2014-03-29 23:09:45 +08:00
|
|
|
// RUN: %clang_cc1 -triple arm64-linux-gnu -emit-llvm -o - %s | FileCheck --check-prefix=CHECK --check-prefix=CHECK-LE %s
|
2014-07-23 20:57:31 +08:00
|
|
|
// RUN: %clang_cc1 -triple aarch64_be-linux-gnu -emit-llvm -o - %s | FileCheck --check-prefix=CHECK --check-prefix=CHECK-BE %s
|
2014-03-14 19:51:06 +08:00
|
|
|
|
2013-01-31 20:13:10 +08:00
|
|
|
#include <stdarg.h>
|
|
|
|
|
|
|
|
|
|
// Obviously there's more than one way to implement va_arg. This test should at
|
|
|
|
|
// least prevent unintentional regressions caused by refactoring.
|
|
|
|
|
|
|
|
|
|
va_list the_list;
|
|
|
|
|
|
|
|
|
|
int simple_int(void) {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i32 @simple_int
|
2013-01-31 20:13:10 +08:00
|
|
|
return va_arg(the_list, int);
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
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
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK-BE: [[REG_ADDR_ALIGNED:%[0-9]+]] = getelementptr inbounds i8, i8* [[REG_ADDR]], i64 4
|
|
|
|
|
// CHECK-BE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR_ALIGNED]] to i32*
|
|
|
|
|
// CHECK-LE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i32*
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
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
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
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
|
|
|
// CHECK-BE: [[STACK_ALIGNED:%[a-z_0-9]*]] = getelementptr inbounds i8, i8* [[STACK]], i64 4
|
|
|
|
|
// CHECK-BE: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK_ALIGNED]] to i32*
|
|
|
|
|
// CHECK-LE: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to i32*
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i32* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i32, i32* [[ADDR]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: ret i32 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
__int128 aligned_int(void) {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i128 @aligned_int
|
2013-01-31 20:13:10 +08:00
|
|
|
return va_arg(the_list, __int128);
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[ALIGN_REGOFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_REGOFFS:%[a-z_0-9]+]] = and i32 [[ALIGN_REGOFFS]], -16
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[ALIGNED_REGOFFS]], 16
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
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
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[ALIGNED_REGOFFS]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i128*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint i8* [[STACK]] to i64
|
|
|
|
|
// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to i8*
|
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
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[ALIGNED_STACK_PTR]], i64 16
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[ALIGNED_STACK_PTR]] to i128*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i128* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i128, i128* [[ADDR]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: ret i128 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
struct bigstruct {
|
|
|
|
|
int a[10];
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct bigstruct simple_indirect(void) {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define void @simple_indirect
|
2013-01-31 20:13:10 +08:00
|
|
|
return va_arg(the_list, struct bigstruct);
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK-NOT: and i32
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
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
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.bigstruct**
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK-NOT: and i64
|
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
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.bigstruct**
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.bigstruct** [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: load %struct.bigstruct*, %struct.bigstruct** [[ADDR]]
|
2013-01-31 20:13:10 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
struct aligned_bigstruct {
|
|
|
|
|
float a;
|
|
|
|
|
long double b;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct aligned_bigstruct simple_aligned_indirect(void) {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define void @simple_aligned_indirect
|
2013-01-31 20:13:10 +08:00
|
|
|
return va_arg(the_list, struct aligned_bigstruct);
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
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
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.aligned_bigstruct**
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
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
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.aligned_bigstruct**
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.aligned_bigstruct** [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: load %struct.aligned_bigstruct*, %struct.aligned_bigstruct** [[ADDR]]
|
2013-01-31 20:13:10 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
|
|
double simple_double(void) {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define double @simple_double
|
2013-01-31 20:13:10 +08:00
|
|
|
return va_arg(the_list, double);
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[VR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 4)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[VR_OFFS]], 0
|
2014-03-15 00:30:06 +08:00
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG]]
|
2013-01-31 20:13:10 +08:00
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[VR_OFFS]], 16
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 4)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 2)
|
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
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[VR_OFFS]]
|
|
|
|
|
// CHECK-BE: [[REG_ADDR_ALIGNED:%[a-z_0-9]*]] = getelementptr inbounds i8, i8* [[REG_ADDR]], i64 8
|
|
|
|
|
// CHECK-BE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR_ALIGNED]] to double*
|
|
|
|
|
// CHECK-LE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to double*
|
2014-03-15 00:30:06 +08:00
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
2013-01-31 20:13:10 +08:00
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
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
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to double*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi double* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load double, double* [[ADDR]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: ret double [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
struct hfa {
|
|
|
|
|
float a, b;
|
|
|
|
|
};
|
|
|
|
|
|
|
|
|
|
struct hfa simple_hfa(void) {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define %struct.hfa @simple_hfa
|
2013-01-31 20:13:10 +08:00
|
|
|
return va_arg(the_list, struct hfa);
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[VR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 4)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[VR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[VR_OFFS]], 32
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 4)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 2)
|
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
|
|
|
// CHECK: [[FIRST_REG:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[VR_OFFS]]
|
|
|
|
|
// CHECK-LE: [[EL_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[FIRST_REG]], i64 0
|
|
|
|
|
// CHECK-BE: [[EL_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[FIRST_REG]], i64 12
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EL_TYPED:%[a-z_0-9]+]] = bitcast i8* [[EL_ADDR]] to float*
|
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
|
|
|
// CHECK: [[EL_TMPADDR:%[a-z_0-9]+]] = getelementptr inbounds [2 x float], [2 x float]* %[[TMP_HFA:[a-z_.0-9]+]], i64 0, i64 0
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[EL:%[a-z_0-9]+]] = load float, float* [[EL_TYPED]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: store float [[EL]], float* [[EL_TMPADDR]]
|
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
|
|
|
// CHECK-LE: [[EL_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[FIRST_REG]], i64 16
|
|
|
|
|
// CHECK-BE: [[EL_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[FIRST_REG]], i64 28
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[EL_TYPED:%[a-z_0-9]+]] = bitcast i8* [[EL_ADDR]] to float*
|
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
|
|
|
// CHECK: [[EL_TMPADDR:%[a-z_0-9]+]] = getelementptr inbounds [2 x float], [2 x float]* %[[TMP_HFA]], i64 0, i64 1
|
2015-02-28 05:19:58 +08:00
|
|
|
// CHECK: [[EL:%[a-z_0-9]+]] = load float, float* [[EL_TYPED]]
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: store float [[EL]], float* [[EL_TMPADDR]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast [2 x float]* %[[TMP_HFA]] to %struct.hfa*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
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
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
2015-03-14 02:21:46 +08:00
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
2013-01-31 20:13:10 +08:00
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.hfa*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.hfa* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
2019-05-22 19:42:54 +08:00
|
|
|
// Over and under alignment on fundamental types has no effect on parameter
|
|
|
|
|
// passing, so the code generated for va_arg should be the same as for
|
|
|
|
|
// non-aligned fundamental types.
|
|
|
|
|
|
|
|
|
|
typedef int underaligned_int __attribute__((packed,aligned(2)));
|
|
|
|
|
underaligned_int underaligned_int_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i32 @underaligned_int_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_int);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK-BE: [[REG_ADDR_ALIGNED:%[0-9]+]] = getelementptr inbounds i8, i8* [[REG_ADDR]], i64 4
|
|
|
|
|
// CHECK-BE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR_ALIGNED]] to i32*
|
|
|
|
|
// CHECK-LE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i32*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK-BE: [[STACK_ALIGNED:%[a-z_0-9]*]] = getelementptr inbounds i8, i8* [[STACK]], i64 4
|
|
|
|
|
// CHECK-BE: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK_ALIGNED]] to i32*
|
|
|
|
|
// CHECK-LE: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to i32*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i32* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i32, i32* [[ADDR]]
|
|
|
|
|
// CHECK: ret i32 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef int overaligned_int __attribute__((aligned(32)));
|
|
|
|
|
overaligned_int overaligned_int_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i32 @overaligned_int_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_int);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK-BE: [[REG_ADDR_ALIGNED:%[0-9]+]] = getelementptr inbounds i8, i8* [[REG_ADDR]], i64 4
|
|
|
|
|
// CHECK-BE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR_ALIGNED]] to i32*
|
|
|
|
|
// CHECK-LE: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i32*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK-BE: [[STACK_ALIGNED:%[a-z_0-9]*]] = getelementptr inbounds i8, i8* [[STACK]], i64 4
|
|
|
|
|
// CHECK-BE: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK_ALIGNED]] to i32*
|
|
|
|
|
// CHECK-LE: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to i32*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i32* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i32, i32* [[ADDR]]
|
|
|
|
|
// CHECK: ret i32 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef long long underaligned_long_long __attribute__((packed,aligned(2)));
|
|
|
|
|
underaligned_long_long underaligned_long_long_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i64 @underaligned_long_long_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_long_long);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i64*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to i64*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i64* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i64, i64* [[ADDR]]
|
|
|
|
|
// CHECK: ret i64 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef long long overaligned_long_long __attribute__((aligned(32)));
|
|
|
|
|
overaligned_long_long overaligned_long_long_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i64 @overaligned_long_long_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_long_long);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i64*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to i64*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i64* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i64, i64* [[ADDR]]
|
|
|
|
|
// CHECK: ret i64 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef __int128 underaligned_int128 __attribute__((packed,aligned(2)));
|
|
|
|
|
underaligned_int128 underaligned_int128_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i128 @underaligned_int128_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_int128);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[ALIGN_REGOFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_REGOFFS:%[a-z_0-9]+]] = and i32 [[ALIGN_REGOFFS]], -16
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[ALIGNED_REGOFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[ALIGNED_REGOFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i128*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint i8* [[STACK]] to i64
|
|
|
|
|
// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to i8*
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[ALIGNED_STACK_PTR]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[ALIGNED_STACK_PTR]] to i128*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i128* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i128, i128* [[ADDR]]
|
|
|
|
|
// CHECK: ret i128 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef __int128 overaligned_int128 __attribute__((aligned(32)));
|
|
|
|
|
overaligned_int128 overaligned_int128_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i128 @overaligned_int128_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_int128);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[ALIGN_REGOFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_REGOFFS:%[a-z_0-9]+]] = and i32 [[ALIGN_REGOFFS]], -16
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[ALIGNED_REGOFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[ALIGNED_REGOFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to i128*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint i8* [[STACK]] to i64
|
|
|
|
|
// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to i8*
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[ALIGNED_STACK_PTR]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[ALIGNED_STACK_PTR]] to i128*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi i128* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
// CHECK: [[RESULT:%[a-z_0-9]+]] = load i128, i128* [[ADDR]]
|
|
|
|
|
// CHECK: ret i128 [[RESULT]]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// The way that attributes applied to a struct change parameter passing is a
|
|
|
|
|
// little strange, in that the alignment due to attributes is used when
|
|
|
|
|
// calculating the size of the struct, but the alignment is based only on the
|
|
|
|
|
// alignment of the members (which can be affected by attributes). What this
|
|
|
|
|
// means is:
|
|
|
|
|
// * The only effect of the aligned attribute on a struct is to increase its
|
|
|
|
|
// size if the alignment is greater than the member alignment.
|
|
|
|
|
// * The packed attribute is considered as applying to the members, so it will
|
|
|
|
|
// affect the alignment.
|
|
|
|
|
// Additionally the alignment can't go below 8 or above 16, so it's only
|
|
|
|
|
// __int128 that can be affected by a change in alignment.
|
|
|
|
|
|
|
|
|
|
typedef struct __attribute__((packed,aligned(2))) {
|
|
|
|
|
int val;
|
|
|
|
|
} underaligned_int_struct;
|
|
|
|
|
underaligned_int_struct underaligned_int_struct_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i64 @underaligned_int_struct_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_int_struct);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.underaligned_int_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.underaligned_int_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.underaligned_int_struct* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct __attribute__((aligned(16))) {
|
|
|
|
|
int val;
|
|
|
|
|
} overaligned_int_struct;
|
|
|
|
|
overaligned_int_struct overaligned_int_struct_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i128 @overaligned_int_struct_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_int_struct);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.overaligned_int_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.overaligned_int_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.overaligned_int_struct* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct __attribute__((packed,aligned(2))) {
|
|
|
|
|
long long val;
|
|
|
|
|
} underaligned_long_long_struct;
|
|
|
|
|
underaligned_long_long_struct underaligned_long_long_struct_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i64 @underaligned_long_long_struct_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_long_long_struct);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.underaligned_long_long_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.underaligned_long_long_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.underaligned_long_long_struct* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct __attribute__((aligned(16))) {
|
|
|
|
|
long long val;
|
|
|
|
|
} overaligned_long_long_struct;
|
|
|
|
|
overaligned_long_long_struct overaligned_long_long_struct_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i128 @overaligned_long_long_struct_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_long_long_struct);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.overaligned_long_long_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.overaligned_long_long_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.overaligned_long_long_struct* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct __attribute__((packed,aligned(2))) {
|
|
|
|
|
__int128 val;
|
|
|
|
|
} underaligned_int128_struct;
|
|
|
|
|
underaligned_int128_struct underaligned_int128_struct_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define [2 x i64] @underaligned_int128_struct_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_int128_struct);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.underaligned_int128_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.underaligned_int128_struct*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.underaligned_int128_struct* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Overaligning to 32 bytes causes it to be passed indirectly via a pointer
|
|
|
|
|
typedef struct __attribute__((aligned(32))) {
|
|
|
|
|
__int128 val;
|
|
|
|
|
} overaligned_int128_struct;
|
|
|
|
|
overaligned_int128_struct overaligned_int128_struct_test() {
|
2020-09-29 21:33:55 +08:00
|
|
|
// CHECK-LABEL: define void @overaligned_int128_struct_test(%struct.overaligned_int128_struct* noalias sret(%struct.overaligned_int128_struct) align 32 %agg.result)
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_int128_struct);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.overaligned_int128_struct**
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.overaligned_int128_struct**
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.overaligned_int128_struct** [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Overaligning or underaligning a struct member changes both its alignment and
|
|
|
|
|
// size when passed as an argument.
|
|
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
int val __attribute__((packed,aligned(2)));
|
|
|
|
|
} underaligned_int_struct_member;
|
|
|
|
|
underaligned_int_struct_member underaligned_int_struct_member_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i64 @underaligned_int_struct_member_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_int_struct_member);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.underaligned_int_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.underaligned_int_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.underaligned_int_struct_member* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
int val __attribute__((aligned(16)));
|
|
|
|
|
} overaligned_int_struct_member;
|
|
|
|
|
overaligned_int_struct_member overaligned_int_struct_member_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i128 @overaligned_int_struct_member_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_int_struct_member);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[ALIGN_REGOFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_REGOFFS:%[a-z_0-9]+]] = and i32 [[ALIGN_REGOFFS]], -16
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[ALIGNED_REGOFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[ALIGNED_REGOFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.overaligned_int_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint i8* [[STACK]] to i64
|
|
|
|
|
// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to i8*
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[ALIGNED_STACK_PTR]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[ALIGNED_STACK_PTR]] to %struct.overaligned_int_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.overaligned_int_struct_member* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
long long val __attribute__((packed,aligned(2)));
|
|
|
|
|
} underaligned_long_long_struct_member;
|
|
|
|
|
underaligned_long_long_struct_member underaligned_long_long_struct_member_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i64 @underaligned_long_long_struct_member_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_long_long_struct_member);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.underaligned_long_long_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.underaligned_long_long_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.underaligned_long_long_struct_member* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
long long val __attribute__((aligned(16)));
|
|
|
|
|
} overaligned_long_long_struct_member;
|
|
|
|
|
overaligned_long_long_struct_member overaligned_long_long_struct_member_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define i128 @overaligned_long_long_struct_member_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_long_long_struct_member);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[ALIGN_REGOFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_REGOFFS:%[a-z_0-9]+]] = and i32 [[ALIGN_REGOFFS]], -16
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[ALIGNED_REGOFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[ALIGNED_REGOFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.overaligned_long_long_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[STACKINT:%[a-z_0-9]+]] = ptrtoint i8* [[STACK]] to i64
|
|
|
|
|
// CHECK: [[ALIGN_STACK:%[a-z_0-9]+]] = add i64 [[STACKINT]], 15
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_INT:%[a-z_0-9]+]] = and i64 [[ALIGN_STACK]], -16
|
|
|
|
|
// CHECK: [[ALIGNED_STACK_PTR:%[a-z_0-9]+]] = inttoptr i64 [[ALIGNED_STACK_INT]] to i8*
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[ALIGNED_STACK_PTR]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[ALIGNED_STACK_PTR]] to %struct.overaligned_long_long_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.overaligned_long_long_struct_member* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
typedef struct {
|
|
|
|
|
__int128 val __attribute__((packed,aligned(2)));
|
|
|
|
|
} underaligned_int128_struct_member;
|
|
|
|
|
underaligned_int128_struct_member underaligned_int128_struct_member_test() {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define [2 x i64] @underaligned_int128_struct_member_test()
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, underaligned_int128_struct_member);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 16
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.underaligned_int128_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 16
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.underaligned_int128_struct_member*
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.underaligned_int128_struct_member* [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
// Overaligning to 32 bytes causes it to be passed indirectly via a pointer
|
|
|
|
|
typedef struct {
|
|
|
|
|
__int128 val __attribute__((aligned(32)));
|
|
|
|
|
} overaligned_int128_struct_member;
|
|
|
|
|
overaligned_int128_struct_member overaligned_int128_struct_member_test() {
|
2020-09-29 21:33:55 +08:00
|
|
|
// CHECK-LABEL: define void @overaligned_int128_struct_member_test(%struct.overaligned_int128_struct_member* noalias sret(%struct.overaligned_int128_struct_member) align 32 %agg.result)
|
2019-05-22 19:42:54 +08:00
|
|
|
return va_arg(the_list, overaligned_int128_struct_member);
|
|
|
|
|
// CHECK: [[GR_OFFS:%[a-z_0-9]+]] = load i32, i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[EARLY_ONSTACK:%[a-z_0-9]+]] = icmp sge i32 [[GR_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[EARLY_ONSTACK]], label %[[VAARG_ON_STACK:[a-z_.0-9]+]], label %[[VAARG_MAYBE_REG:[a-z_.0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_MAYBE_REG]]
|
|
|
|
|
// CHECK: [[NEW_REG_OFFS:%[a-z_0-9]+]] = add i32 [[GR_OFFS]], 8
|
|
|
|
|
// CHECK: store i32 [[NEW_REG_OFFS]], i32* getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 3)
|
|
|
|
|
// CHECK: [[INREG:%[a-z_0-9]+]] = icmp sle i32 [[NEW_REG_OFFS]], 0
|
|
|
|
|
// CHECK: br i1 [[INREG]], label %[[VAARG_IN_REG:[a-z_.0-9]+]], label %[[VAARG_ON_STACK]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_IN_REG]]
|
|
|
|
|
// CHECK: [[REG_TOP:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 1)
|
|
|
|
|
// CHECK: [[REG_ADDR:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[REG_TOP]], i32 [[GR_OFFS]]
|
|
|
|
|
// CHECK: [[FROMREG_ADDR:%[a-z_0-9]+]] = bitcast i8* [[REG_ADDR]] to %struct.overaligned_int128_struct_member**
|
|
|
|
|
// CHECK: br label %[[VAARG_END:[a-z._0-9]+]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_ON_STACK]]
|
|
|
|
|
// CHECK: [[STACK:%[a-z_0-9]+]] = load i8*, i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[NEW_STACK:%[a-z_0-9]+]] = getelementptr inbounds i8, i8* [[STACK]], i64 8
|
|
|
|
|
// CHECK: store i8* [[NEW_STACK]], i8** getelementptr inbounds (%struct.__va_list, %struct.__va_list* @the_list, i32 0, i32 0)
|
|
|
|
|
// CHECK: [[FROMSTACK_ADDR:%[a-z_0-9]+]] = bitcast i8* [[STACK]] to %struct.overaligned_int128_struct_member**
|
|
|
|
|
// CHECK: br label %[[VAARG_END]]
|
|
|
|
|
|
|
|
|
|
// CHECK: [[VAARG_END]]
|
|
|
|
|
// CHECK: [[ADDR:%[a-z._0-9]+]] = phi %struct.overaligned_int128_struct_member** [ [[FROMREG_ADDR]], %[[VAARG_IN_REG]] ], [ [[FROMSTACK_ADDR]], %[[VAARG_ON_STACK]] ]
|
|
|
|
|
}
|
|
|
|
|
|
2013-01-31 20:13:10 +08:00
|
|
|
void check_start(int n, ...) {
|
2020-02-04 02:09:39 +08:00
|
|
|
// CHECK-LABEL: define void @check_start(i32 %n, ...)
|
2013-01-31 20:13:10 +08:00
|
|
|
|
|
|
|
|
va_list the_list;
|
|
|
|
|
va_start(the_list, n);
|
|
|
|
|
// CHECK: [[THE_LIST:%[a-z_0-9]+]] = alloca %struct.__va_list
|
|
|
|
|
// CHECK: [[VOIDP_THE_LIST:%[a-z_0-9]+]] = bitcast %struct.__va_list* [[THE_LIST]] to i8*
|
|
|
|
|
// CHECK: call void @llvm.va_start(i8* [[VOIDP_THE_LIST]])
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|