fix Windows low quality host RNG due to 15bit rand(), fix #31

fangq · Mar 11, 2018 · a0d445b · a0d445b
1 parent 521fb85
commit a0d445b
Show file tree

Hide file tree

Showing 2 changed files with 6 additions and 4 deletions.
diff --git a/src/Makefile b/src/Makefile
@@ -112,12 +112,13 @@ posix:      LINKOPT+=-DUSE_POSIX_RAND
 fast:       CUCCOPT+=-DUSE_XORSHIFT128P_RAND -use_fast_math
 log:        CUCCOPT+=
 debugxor:   xor
+debugxor:   LINKOPT+=$(OMP)
 debuglog:   CUCCOPT+=
 racing:     CUCCOPT+=-DTEST_RACING
 xoratomic:  CUCCOPT+=-DUSE_XORSHIFT128P_RAND -DUSE_ATOMIC -use_fast_math $(CUGENCODE) MCX_TARGET_NAME='"Xorshift128+ Atomic MCX"'
 logatomic:  CUCCOPT+=-DUSE_ATOMIC -use_fast_math $(CUGENCODE) -DMCX_TARGET_NAME='"LL5 Atomic MCX"'
 fermi fermimex fermioct:      CUCCOPT+=-DUSE_ATOMIC -use_fast_math
-fermi fermimex fermioct:      xoro
+fermi fermimex fermioct:      xor
 fermimex fermioct: OUTPUTFLAG:=-output
 
 fermi:      LINKOPT+=$(CUOMPLINK) $(OMP)

diff --git a/src/mcx_core.cu b/src/mcx_core.cu
@@ -1658,8 +1658,9 @@ void mcx_run_simulation(Config *cfg,GPUInfo *gpu){
            param.twin0=cfg->tstart;
            param.twin1=cfg->tend;
            Pseed=(uint*)malloc(sizeof(RandType)*RAND_BUF_LEN);
-           for (i=0; i<(int)(((sizeof(RandType)*RAND_BUF_LEN)>>2)); i++)
-		Pseed[i]=rand();
+           for (i=0; i<(int)(((sizeof(RandType)*RAND_BUF_LEN)>>2)); i++){
+		Pseed[i]=((rand() << 16) | (rand() << 1) | (rand() >> 14));
+	   }
            CUDA_ASSERT(cudaMalloc((void **) &gPseed, sizeof(RandType)*RAND_BUF_LEN));
 	   CUDA_ASSERT(cudaMemcpy(gPseed, Pseed, sizeof(RandType)*RAND_BUF_LEN,  cudaMemcpyHostToDevice));
            CUDA_ASSERT(cudaMalloc((void **) &gfield, sizeof(float)*fieldlen));
@@ -1880,7 +1881,7 @@ void mcx_run_simulation(Config *cfg,GPUInfo *gpu){
 
            if(cfg->seed!=SEED_FROM_FILE){
              for (i=0; i<gpu[gpuid].autothread*((int)(sizeof(RandType)*RAND_BUF_LEN)>>2); i++)
-               Pseed[i]=rand();
+               Pseed[i]=((rand() << 16) | (rand() << 1) | (rand() >> 14));
 	     CUDA_ASSERT(cudaMemcpy(gPseed, Pseed, sizeof(RandType)*gpu[gpuid].autothread*RAND_BUF_LEN,  cudaMemcpyHostToDevice));
            }
            tic0=GetTimeMillis();