |= __builtin_arm_stlex(&var, (struct Simple **)addr);

res |= __builtin_arm_stlex(42, (volatile char *)addr);
res |= __builtin_arm_stlex(42, (char *const)addr);
res |= __builtin_arm_stlex(42, (const char *)addr); // expected-warning {{passing 'const char *' to parameter of type 'volatile char *' discards qualifiers}}


res |= __builtin_arm_stlex(var, (struct Simple *)addr); // expected-error {{address argument to atomic builtin must be a pointer to}}
res |= __builtin_arm_stlex(var, (struct Simple **)addr); // expected-error {{passing 'struct Simple' to parameter of incompatible type 'struct Simple *'}}
res |= __builtin_arm_stlex(&var, (struct Simple **)addr).a; // expected-error {{is not a structure or union}}

res |= __builtin_arm_stlex(1, (__int128 *)addr); // expected-error {{__int128 is not supported on this target}} expected-error {{address argument to load or store exclusive builtin must be a pointer to 1,2,4 or 8 byte type}}

__builtin_arm_stlex(1); // expected-error {{too few arguments to function call}}
__builtin_arm_stlex(1, 2, 3); // expected-error {{too many arguments to function call}}
return res;
}

void test_clrex() {
__builtin_arm_clrex();
__builtin_arm_clrex(1); // expected-error {{too many arguments to function call}}
Expand Down
49 changes: 49 additions & 0 deletions clang/test/Sema/builtins-arm64-exclusive.c
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,55 @@ int test_strex(char *addr) {
return res;
}

int test_ldaex(char *addr) {
int sum = 0;
sum += __builtin_arm_ldaex(addr);
sum += __builtin_arm_ldaex((short *)addr);
sum += __builtin_arm_ldaex((int *)addr);
sum += __builtin_arm_ldaex((long long *)addr);
sum += __builtin_arm_ldaex((__int128 *)addr);
sum += __builtin_arm_ldaex((float *)addr);
sum += __builtin_arm_ldaex((double *)addr);
sum += *__builtin_arm_ldaex((int **)addr);
sum += __builtin_arm_ldaex((struct Simple **)addr)->a;
sum += __builtin_arm_ldaex((volatile char *)addr);
sum += __builtin_arm_ldaex((const volatile char *)addr);

// In principle this might be valid, but stick to ints and floats for scalar
// types at the moment.
sum += __builtin_arm_ldaex((struct Simple *)addr).a; // expected-error {{address argument to atomic builtin must be a pointer to}}

__builtin_arm_ldaex(); // expected-error {{too few arguments to function call}}
__builtin_arm_ldaex(1, 2); // expected-error {{too many arguments to function call}}
return sum;
}

int test_stlex(char *addr) {
int res = 0;
struct Simple var = {0};
res |= __builtin_arm_stlex(4, addr);
res |= __builtin_arm_stlex(42, (short *)addr);
res |= __builtin_arm_stlex(42, (int *)addr);
res |= __builtin_arm_stlex(42, (long long *)addr);
res |= __builtin_arm_stlex(42, (__int128 *)addr);
res |= __builtin_arm_stlex(2.71828f, (float *)addr);
res |= __builtin_arm_stlex(3.14159, (double *)addr);
res |= __builtin_arm_stlex(&var, (struct Simple **)addr);

res |= __builtin_arm_stlex(42, (volatile char *)addr);
res |= __builtin_arm_stlex(42, (char *const)addr);
res |= __builtin_arm_stlex(42, (const char *)addr); // expected-warning {{passing 'const char *' to parameter of type 'volatile char *' discards qualifiers}}


res |= __builtin_arm_stlex(var, (struct Simple *)addr); // expected-error {{address argument to atomic builtin must be a pointer to}}
res |= __builtin_arm_stlex(var, (struct Simple **)addr); // expected-error {{passing 'struct Simple' to parameter of incompatible type 'struct Simple *'}}
res |= __builtin_arm_stlex(&var, (struct Simple **)addr).a; // expected-error {{is not a structure or union}}

__builtin_arm_stlex(1); // expected-error {{too few arguments to function call}}
__builtin_arm_stlex(1, 2, 3); // expected-error {{too many arguments to function call}}
return res;
}

void test_clrex() {
__builtin_arm_clrex();
__builtin_arm_clrex(1); // expected-error {{too many arguments to function call}}
Expand Down