def main(vlength = 128,loops = 1):
n2 = vlength ## Vector length
h_S = (c_float*n2)()
h_X = (c_float*n2)()
h_T = (c_float*n2)()
h_C = (c_float*n2)()
h_P = (c_float*n2)()
randInit(h_S,5.,30.)
randInit(h_X,1.,100.)
randInit(h_T,.25,10.)
R,V = .03,.3
d_S = getMemory(h_S)
d_X = getMemory(h_X)
d_T = getMemory(h_T)
d_C = getMemory(h_C)
d_P = getMemory(h_P)
blockDim = dim3(BLOCK_SIZE,1,1)
gridDim = dim3(GRID_SIZE,1,1)
cudaThreadSynchronize()
t0 = time()
for i in range(loops):
cudaConfigureCall(gridDim,blockDim,0,0)
gpuBLSC(d_C,d_P,d_S,d_X,d_T,R,V,n2)
cudaThreadSynchronize()
t0 = time()-t0
flops = (2.e-6*n2)*float(loops)
g_C = (c_float*n2)()
g_P = (c_float*n2)()
cudaMemcpy(g_C,d_C,S4*n2,cudaMemcpyDeviceToHost)
cudaMemcpy(g_P,d_P,S4*n2,cudaMemcpyDeviceToHost)
cudaThreadSynchronize()
cudaFree(d_S)
cudaFree(d_X)
cudaFree(d_T)
cudaFree(d_C)
cudaFree(d_P)
cudaThreadExit()
t1 = time()
for i in range(loops):
cpuBLSC(h_C,h_P,h_S,h_X,h_T,R,V,n2)
t1 = time()-t1
print "%10d%10.2f%10.2f" % (vlength,flops/t1,flops/t0)
if checkErrorFlag:
err,mxe = checkError(h_C,g_C)
print "Avg rel error (call) = %.2e" % (err,)
err,mxe = checkError(h_P,g_P)
print "Avg rel error (put) = %.2e" % (err,)
def main(vlength = 128,loops = 1):
n2 = vlength ## Vector length
h_S = (c_float*n2)()
h_X = (c_float*n2)()
h_T = (c_float*n2)()
h_C = (c_float*n2)()
h_P = (c_float*n2)()
randInit(h_S,5.,30.)
randInit(h_X,1.,100.)
randInit(h_T,.25,10.)
R,V = .03,.3
d_S = getMemory(h_S)
d_X = getMemory(h_X)
d_T = getMemory(h_T)
d_C = getMemory(h_C)
d_P = getMemory(h_P)
blockDim = dim3(BLOCK_SIZE,1,1)
gridDim = dim3(GRID_SIZE,1,1)
cudaThreadSynchronize()
t0 = time()
for i in range(loops):
cudaConfigureCall(gridDim,blockDim,0,0)
gpuBLSC(d_C,d_P,d_S,d_X,d_T,R,V,n2)
cudaThreadSynchronize()
t0 = time()-t0
flops = (2.e-6*n2)*float(loops)
g_C = (c_float*n2)()
g_P = (c_float*n2)()
cudaMemcpy(g_C,d_C,S4*n2,cudaMemcpyDeviceToHost)
cudaMemcpy(g_P,d_P,S4*n2,cudaMemcpyDeviceToHost)
cudaThreadSynchronize()
cudaFree(d_S)
cudaFree(d_X)
cudaFree(d_T)
cudaFree(d_C)
cudaFree(d_P)
cudaThreadExit()
t1 = time()
for i in range(loops):
cpuBLSC(h_C,h_P,h_S,h_X,h_T,R,V,n2)
t1 = time()-t1
print "%10d%10.2f%10.2f" % (vlength,flops/t1,flops/t0)
if checkErrorFlag:
err,mxe = checkError(h_C,g_C)
print "Avg rel error (call) = %.2e" % (err,)
err,mxe = checkError(h_P,g_P)
print "Avg rel error (put) = %.2e" % (err,)
def main(device,vlength = 128,loops = 1):
n2 = vlength ## Vector length
gpuBLSC = device.functions["gpuBLSC"]
h_S = (c_float*n2)()
h_X = (c_float*n2)()
h_T = (c_float*n2)()
h_C = (c_float*n2)()
h_P = (c_float*n2)()
randInit(h_S,5.,30.)
randInit(h_X,1.,100.)
randInit(h_T,.25,10.)
R,V = .03,.3
d_S = getMemory(h_S)
d_X = getMemory(h_X)
d_T = getMemory(h_T)
d_C = getMemory(h_C)
d_P = getMemory(h_P)
cuFuncSetBlockShape(gpuBLSC,BLOCK_SIZE,1,1)
cuParamSeti(gpuBLSC, 0,d_C)
cuParamSeti(gpuBLSC, 4,d_P)
cuParamSeti(gpuBLSC, 8,d_S)
cuParamSeti(gpuBLSC,12,d_X)
cuParamSeti(gpuBLSC,16,d_T)
cuParamSetf(gpuBLSC,20,R)
cuParamSetf(gpuBLSC,24,V)
cuParamSeti(gpuBLSC,28,n2)
cuParamSetSize(gpuBLSC,32)
cuCtxSynchronize()
t0 = time()
for i in range(loops):
cuLaunchGrid(gpuBLSC,GRID_SIZE,1)
cuCtxSynchronize()
t0 = time()-t0
flops = (2.e-6*n2)*float(loops)
g_C = (c_float*n2)()
g_P = (c_float*n2)()
cuMemcpyDtoH(g_C,d_C,n2*S4)
cuMemcpyDtoH(g_P,d_P,n2*S4)
cuCtxSynchronize()
cuMemFree(d_S)
cuMemFree(d_X)
cuMemFree(d_T)
cuMemFree(d_C)
cuMemFree(d_P)
t1 = time()
for i in range(loops):
cpuBLSC(h_C,h_P,h_S,h_X,h_T,R,V,n2)
t1 = time()-t1
print "%10d%10.2f%10.2f" % (vlength,flops/t1,flops/t0)
if checkErrorFlag:
err,mxe = checkError(h_C,g_C)
print "Avg rel error (call) = %.2e" % (err,)
err,mxe = checkError(h_P,g_P)
print "Avg rel error (put) = %.2e" % (err,)