summaryrefslogtreecommitdiff
path: root/src/cuda-sim/instructions.cc
diff options
context:
space:
mode:
Diffstat (limited to 'src/cuda-sim/instructions.cc')
-rw-r--r--src/cuda-sim/instructions.cc8
1 files changed, 7 insertions, 1 deletions
diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index bddaf8c..7407269 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -1493,6 +1493,12 @@ unsigned trunc(unsigned num, unsigned precision) {
}
return num;
}
+void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
+{
+}
+void mma_st_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
+{
+}
void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
{
@@ -1565,7 +1571,7 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
for (i=0;i<16;i++){
for(j=0;j<16;j++){
for(k=0;k<16;k++){
- matrix_d[i][j].f32=matrix_d[i][j].f32+matrix_a[i][k].f32*matrix_b[k][j].f32;
+ matrix_d[i][j].f32=matrix_d[i][j].f32+matrix_a[i][k].f32*matrix_b[j][k].f32;
}
matrix_d[i][j].f32+=matrix_c[i][j].f32;
}