我正在尝试使用OpenCL加速我的神经网络，我遇到的唯一问题是当我将数组的大小从(10*6), (6*3), (6*10)增加到(1000000*6), (6*3), (6*10)时返回的数据的一半我已经检查了内存使用情况，它在GPU的内存中使用的最大值为91MB。有没有办法解决它，还是物理限制？
from __future__ import division
KERNEL_CODE = """
// Thread block size
#define BLOCK_SIZE %(block_size)d
// Matrix dimensions
// (chosen as multiples of the thread block size for simplicity)
#define WA %(w_a)d // Matrix A width
#define HA %(h_a)d // Matrix A height
#define WB %(w_b)d // Matrix B width
#define HB WA // Matrix B height
#define WC WB // Matrix C width
#define HC HA // Matrix C height
/*
* Copyright 1993-2009 NVIDIA Corporation. All rights reserved.
*
* NVIDIA Corporation and its licensors retain all intellectual property and
* proprietary rights in and to this software and related documentation.
* Any use, reproduction, disclosure, or distribution of this software
* and related documentation without an express license agreement from
* NVIDIA Corporation is strictly prohibited.
*
* Please refer to the applicable NVIDIA end user license agreement (EULA)
* associated with this source code for terms and conditions that govern
* your use of this NVIDIA software.
*
*/
/* Matrix multiplication: C = A * B.
* Device code.
*/
#define AS(j, i) As[i + j * BLOCK_SIZE]
#define BS(j, i) Bs[i + j * BLOCK_SIZE]
////////////////////////////////////////////////////////////////////////////////
//! Matrix multiplication on the device: C = A * B
//! WA is A's width and WB is B's width
////////////////////////////////////////////////////////////////////////////////
__kernel __attribute__((reqd_work_group_size(16,16,1)))
void
matrixMul( __global float* C, __global float* A, __global float* B)
{
__local float As[BLOCK_SIZE*BLOCK_SIZE];
__local float Bs[BLOCK_SIZE*BLOCK_SIZE];
// Block index
int bx = get_group_id(0);
int by = get_group_id(1);
// Thread index
int tx = get_local_id(0);
int ty = get_local_id(1);
// Index of the first sub-matrix of A processed by the block
int aBegin = WA * BLOCK_SIZE * by;
// Index of the last sub-matrix of A processed by the block
int aEnd = aBegin + WA - 1;
// Step size used to iterate through the sub-matrices of A
int aStep = BLOCK_SIZE;
// Index of the first sub-matrix of B processed by the block
int bBegin = BLOCK_SIZE * bx;
// Step size used to iterate through the sub-matrices of B
int bStep = BLOCK_SIZE * WB;
// Csub is used to store the element of the block sub-matrix
// that is computed by the thread
float Csub = 0.0f;
// Loop over all the sub-matrices of A and B
// required to compute the block sub-matrix
for (int a = aBegin, b = bBegin;
a <= aEnd;
a += aStep, b += bStep) {
// Load the matrices from device memory
// to shared memory; each thread loads
// one element of each matrix
AS(ty, tx) = A[a + WA * ty + tx];
BS(ty, tx) = B[b + WB * ty + tx];
// Synchronize to make sure the matrices are loaded
barrier(CLK_LOCAL_MEM_FENCE);
// Multiply the two matrices together;
// each thread computes one element
// of the block sub-matrix
for (int k = 0; k < BLOCK_SIZE; ++k)
Csub += AS(ty, k) * BS(k, tx);
// Synchronize to make sure that the preceding
// computation is done before loading two new
// sub-matrices of A and B in the next iteration
barrier(CLK_LOCAL_MEM_FENCE);
}
// Write the block sub-matrix to device memory;
// each thread writes one element
C[get_global_id(1) * get_global_size(0) + get_global_id(0)] = Csub;
}
"""
import pyopencl as cl
from time import time
import numpy
block_size = 16
ctx = cl.create_some_context()
for dev in ctx.devices:
assert dev.local_mem_size > 0
queue = cl.CommandQueue(ctx,
properties=cl.command_queue_properties.PROFILING_ENABLE)
#queue = cl.CommandQueue(ctx)
a_width = 16
a_height = 1000000*16
b_width = 16
b_height = a_width
c_width = b_width
c_height = a_height
h_a = numpy.random.rand(a_height, a_width).astype(numpy.float32)
h_b = numpy.random.rand(b_height, b_width).astype(numpy.float32)
h_c = numpy.empty((c_height, c_width)).astype(numpy.float32)
kernel_params = {"block_size": block_size,
"w_a":a_width, "h_a":a_height, "w_b":b_width}
prg = cl.Program(ctx, KERNEL_CODE % kernel_params,
).build(options="-cl-mad-enable -cl-fast-relaxed-math")
kernel = prg.matrixMul
#print prg.binaries[0]
assert a_width % block_size == 0
assert a_height % block_size == 0
assert b_width % block_size == 0
# transfer host -> device -----------------------------------------------------
mf = cl.mem_flags
t1 = time()
d_a_buf = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=h_a)
d_b_buf = cl.Buffer(ctx, mf.READ_ONLY | mf.COPY_HOST_PTR, hostbuf=h_b)
d_c_buf = cl.Buffer(ctx, mf.WRITE_ONLY, size=h_c.nbytes)
push_time = time()-t1
# warmup ----------------------------------------------------------------------
for i in range(1):
event = kernel(queue, h_c.shape, (block_size, block_size),
d_c_buf, d_a_buf, d_b_buf)
event.wait()
queue.finish()
# actual benchmark ------------------------------------------------------------
t1 = time()
count = 20
for i in range(count):
event = kernel(queue, h_c.shape, (block_size, block_size),
d_c_buf, d_a_buf, d_b_buf)
event.wait()
gpu_time = (time()-t1)/count
# transfer device -> host -----------------------------------------------------
t1 = time()
cl.enqueue_read_buffer(queue, d_c_buf, h_c).wait()
pull_time = time()-t1
# timing output ---------------------------------------------------------------
gpu_total_time = gpu_time+push_time+pull_time
print "GPU push+compute+pull total [s]:", gpu_total_time
print "GPU push [s]:", push_time
print "GPU pull [s]:", pull_time
print "GPU compute (host-timed) [s]:", gpu_time
print "GPU compute (event-timed) [s]: ", (event.profile.end-event.profile.start)*1e-9
gflop = h_c.size * (a_width * 2.) / (1000**3.)
gflops = gflop / gpu_time
print
print "GFlops/s:", gflops
# cpu comparison --------------------------------------------------------------
t1 = time()
h_c_cpu = numpy.dot(h_a,h_b)
cpu_time = time()-t1
print
print "GPU==CPU:",numpy.allclose(h_c, h_c_cpu)
print
print "CPU time (s)", cpu_time
print
print "GPU speedup (with transfer): ", cpu_time/gpu_total_time
print "GPU speedup (without transfer): ", cpu_time/gpu_time
print h_c