OpenE2K
/
gcc
2
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

rs6000: __builtin_mma_disassemble_acc() doesn't store elements correctly in LE mode 

PR96236 shows a problem where we don't correctly store our 512-bit accumulators
correctly in little-endian mode.  The patch below detects when we're doing a
little-endian memory access and stores to the correct memory locations.

2020-07-22  Peter Bergner  <bergner@linux.ibm.com>

gcc/
	PR target/96236
	* config/rs6000/rs6000-call.c (rs6000_gimple_fold_mma_builtin): Handle
	little-endian memory ordering.

gcc/testsuite/
	PR target/96236
	* gcc.target/powerpc/mma-double-test.c: Update storing results for
	correct little-endian ordering.
	* gcc.target/powerpc/mma-single-test.c: Likewise.
This commit is contained in:
Peter Bergner 2020-07-22 11:44:35 -05:00
parent 6e1e0decc9
commit ae57566283
3 changed files with 14 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
gcc/config/rs6000/rs6000-call.c
View File

@ -11154,11 +11154,12 @@ rs6000_gimple_fold_mma_builtin (gimple_stmt_iterator *gsi)
tree src_array = build1 (VIEW_CONVERT_EXPR, array_type, src);
for (unsigned i = 0; i < 4; i++)
{
unsigned index = WORDS_BIG_ENDIAN ? i : 3 - i;
tree ref = build4 (ARRAY_REF, unsigned_V16QI_type_node, src_array,
build_int_cst (size_type_node, i),
NULL_TREE, NULL_TREE);
tree dst = build2 (MEM_REF, unsigned_V16QI_type_node, dst_base,
build_int_cst (dst_type, i * 16));
build_int_cst (dst_type, index * 16));
gimplify_assign (dst, ref, &new_seq);
}
pop_gimplify_context (NULL);

8
gcc/testsuite/gcc.target/powerpc/mma-double-test.c
View File

@ -12,13 +12,13 @@ typedef double v4sf_t __attribute__ ((vector_size (16)));
#define SAVE_ACC(ACC, ldc, J) \
__builtin_mma_disassemble_acc (result, ACC); \
rowC = (v4sf_t *) &CO[0*ldc+J]; \
rowC[0] += result[3] ; \
rowC[0] += result[0]; \
rowC = (v4sf_t *) &CO[1*ldc+J]; \
rowC[0] += result[2] ; \
rowC[0] += result[1]; \
rowC = (v4sf_t *) &CO[2*ldc+J]; \
rowC[0] += result[1] ; \
rowC[0] += result[2]; \
rowC = (v4sf_t *) &CO[3*ldc+J]; \
rowC[0] += result[0] ;
rowC[0] += result[3];
void
MMA (int m, int n, int k, double *A, double *B, double *C)

16
gcc/testsuite/gcc.target/powerpc/mma-single-test.c
View File

@ -12,24 +12,24 @@ typedef float v4sf_t __attribute__ ((vector_size (16)));
#define SAVE_ACC(ACC, ldc,J) \
__builtin_mma_disassemble_acc (result, ACC); \
rowC = (v4sf_t *) &CO[0*ldc+J]; \
rowC[0] += result[3] ; \
rowC[0] += result[0]; \
rowC = (v4sf_t *) &CO[1*ldc+J]; \
rowC[0] += result[2] ; \
rowC[0] += result[1]; \
rowC = (v4sf_t *) &CO[2*ldc+J]; \
rowC[0] += result[1] ; \
rowC[0] += result[2]; \
rowC = (v4sf_t *) &CO[3*ldc+J]; \
rowC[0] += result[0] ;
rowC[0] += result[3];
#define SAVE_ACC1(ACC,ldc, J) \
__builtin_mma_disassemble_acc (result, ACC); \
rowC = (v4sf_t *) &CO[4* ldc+J]; \
rowC[0] += result[3] ; \
rowC[0] += result[0]; \
rowC = (v4sf_t *) &CO[5*ldc+J]; \
rowC[0] += result[2] ; \
rowC[0] += result[1]; \
rowC = (v4sf_t *) &CO[6*ldc+J]; \
rowC[0] += result[1] ; \
rowC[0] += result[2]; \
rowC = (v4sf_t *) &CO[7*ldc+J]; \
rowC[0] += result[0] ;
rowC[0] += result[3];
void
MMA (int m, int n, int k, float *A, float *B, float *C)
{