Skip to content

Commit 03aaef8

Browse files
ecnelisesecnelises
committed
[PowerPC][Clang] Remove use of malloc in mm_malloc
Remove dependency of malloc in implementation of mm_malloc function in PowerPC intrinsics and alignment assumption on glibc. Reviewed By: Hal Finkel Differential Revision: https://reviews.llvm.org/D64850 llvm-svn: 366406
1 parent 8da0402 commit 03aaef8

File tree

3 files changed

+8
-44
lines changed

3 files changed

+8
-44
lines changed

‎clang/lib/Headers/ppc_wrappers/mm_malloc.h

Lines changed: 0 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -25,12 +25,8 @@ _mm_malloc (size_t size, size_t alignment)
2525
{
2626
/* PowerPC64 ELF V2 ABI requires quadword alignment. */
2727
size_t vec_align = sizeof (__vector float);
28-
/* Linux GLIBC malloc alignment is at least 2 X ptr size. */
29-
size_t malloc_align = (sizeof (void *) + sizeof (void *));
3028
void *ptr;
3129

32-
if (alignment == malloc_align && alignment == vec_align)
33-
return malloc (size);
3430
if (alignment < vec_align)
3531
alignment = vec_align;
3632
if (posix_memalign (&ptr, alignment, size) == 0)

‎clang/test/CodeGen/ppc-mm-malloc-le.c

Lines changed: 4 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -24,30 +24,14 @@ test_mm_malloc() {
2424
// CHECK: store i64 [[REG1]], i64* [[REG4:[0-9a-zA-Z_%.]+]], align 8
2525
// CHECK-NEXT: store i64 [[REG2]], i64* [[REG5:[0-9a-zA-Z_%.]+]], align 8
2626
// CHECK-NEXT: store i64 16, i64* [[REG6:[0-9a-zA-Z_%.]+]], align 8
27-
// CHECK-NEXT: store i64 16, i64* [[REG7:[0-9a-zA-Z_%.]+]], align 8
2827
// CHECK-NEXT: [[REG8:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
29-
// CHECK-NEXT: [[REG9:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG7]], align 8
30-
// CHECK-NEXT: [[REG10:[0-9a-zA-Z_%.]+]] = icmp eq i64 [[REG8]], [[REG9]]
31-
// CHECK-NEXT: br i1 [[REG10]], label %[[REG11:[0-9a-zA-Z_%.]+]], label %[[REG12:[0-9a-zA-Z_%.]+]]
32-
// CHECK: [[REG11]]:
33-
// CHECK-NEXT: [[REG13:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
34-
// CHECK-NEXT: [[REG14:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
35-
// CHECK-NEXT: [[REG15:[0-9a-zA-Z_%.]+]] = icmp eq i64 [[REG13]], [[REG14]]
36-
// CHECK-NEXT: br i1 [[REG15]], label %[[REG16:[0-9a-zA-Z_%.]+]], label %[[REG12:[0-9a-zA-Z_%.]+]]
37-
// CHECK: [[REG16]]:
38-
// CHECK-NEXT: [[REG17:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG4]], align 8
39-
// CHECK-NEXT: [[REG18:[0-9a-zA-Z_%.]+]] = call noalias i8* @malloc(i64 [[REG17]])
40-
// CHECK-NEXT: store i8* [[REG18]], i8** [[REG3]], align 8
41-
// CHECK-NEXT: br label %[[REG19:[0-9a-zA-Z_%.]+]]
42-
// CHECK: [[REG12]]:
43-
// CHECK-NEXT: [[REG20:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
44-
// CHECK-NEXT: [[REG21:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
45-
// CHECK-NEXT: [[REG22:[0-9a-zA-Z_%.]+]] = icmp ult i64 [[REG20]], [[REG21]]
46-
// CHECK-NEXT: br i1 [[REG22]], label %[[REG23:[0-9a-zA-Z_%.]+]], label %[[REG24:[0-9a-zA-Z_%.]+]]
28+
// CHECK-NEXT: [[REG9:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
29+
// CHECK-NEXT: [[REG10:[0-9a-zA-Z_%.]+]] = icmp ult i64 [[REG8]], [[REG9]]
30+
// CHECK-NEXT: br i1 [[REG10]], label %[[REG23:[0-9a-zA-Z_%.]+]], label %[[REG24:[0-9a-zA-Z_%.]+]]
4731
// CHECK: [[REG23]]:
4832
// CHECK-NEXT: [[REG25:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
4933
// CHECK-NEXT: store i64 [[REG25]], i64* [[REG5]], align 8
50-
// CHECK-NEXT: br label %[[REG12:[0-9a-zA-Z_%.]+]]4
34+
// CHECK-NEXT: br label %[[REG24:[0-9a-zA-Z_%.]+]]
5135
// CHECK: [[REG24]]:
5236
// CHECK-NEXT: [[REG26:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
5337
// CHECK-NEXT: [[REG27:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG4]], align 8

‎clang/test/CodeGen/ppc-mm-malloc.c

Lines changed: 4 additions & 20 deletions
Original file line numberDiff line numberDiff line change
@@ -24,30 +24,14 @@ test_mm_malloc() {
2424
// CHECK: store i64 [[REG1]], i64* [[REG4:[0-9a-zA-Z_%.]+]], align 8
2525
// CHECK-NEXT: store i64 [[REG2]], i64* [[REG5:[0-9a-zA-Z_%.]+]], align 8
2626
// CHECK-NEXT: store i64 16, i64* [[REG6:[0-9a-zA-Z_%.]+]], align 8
27-
// CHECK-NEXT: store i64 16, i64* [[REG7:[0-9a-zA-Z_%.]+]], align 8
2827
// CHECK-NEXT: [[REG8:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
29-
// CHECK-NEXT: [[REG9:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG7]], align 8
30-
// CHECK-NEXT: [[REG10:[0-9a-zA-Z_%.]+]] = icmp eq i64 [[REG8]], [[REG9]]
31-
// CHECK-NEXT: br i1 [[REG10]], label %[[REG11:[0-9a-zA-Z_%.]+]], label %[[REG12:[0-9a-zA-Z_%.]+]]
32-
// CHECK: [[REG11]]:
33-
// CHECK-NEXT: [[REG13:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
34-
// CHECK-NEXT: [[REG14:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
35-
// CHECK-NEXT: [[REG15:[0-9a-zA-Z_%.]+]] = icmp eq i64 [[REG13]], [[REG14]]
36-
// CHECK-NEXT: br i1 [[REG15]], label %[[REG16:[0-9a-zA-Z_%.]+]], label %[[REG12:[0-9a-zA-Z_%.]+]]
37-
// CHECK: [[REG16]]:
38-
// CHECK-NEXT: [[REG17:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG4]], align 8
39-
// CHECK-NEXT: [[REG18:[0-9a-zA-Z_%.]+]] = call noalias i8* @malloc(i64 [[REG17]])
40-
// CHECK-NEXT: store i8* [[REG18]], i8** [[REG3]], align 8
41-
// CHECK-NEXT: br label %[[REG19:[0-9a-zA-Z_%.]+]]
42-
// CHECK: [[REG12]]:
43-
// CHECK-NEXT: [[REG20:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
44-
// CHECK-NEXT: [[REG21:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
45-
// CHECK-NEXT: [[REG22:[0-9a-zA-Z_%.]+]] = icmp ult i64 [[REG20]], [[REG21]]
46-
// CHECK-NEXT: br i1 [[REG22]], label %[[REG23:[0-9a-zA-Z_%.]+]], label %[[REG24:[0-9a-zA-Z_%.]+]]
28+
// CHECK-NEXT: [[REG9:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
29+
// CHECK-NEXT: [[REG10:[0-9a-zA-Z_%.]+]] = icmp ult i64 [[REG8]], [[REG9]]
30+
// CHECK-NEXT: br i1 [[REG10]], label %[[REG23:[0-9a-zA-Z_%.]+]], label %[[REG24:[0-9a-zA-Z_%.]+]]
4731
// CHECK: [[REG23]]:
4832
// CHECK-NEXT: [[REG25:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG6]], align 8
4933
// CHECK-NEXT: store i64 [[REG25]], i64* [[REG5]], align 8
50-
// CHECK-NEXT: br label %[[REG12:[0-9a-zA-Z_%.]+]]4
34+
// CHECK-NEXT: br label %[[REG24:[0-9a-zA-Z_%.]+]]
5135
// CHECK: [[REG24]]:
5236
// CHECK-NEXT: [[REG26:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG5]], align 8
5337
// CHECK-NEXT: [[REG27:[0-9a-zA-Z_%.]+]] = load i64, i64* [[REG4]], align 8

0 commit comments

Comments
 (0)