llvm-project/clang/test/CodeGen/arm-arguments.c

221 lines
7.3 KiB
C
Raw Normal View History

// REQUIRES: arm-registered-target
// RUN: %clang_cc1 -triple armv7-apple-darwin9 -target-feature +neon -target-abi apcs-gnu -emit-llvm -w -o - %s | FileCheck -check-prefix=APCS-GNU %s
// RUN: %clang_cc1 -triple armv7-apple-darwin9 -target-feature +neon -target-abi aapcs -emit-llvm -w -o - %s | FileCheck -check-prefix=AAPCS %s
// APCS-GNU-LABEL: define signext i8 @f0()
// AAPCS-LABEL: define arm_aapcscc signext i8 @f0()
char f0(void) {
return 0;
}
// APCS-GNU-LABEL: define i8 @f1()
// AAPCS-LABEL: define arm_aapcscc i8 @f1()
struct s1 { char f0; };
struct s1 f1(void) {}
// APCS-GNU-LABEL: define i16 @f2()
// AAPCS-LABEL: define arm_aapcscc i16 @f2()
struct s2 { short f0; };
struct s2 f2(void) {}
// APCS-GNU-LABEL: define i32 @f3()
// AAPCS-LABEL: define arm_aapcscc i32 @f3()
struct s3 { int f0; };
struct s3 f3(void) {}
// APCS-GNU-LABEL: define i32 @f4()
// AAPCS-LABEL: define arm_aapcscc i32 @f4()
struct s4 { struct s4_0 { int f0; } f0; };
struct s4 f4(void) {}
// APCS-GNU-LABEL: define void @f5(
// APCS-GNU: struct.s5* noalias sret
// AAPCS-LABEL: define arm_aapcscc i32 @f5()
struct s5 { struct { } f0; int f1; };
struct s5 f5(void) {}
// APCS-GNU-LABEL: define void @f6(
// APCS-GNU: struct.s6* noalias sret
// AAPCS-LABEL: define arm_aapcscc i32 @f6()
struct s6 { int f0[1]; };
struct s6 f6(void) {}
// APCS-GNU-LABEL: define void @f7()
// AAPCS-LABEL: define arm_aapcscc void @f7()
struct s7 { struct { int : 0; } f0; };
struct s7 f7(void) {}
// APCS-GNU-LABEL: define void @f8(
// APCS-GNU: struct.s8* noalias sret
// AAPCS-LABEL: define arm_aapcscc void @f8()
struct s8 { struct { int : 0; } f0[1]; };
struct s8 f8(void) {}
// APCS-GNU-LABEL: define i32 @f9()
// AAPCS-LABEL: define arm_aapcscc i32 @f9()
struct s9 { int f0; int : 0; };
struct s9 f9(void) {}
// APCS-GNU-LABEL: define i32 @f10()
// AAPCS-LABEL: define arm_aapcscc i32 @f10()
struct s10 { int f0; int : 0; int : 0; };
struct s10 f10(void) {}
// APCS-GNU-LABEL: define void @f11(
// APCS-GNU: struct.s11* noalias sret
// AAPCS-LABEL: define arm_aapcscc i32 @f11()
struct s11 { int : 0; int f0; };
struct s11 f11(void) {}
// APCS-GNU-LABEL: define i32 @f12()
// AAPCS-LABEL: define arm_aapcscc i32 @f12()
union u12 { char f0; short f1; int f2; };
union u12 f12(void) {}
// APCS-GNU-LABEL: define void @f13(
// APCS-GNU: struct.s13* noalias sret
// FIXME: This should return a float.
// AAPCS-FIXME: darm_aapcscc efine float @f13()
struct s13 { float f0; };
struct s13 f13(void) {}
// APCS-GNU-LABEL: define void @f14(
// APCS-GNU: union.u14* noalias sret
// AAPCS-LABEL: define arm_aapcscc i32 @f14()
union u14 { float f0; };
union u14 f14(void) {}
// APCS-GNU-LABEL: define void @f15()
// AAPCS-LABEL: define arm_aapcscc void @f15()
void f15(struct s7 a0) {}
// APCS-GNU-LABEL: define void @f16()
// AAPCS-LABEL: define arm_aapcscc void @f16()
void f16(struct s8 a0) {}
// APCS-GNU-LABEL: define i32 @f17()
// AAPCS-LABEL: define arm_aapcscc i32 @f17()
struct s17 { short f0 : 13; char f1 : 4; };
struct s17 f17(void) {}
// APCS-GNU-LABEL: define i32 @f18()
// AAPCS-LABEL: define arm_aapcscc i32 @f18()
struct s18 { short f0; char f1 : 4; };
struct s18 f18(void) {}
// APCS-GNU-LABEL: define void @f19(
// APCS-GNU: struct.s19* noalias sret
// AAPCS-LABEL: define arm_aapcscc i32 @f19()
struct s19 { int f0; struct s8 f1; };
struct s19 f19(void) {}
// APCS-GNU-LABEL: define void @f20(
// APCS-GNU: struct.s20* noalias sret
// AAPCS-LABEL: define arm_aapcscc i32 @f20()
struct s20 { struct s8 f1; int f0; };
struct s20 f20(void) {}
// APCS-GNU-LABEL: define i8 @f21()
// AAPCS-LABEL: define arm_aapcscc i32 @f21()
struct s21 { struct {} f1; int f0 : 4; };
struct s21 f21(void) {}
// APCS-GNU-LABEL: define i16 @f22()
// APCS-GNU-LABEL: define i32 @f23()
// APCS-GNU-LABEL: define i64 @f24()
// APCS-GNU-LABEL: define i128 @f25()
// APCS-GNU-LABEL: define i64 @f26()
// APCS-GNU-LABEL: define i128 @f27()
// AAPCS-LABEL: define arm_aapcscc i16 @f22()
// AAPCS-LABEL: define arm_aapcscc i32 @f23()
// AAPCS: define arm_aapcscc void @f24({{.*}} noalias sret
// AAPCS: define arm_aapcscc void @f25({{.*}} noalias sret
// AAPCS: define arm_aapcscc void @f26({{.*}} noalias sret
// AAPCS: define arm_aapcscc void @f27({{.*}} noalias sret
_Complex char f22(void) {}
_Complex short f23(void) {}
_Complex int f24(void) {}
_Complex long long f25(void) {}
_Complex float f26(void) {}
_Complex double f27(void) {}
// APCS-GNU-LABEL: define i16 @f28()
// AAPCS-LABEL: define arm_aapcscc i16 @f28()
struct s28 { _Complex char f0; };
struct s28 f28() {}
// APCS-GNU-LABEL: define i32 @f29()
// AAPCS-LABEL: define arm_aapcscc i32 @f29()
struct s29 { _Complex short f0; };
struct s29 f29() {}
// APCS-GNU: define void @f30({{.*}} noalias sret
// AAPCS: define arm_aapcscc void @f30({{.*}} noalias sret
struct s30 { _Complex int f0; };
struct s30 f30() {}
// PR11905
struct s31 { char x; };
void f31(struct s31 s) { }
// AAPCS: @f31([1 x i32] %s.coerce)
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
// AAPCS: %s = alloca %struct.s31, align 1
// AAPCS: [[TEMP:%.*]] = alloca [1 x i32], align 4
// AAPCS: store [1 x i32] %s.coerce, [1 x i32]* [[TEMP]], align 4
// APCS-GNU: @f31([1 x i32] %s.coerce)
Compute and preserve alignment more faithfully in IR-generation. Introduce an Address type to bundle a pointer value with an alignment. Introduce APIs on CGBuilderTy to work with Address values. Change core APIs on CGF/CGM to traffic in Address where appropriate. Require alignments to be non-zero. Update a ton of code to compute and propagate alignment information. As part of this, I've promoted CGBuiltin's EmitPointerWithAlignment helper function to CGF and made use of it in a number of places in the expression emitter. The end result is that we should now be significantly more correct when performing operations on objects that are locally known to be under-aligned. Since alignment is not reliably tracked in the type system, there are inherent limits to this, but at least we are no longer confused by standard operations like derived-to-base conversions and array-to-pointer decay. I've also fixed a large number of bugs where we were applying the complete-object alignment to a pointer instead of the non-virtual alignment, although most of these were hidden by the very conservative approach we took with member alignment. Also, because IRGen now reliably asserts on zero alignments, we should no longer be subject to an absurd but frustrating recurring bug where an incomplete type would report a zero alignment and then we'd naively do a alignmentAtOffset on it and emit code using an alignment equal to the largest power-of-two factor of the offset. We should also now be emitting much more aggressive alignment attributes in the presence of over-alignment. In particular, field access now uses alignmentAtOffset instead of min. Several times in this patch, I had to change the existing code-generation pattern in order to more effectively use the Address APIs. For the most part, this seems to be a strict improvement, like doing pointer arithmetic with GEPs instead of ptrtoint. That said, I've tried very hard to not change semantics, but it is likely that I've failed in a few places, for which I apologize. ABIArgInfo now always carries the assumed alignment of indirect and indirect byval arguments. In order to cut down on what was already a dauntingly large patch, I changed the code to never set align attributes in the IR on non-byval indirect arguments. That is, we still generate code which assumes that indirect arguments have the given alignment, but we don't express this information to the backend except where it's semantically required (i.e. on byvals). This is likely a minor regression for those targets that did provide this information, but it'll be trivial to add it back in a later patch. I partially punted on applying this work to CGBuiltin. Please do not add more uses of the CreateDefaultAligned{Load,Store} APIs; they will be going away eventually. llvm-svn: 246985
2015-09-08 16:05:57 +08:00
// APCS-GNU: %s = alloca %struct.s31, align 1
// APCS-GNU: [[TEMP:%.*]] = alloca [1 x i32], align 4
// APCS-GNU: store [1 x i32] %s.coerce, [1 x i32]* [[TEMP]], align 4
// PR13562
struct s32 { double x; };
void f32(struct s32 s) { }
// AAPCS: @f32([1 x i64] %s.coerce)
// APCS-GNU: @f32([2 x i32] %s.coerce)
// PR13350
struct s33 { char buf[32*32]; };
void f33(struct s33 s) { }
// APCS-GNU-LABEL: define void @f33(%struct.s33* byval align 4 %s)
// AAPCS-LABEL: define arm_aapcscc void @f33(%struct.s33* byval align 4 %s)
// PR14048
struct s34 { char c; };
void f34(struct s34 s);
void g34(struct s34 *s) { f34(*s); }
// AAPCS: @g34(%struct.s34* %s)
// AAPCS: %[[a:.*]] = alloca [1 x i32]
// AAPCS: load [1 x i32], [1 x i32]* %[[a]]
// rdar://12596507
struct s35
{
float v[18]; //make sure byval is on.
} __attribute__((aligned(16)));
typedef struct s35 s35_with_align;
typedef __attribute__((neon_vector_type(4))) float float32x4_t;
static __attribute__((__always_inline__, __nodebug__)) float32x4_t vaddq_f32(
float32x4_t __a, float32x4_t __b) {
return __a + __b;
}
float32x4_t f35(int i, s35_with_align s1, s35_with_align s2) {
float32x4_t v = vaddq_f32(*(float32x4_t *)&s1,
*(float32x4_t *)&s2);
return v;
}
// APCS-GNU-LABEL: define <4 x float> @f35(i32 %i, %struct.s35* byval align 4, %struct.s35* byval align 4)
// APCS-GNU: %[[a:.*]] = alloca %struct.s35, align 16
// APCS-GNU: %[[b:.*]] = bitcast %struct.s35* %[[a]] to i8*
// APCS-GNU: %[[c:.*]] = bitcast %struct.s35* %0 to i8*
// APCS-GNU: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align {{[0-9]+}} %[[b]], i8* align {{[0-9]+}} %[[c]]
// APCS-GNU: %[[d:.*]] = bitcast %struct.s35* %[[a]] to <4 x float>*
// APCS-GNU: load <4 x float>, <4 x float>* %[[d]], align 16
// AAPCS-LABEL: define arm_aapcscc <4 x float> @f35(i32 %i, %struct.s35* byval align 8, %struct.s35* byval align 8)
// AAPCS: %[[a:.*]] = alloca %struct.s35, align 16
// AAPCS: %[[b:.*]] = bitcast %struct.s35* %[[a]] to i8*
// AAPCS: %[[c:.*]] = bitcast %struct.s35* %0 to i8*
// AAPCS: call void @llvm.memcpy.p0i8.p0i8.i32(i8* align 16 %[[b]], i8* align 8 %[[c]]
// AAPCS: %[[d:.*]] = bitcast %struct.s35* %[[a]] to <4 x float>*
// AAPCS: load <4 x float>, <4 x float>* %[[d]], align 16