maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/clang/test/CodeGen/arm64-abi-vector.c

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

447 lines
14 KiB
C
Raw Normal View History

ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// RUN: %clang_cc1 -triple arm64-apple-ios7 -target-abi darwinpcs -emit-llvm -o - %s | FileCheck %s
Add support for Android Vector calling convention for AArch64 This modification applies the following Android commit when we have an Android environment. This is the sole non-renderscript in the Android repo commit 9212d4fb30a3ca2f4ee966dd2748c35573d9682c Author: Tim Murray <timmurray@google.com> Date: Fri Aug 15 16:00:15 2014 -0700 Update vector calling convention for AArch64. bug 16846318 Change-Id: I3cfd167758b4bd634d8480ee6ba6bb55d61f82a7 Reviewers: srhines, jyknight Subscribers: mcrosier, aemerson, rengolin, tberghammer, danalbert, srhines Differential Revision: http://reviews.llvm.org/D17448 llvm-svn: 261533
2016-02-23 00:48:42 +08:00
// RUN: %clang_cc1 -triple aarch64-linux-android -emit-llvm -o - %s | FileCheck -check-prefix=ANDROID %s
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
#include <stdarg.h>
Add support for Android Vector calling convention for AArch64 This modification applies the following Android commit when we have an Android environment. This is the sole non-renderscript in the Android repo commit 9212d4fb30a3ca2f4ee966dd2748c35573d9682c Author: Tim Murray <timmurray@google.com> Date: Fri Aug 15 16:00:15 2014 -0700 Update vector calling convention for AArch64. bug 16846318 Change-Id: I3cfd167758b4bd634d8480ee6ba6bb55d61f82a7 Reviewers: srhines, jyknight Subscribers: mcrosier, aemerson, rengolin, tberghammer, danalbert, srhines Differential Revision: http://reviews.llvm.org/D17448 llvm-svn: 261533
2016-02-23 00:48:42 +08:00
typedef __attribute__(( ext_vector_type(2) )) char __char2;
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
typedef __attribute__(( ext_vector_type(3) )) char __char3;
typedef __attribute__(( ext_vector_type(4) )) char __char4;
typedef __attribute__(( ext_vector_type(5) )) char __char5;
typedef __attribute__(( ext_vector_type(9) )) char __char9;
typedef __attribute__(( ext_vector_type(19) )) char __char19;
typedef __attribute__(( ext_vector_type(3) )) short __short3;
typedef __attribute__(( ext_vector_type(5) )) short __short5;
typedef __attribute__(( ext_vector_type(3) )) int __int3;
typedef __attribute__(( ext_vector_type(5) )) int __int5;
typedef __attribute__(( ext_vector_type(3) )) double __double3;
Add support for Android Vector calling convention for AArch64 This modification applies the following Android commit when we have an Android environment. This is the sole non-renderscript in the Android repo commit 9212d4fb30a3ca2f4ee966dd2748c35573d9682c Author: Tim Murray <timmurray@google.com> Date: Fri Aug 15 16:00:15 2014 -0700 Update vector calling convention for AArch64. bug 16846318 Change-Id: I3cfd167758b4bd634d8480ee6ba6bb55d61f82a7 Reviewers: srhines, jyknight Subscribers: mcrosier, aemerson, rengolin, tberghammer, danalbert, srhines Differential Revision: http://reviews.llvm.org/D17448 llvm-svn: 261533
2016-02-23 00:48:42 +08:00
// Passing legal vector types as varargs. Check that we've allocated the appropriate size
double varargs_vec_2c(int fixed, ...) {
// ANDROID: varargs_vec_2c
// ANDROID: [[VAR:%.*]] = alloca <2 x i8>, align 2
// ANDROID: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
// ANDROID: bitcast i8* [[AP_CUR]] to <2 x i8>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__char2 c3 = va_arg(ap, __char2);
sum = sum + c3.x + c3.y;
va_end(ap);
return sum;
}
double test_2c(__char2 *in) {
// ANDROID: call double (i32, ...) @varargs_vec_2c(i32 3, i16 {{%.*}})
return varargs_vec_2c(3, *in);
}
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double varargs_vec_3c(int fixed, ...) {
// CHECK: varargs_vec_3c
// CHECK: alloca <3 x i8>, align 4
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_CUR]] to <3 x i8>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__char3 c3 = va_arg(ap, __char3);
sum = sum + c3.x + c3.y;
va_end(ap);
return sum;
}
double test_3c(__char3 *in) {
// CHECK: test_3c
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_3c(i32 3, i32 {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_3c(3, *in);
}
double varargs_vec_4c(int fixed, ...) {
// CHECK: varargs_vec_4c
// CHECK: alloca <4 x i8>, align 4
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_CUR]] to <4 x i8>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__char4 c4 = va_arg(ap, __char4);
sum = sum + c4.x + c4.y;
va_end(ap);
return sum;
}
double test_4c(__char4 *in) {
// CHECK: test_4c
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_4c(i32 4, i32 {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_4c(4, *in);
}
double varargs_vec_5c(int fixed, ...) {
// CHECK: varargs_vec_5c
// CHECK: alloca <5 x i8>, align 8
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_CUR]] to <5 x i8>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__char5 c5 = va_arg(ap, __char5);
sum = sum + c5.x + c5.y;
va_end(ap);
return sum;
}
double test_5c(__char5 *in) {
// CHECK: test_5c
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_5c(i32 5, <2 x i32> {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_5c(5, *in);
}
double varargs_vec_9c(int fixed, ...) {
// CHECK: varargs_vec_9c
// CHECK: alloca <9 x i8>, align 16
// CHECK: [[ALIGN:%.*]] = and i64 {{%.*}}, -16
// CHECK: [[AP_ALIGN:%.*]] = inttoptr i64 [[ALIGN]] 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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i64 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_ALIGN]] to <9 x i8>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__char9 c9 = va_arg(ap, __char9);
sum = sum + c9.x + c9.y;
va_end(ap);
return sum;
}
double test_9c(__char9 *in) {
// CHECK: test_9c
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_9c(i32 9, <4 x i32> {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_9c(9, *in);
}
double varargs_vec_19c(int fixed, ...) {
// CHECK: varargs_vec_19c
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
// CHECK: [[VAR:%.*]] = bitcast i8* [[AP_CUR]] to <19 x i8>**
// CHECK: [[VAR2:%.*]] = load <19 x i8>*, <19 x i8>** [[VAR]]
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__char19 c19 = va_arg(ap, __char19);
sum = sum + c19.x + c19.y;
va_end(ap);
return sum;
}
double test_19c(__char19 *in) {
// CHECK: test_19c
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_19c(i32 19, <19 x i8>* {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_19c(19, *in);
}
double varargs_vec_3s(int fixed, ...) {
// CHECK: varargs_vec_3s
// CHECK: alloca <3 x i16>, align 8
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_CUR]] to <3 x i16>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__short3 c3 = va_arg(ap, __short3);
sum = sum + c3.x + c3.y;
va_end(ap);
return sum;
}
double test_3s(__short3 *in) {
// CHECK: test_3s
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_3s(i32 3, <2 x i32> {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_3s(3, *in);
}
double varargs_vec_5s(int fixed, ...) {
// CHECK: varargs_vec_5s
// CHECK: alloca <5 x i16>, align 16
// CHECK: [[ALIGN:%.*]] = and i64 {{%.*}}, -16
// CHECK: [[AP_ALIGN:%.*]] = inttoptr i64 [[ALIGN]] 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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i64 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_ALIGN]] to <5 x i16>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__short5 c5 = va_arg(ap, __short5);
sum = sum + c5.x + c5.y;
va_end(ap);
return sum;
}
double test_5s(__short5 *in) {
// CHECK: test_5s
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_5s(i32 5, <4 x i32> {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_5s(5, *in);
}
double varargs_vec_3i(int fixed, ...) {
// CHECK: varargs_vec_3i
// CHECK: alloca <3 x i32>, align 16
// CHECK: [[ALIGN:%.*]] = and i64 {{%.*}}, -16
// CHECK: [[AP_ALIGN:%.*]] = inttoptr i64 [[ALIGN]] 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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i64 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_ALIGN]] to <3 x i32>*
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__int3 c3 = va_arg(ap, __int3);
sum = sum + c3.x + c3.y;
va_end(ap);
return sum;
}
double test_3i(__int3 *in) {
// CHECK: test_3i
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_3i(i32 3, <4 x i32> {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_3i(3, *in);
}
double varargs_vec_5i(int fixed, ...) {
// CHECK: varargs_vec_5i
// CHECK: alloca <5 x i32>, align 16
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
// CHECK: [[VAR:%.*]] = bitcast i8* [[AP_CUR]] to <5 x i32>**
// CHECK: [[VAR2:%.*]] = load <5 x i32>*, <5 x i32>** [[VAR]]
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__int5 c5 = va_arg(ap, __int5);
sum = sum + c5.x + c5.y;
va_end(ap);
return sum;
}
double test_5i(__int5 *in) {
// CHECK: test_5i
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_5i(i32 5, <5 x i32>* {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_5i(5, *in);
}
double varargs_vec_3d(int fixed, ...) {
// CHECK: varargs_vec_3d
// CHECK: alloca <3 x double>, align 16
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
// CHECK: [[VAR:%.*]] = bitcast i8* [[AP_CUR]] to <3 x double>**
// CHECK: [[VAR2:%.*]] = load <3 x double>*, <3 x double>** [[VAR]]
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__double3 c3 = va_arg(ap, __double3);
sum = sum + c3.x + c3.y;
va_end(ap);
return sum;
}
double test_3d(__double3 *in) {
// CHECK: test_3d
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec_3d(i32 3, <3 x double>* {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return varargs_vec_3d(3, *in);
}
double varargs_vec(int fixed, ...) {
// CHECK: varargs_vec
va_list ap;
double sum = fixed;
va_start(ap, fixed);
__char3 c3 = va_arg(ap, __char3);
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_CUR]] to <3 x i8>*
sum = sum + c3.x + c3.y;
__char5 c5 = va_arg(ap, __char5);
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_CUR]] to <5 x i8>*
sum = sum + c5.x + c5.y;
__char9 c9 = va_arg(ap, __char9);
// CHECK: [[ALIGN:%.*]] = and i64 {{%.*}}, -16
// CHECK: [[AP_ALIGN:%.*]] = inttoptr i64 [[ALIGN]] 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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i64 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_ALIGN]] to <9 x i8>*
sum = sum + c9.x + c9.y;
__char19 c19 = va_arg(ap, __char19);
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
// CHECK: [[VAR:%.*]] = bitcast i8* [[AP_CUR]] to <19 x i8>**
// CHECK: [[VAR2:%.*]] = load <19 x i8>*, <19 x i8>** [[VAR]]
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
sum = sum + c19.x + c19.y;
__short3 s3 = va_arg(ap, __short3);
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_CUR]] to <3 x i16>*
sum = sum + s3.x + s3.y;
__short5 s5 = va_arg(ap, __short5);
// CHECK: [[ALIGN:%.*]] = and i64 {{%.*}}, -16
// CHECK: [[AP_ALIGN:%.*]] = inttoptr i64 [[ALIGN]] 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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i64 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_ALIGN]] to <5 x i16>*
sum = sum + s5.x + s5.y;
__int3 i3 = va_arg(ap, __int3);
// CHECK: [[ALIGN:%.*]] = and i64 {{%.*}}, -16
// CHECK: [[AP_ALIGN:%.*]] = inttoptr i64 [[ALIGN]] 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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_ALIGN]], i64 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: bitcast i8* [[AP_ALIGN]] to <3 x i32>*
sum = sum + i3.x + i3.y;
__int5 i5 = va_arg(ap, __int5);
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
// CHECK: [[VAR:%.*]] = bitcast i8* [[AP_CUR]] to <5 x i32>**
// CHECK: [[VAR2:%.*]] = load <5 x i32>*, <5 x i32>** [[VAR]]
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
sum = sum + i5.x + i5.y;
__double3 d3 = va_arg(ap, __double3);
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: [[AP_NEXT:%.*]] = getelementptr inbounds i8, i8* [[AP_CUR:%.*]], i64 8
// CHECK: [[VAR:%.*]] = bitcast i8* [[AP_CUR]] to <3 x double>**
// CHECK: [[VAR2:%.*]] = load <3 x double>*, <3 x double>** [[VAR]]
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
sum = sum + d3.x + d3.y;
va_end(ap);
return sum;
}
double test(__char3 *c3, __char5 *c5, __char9 *c9, __char19 *c19,
__short3 *s3, __short5 *s5, __int3 *i3, __int5 *i5,
__double3 *d3) {
double ret = varargs_vec(3, *c3, *c5, *c9, *c19, *s3, *s5, *i3, *i5, *d3);
[opaque pointer types] Explicit non-pointer type for call expressions (migration for recent LLVM change to textual IR for calls) llvm-svn: 235147
2015-04-17 07:25:00 +08:00
// CHECK: call double (i32, ...) @varargs_vec(i32 3, i32 {{%.*}}, <2 x i32> {{%.*}}, <4 x i32> {{%.*}}, <19 x i8>* {{%.*}}, <2 x i32> {{%.*}}, <4 x i32> {{%.*}}, <4 x i32> {{%.*}}, <5 x i32>* {{%.*}}, <3 x double>* {{%.*}})
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
return ret;
}
__attribute__((noinline)) double args_vec_3c(int fixed, __char3 c3) {
// CHECK: args_vec_3c
// CHECK: [[C3:%.*]] = alloca <3 x i8>, align 4
// CHECK: [[TMP:%.*]] = bitcast <3 x i8>* [[C3]] to i32*
// CHECK: store i32 {{%.*}}, i32* [[TMP]]
double sum = fixed;
sum = sum + c3.x + c3.y;
return sum;
}
double fixed_3c(__char3 *in) {
// CHECK: fixed_3c
// CHECK: call double @args_vec_3c(i32 3, i32 {{%.*}})
return args_vec_3c(3, *in);
}
__attribute__((noinline)) double args_vec_5c(int fixed, __char5 c5) {
// CHECK: args_vec_5c
// CHECK: [[C5:%.*]] = alloca <5 x i8>, align 8
// CHECK: [[TMP:%.*]] = bitcast <5 x i8>* [[C5]] to <2 x i32>*
Respect alignment when loading up a coerced function argument Code in CGCall.cpp that loads up function arguments that need to be coerced to a different type may in some cases ignore the fact that the source of the argument is not naturally aligned. This may cause incorrect code to be generated. In some places in CreateCoercedLoad, we already have setAlignment calls to address this, but I ran into one where it was missing, causing wrong code generation on SystemZ. However, in that location, we do not actually know what alignment of the source location we can rely on; the callers do not pass anything to this routine. This is already an issue in other places in CreateCoercedLoad; and the same problem exists for CreateCoercedStore. To avoid pessimising code, and to fix the FIXMEs already in place, this patch also adds an alignment argument to the CreateCoerced* routines and uses it instead of forcing an alignment of 1. The callers are changed to pass in the best information they have. This actually requires changes in a number of existing test cases since we now get better alignment in many places. Differential Revision: http://reviews.llvm.org/D11033 llvm-svn: 241898
2015-07-10 19:31:43 +08:00
// CHECK: store <2 x i32> {{%.*}}, <2 x i32>* [[TMP]], align 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double sum = fixed;
sum = sum + c5.x + c5.y;
return sum;
}
double fixed_5c(__char5 *in) {
// CHECK: fixed_5c
// CHECK: call double @args_vec_5c(i32 5, <2 x i32> {{%.*}})
return args_vec_5c(5, *in);
}
__attribute__((noinline)) double args_vec_9c(int fixed, __char9 c9) {
// CHECK: args_vec_9c
// CHECK: [[C9:%.*]] = alloca <9 x i8>, align 16
// CHECK: [[TMP:%.*]] = bitcast <9 x i8>* [[C9]] to <4 x i32>*
Respect alignment when loading up a coerced function argument Code in CGCall.cpp that loads up function arguments that need to be coerced to a different type may in some cases ignore the fact that the source of the argument is not naturally aligned. This may cause incorrect code to be generated. In some places in CreateCoercedLoad, we already have setAlignment calls to address this, but I ran into one where it was missing, causing wrong code generation on SystemZ. However, in that location, we do not actually know what alignment of the source location we can rely on; the callers do not pass anything to this routine. This is already an issue in other places in CreateCoercedLoad; and the same problem exists for CreateCoercedStore. To avoid pessimising code, and to fix the FIXMEs already in place, this patch also adds an alignment argument to the CreateCoerced* routines and uses it instead of forcing an alignment of 1. The callers are changed to pass in the best information they have. This actually requires changes in a number of existing test cases since we now get better alignment in many places. Differential Revision: http://reviews.llvm.org/D11033 llvm-svn: 241898
2015-07-10 19:31:43 +08:00
// CHECK: store <4 x i32> {{%.*}}, <4 x i32>* [[TMP]], align 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double sum = fixed;
sum = sum + c9.x + c9.y;
return sum;
}
double fixed_9c(__char9 *in) {
// CHECK: fixed_9c
// CHECK: call double @args_vec_9c(i32 9, <4 x i32> {{%.*}})
return args_vec_9c(9, *in);
}
__attribute__((noinline)) double args_vec_19c(int fixed, __char19 c19) {
// CHECK: args_vec_19c
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: [[C19:%.*]] = load <19 x i8>, <19 x i8>* {{.*}}, align 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double sum = fixed;
sum = sum + c19.x + c19.y;
return sum;
}
double fixed_19c(__char19 *in) {
// CHECK: fixed_19c
// CHECK: call double @args_vec_19c(i32 19, <19 x i8>* {{%.*}})
return args_vec_19c(19, *in);
}
__attribute__((noinline)) double args_vec_3s(int fixed, __short3 c3) {
// CHECK: args_vec_3s
// CHECK: [[C3:%.*]] = alloca <3 x i16>, align 8
// CHECK: [[TMP:%.*]] = bitcast <3 x i16>* [[C3]] to <2 x i32>*
Respect alignment when loading up a coerced function argument Code in CGCall.cpp that loads up function arguments that need to be coerced to a different type may in some cases ignore the fact that the source of the argument is not naturally aligned. This may cause incorrect code to be generated. In some places in CreateCoercedLoad, we already have setAlignment calls to address this, but I ran into one where it was missing, causing wrong code generation on SystemZ. However, in that location, we do not actually know what alignment of the source location we can rely on; the callers do not pass anything to this routine. This is already an issue in other places in CreateCoercedLoad; and the same problem exists for CreateCoercedStore. To avoid pessimising code, and to fix the FIXMEs already in place, this patch also adds an alignment argument to the CreateCoerced* routines and uses it instead of forcing an alignment of 1. The callers are changed to pass in the best information they have. This actually requires changes in a number of existing test cases since we now get better alignment in many places. Differential Revision: http://reviews.llvm.org/D11033 llvm-svn: 241898
2015-07-10 19:31:43 +08:00
// CHECK: store <2 x i32> {{%.*}}, <2 x i32>* [[TMP]], align 8
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double sum = fixed;
sum = sum + c3.x + c3.y;
return sum;
}
double fixed_3s(__short3 *in) {
// CHECK: fixed_3s
// CHECK: call double @args_vec_3s(i32 3, <2 x i32> {{%.*}})
return args_vec_3s(3, *in);
}
__attribute__((noinline)) double args_vec_5s(int fixed, __short5 c5) {
// CHECK: args_vec_5s
// CHECK: [[C5:%.*]] = alloca <5 x i16>, align 16
// CHECK: [[TMP:%.*]] = bitcast <5 x i16>* [[C5]] to <4 x i32>*
Respect alignment when loading up a coerced function argument Code in CGCall.cpp that loads up function arguments that need to be coerced to a different type may in some cases ignore the fact that the source of the argument is not naturally aligned. This may cause incorrect code to be generated. In some places in CreateCoercedLoad, we already have setAlignment calls to address this, but I ran into one where it was missing, causing wrong code generation on SystemZ. However, in that location, we do not actually know what alignment of the source location we can rely on; the callers do not pass anything to this routine. This is already an issue in other places in CreateCoercedLoad; and the same problem exists for CreateCoercedStore. To avoid pessimising code, and to fix the FIXMEs already in place, this patch also adds an alignment argument to the CreateCoerced* routines and uses it instead of forcing an alignment of 1. The callers are changed to pass in the best information they have. This actually requires changes in a number of existing test cases since we now get better alignment in many places. Differential Revision: http://reviews.llvm.org/D11033 llvm-svn: 241898
2015-07-10 19:31:43 +08:00
// CHECK: store <4 x i32> {{%.*}}, <4 x i32>* [[TMP]], align 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double sum = fixed;
sum = sum + c5.x + c5.y;
return sum;
}
double fixed_5s(__short5 *in) {
// CHECK: fixed_5s
// CHECK: call double @args_vec_5s(i32 5, <4 x i32> {{%.*}})
return args_vec_5s(5, *in);
}
__attribute__((noinline)) double args_vec_3i(int fixed, __int3 c3) {
// CHECK: args_vec_3i
// CHECK: [[C3:%.*]] = alloca <3 x i32>, align 16
// CHECK: [[TMP:%.*]] = bitcast <3 x i32>* [[C3]] to <4 x i32>*
Respect alignment when loading up a coerced function argument Code in CGCall.cpp that loads up function arguments that need to be coerced to a different type may in some cases ignore the fact that the source of the argument is not naturally aligned. This may cause incorrect code to be generated. In some places in CreateCoercedLoad, we already have setAlignment calls to address this, but I ran into one where it was missing, causing wrong code generation on SystemZ. However, in that location, we do not actually know what alignment of the source location we can rely on; the callers do not pass anything to this routine. This is already an issue in other places in CreateCoercedLoad; and the same problem exists for CreateCoercedStore. To avoid pessimising code, and to fix the FIXMEs already in place, this patch also adds an alignment argument to the CreateCoerced* routines and uses it instead of forcing an alignment of 1. The callers are changed to pass in the best information they have. This actually requires changes in a number of existing test cases since we now get better alignment in many places. Differential Revision: http://reviews.llvm.org/D11033 llvm-svn: 241898
2015-07-10 19:31:43 +08:00
// CHECK: store <4 x i32> {{%.*}}, <4 x i32>* [[TMP]], align 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double sum = fixed;
sum = sum + c3.x + c3.y;
return sum;
}
double fixed_3i(__int3 *in) {
// CHECK: fixed_3i
// CHECK: call double @args_vec_3i(i32 3, <4 x i32> {{%.*}})
return args_vec_3i(3, *in);
}
__attribute__((noinline)) double args_vec_5i(int fixed, __int5 c5) {
// CHECK: args_vec_5i
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: [[C5:%.*]] = load <5 x i32>, <5 x i32>* {{%.*}}, align 16
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
double sum = fixed;
sum = sum + c5.x + c5.y;
return sum;
}
double fixed_5i(__int5 *in) {
// CHECK: fixed_5i
// CHECK: call double @args_vec_5i(i32 5, <5 x i32>* {{%.*}})
return args_vec_5i(5, *in);
}
__attribute__((noinline)) double args_vec_3d(int fixed, __double3 c3) {
// CHECK: args_vec_3d
// CHECK: [[CAST:%.*]] = bitcast <3 x double>* {{%.*}} to <4 x double>*
Update Clang tests to handle explicitly typed load changes in LLVM. llvm-svn: 230795
2015-02-28 05:19:58 +08:00
// CHECK: [[LOAD:%.*]] = load <4 x double>, <4 x double>* [[CAST]]
ARM64: initial clang support commit. This adds Clang support for the ARM64 backend. There are definitely still some rough edges, so please bring up any issues you see with this patch. As with the LLVM commit though, we think it'll be more useful for merging with AArch64 from within the tree. llvm-svn: 205100
2014-03-29 23:09:45 +08:00
// CHECK: shufflevector <4 x double> [[LOAD]], <4 x double> undef, <3 x i32> <i32 0, i32 1, i32 2>
double sum = fixed;
sum = sum + c3.x + c3.y;
return sum;
}
double fixed_3d(__double3 *in) {
// CHECK: fixed_3d
// CHECK: call double @args_vec_3d(i32 3, <3 x double>* {{%.*}})
return args_vec_3d(3, *in);
}