From 809a387769788d028252139d5bbd58c502c4eb43 Mon Sep 17 00:00:00 2001
From: aamir <araihan@ece.ubc.ca>
Date: Mon, 11 Jun 2018 16:02:19 -0700
Subject: tested all the configuration of mma

---
 src/cuda-sim/instructions.cc | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

(limited to 'src')

diff --git a/src/cuda-sim/instructions.cc b/src/cuda-sim/instructions.cc
index e03cbce..8b66cd0 100644
--- a/src/cuda-sim/instructions.cc
+++ b/src/cuda-sim/instructions.cc
@@ -1701,7 +1701,7 @@ void mma_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
 
 			if(!((i==3)&&(type2==F32_TYPE))){
 				for(k=0;k<2*nelem;k++){
-					if(k%2==0)
+					if(k%2==1)
 						hex_val=(v[k/2].s64&0xffff);
 					else
 						hex_val=((v[k/2].s64&0xffff0000)>>16);
@@ -3094,7 +3094,7 @@ void mma_ld_impl( const ptx_instruction *pI, core_t *core, warp_inst_t inst )
 			num_reg=8;
 
 		for(i=0;i<num_reg;i++){
-			nw_data[i].s64= ((data[2*i].s64 & 0xffff))| ((data[2*i+1].s64 & 0xffff)<<16);
+			nw_data[i].s64= ((data[2*i].s64 & 0xffff)<<16)| ((data[2*i+1].s64 & 0xffff));
 		}
 
 		if(wmma_type==LOAD_C)
-- 
cgit v1.3