compute-runtime/opencl/test/unit_test/test_files/vme_kernels.cl

107 lines
4.8 KiB
Common Lisp

/*
* Copyright (C) 2018-2021 Intel Corporation
*
* SPDX-License-Identifier: MIT
*
*/
/*************************************************************************************************\
Device-side basic vme kernel:
device_side_block_motion_estimate_intel
Description:
Computes motion vectors by comparing a 2d image source with a 2d reference image, producing a
vector field motion vectors. The algorithm searches the best match of each macroblock pixel
block in the source image by searching an image region in the reference image, centered on the
coordinates of that pixel macroblock in the source image (optionally offset by the prediction
motion vectors).
This kernel optionally takes a vector field of motion vector predictors via the
prediction_motion_vector_image kernel argument. The kernel also optionally returns a vector
field of per-macroblock pixel-block information records. Each record contains the best-match
distortion (SAD) value and additional search result information.
This kernel needs to be compiled with following compiler option:
" -D cl_intel_device_side_avc_vme_enable "
\*************************************************************************************************/
__kernel __attribute__((reqd_work_group_size(16, 1, 1)))
void device_side_block_motion_estimate_intel(__read_only image2d_t srcImg,
__read_only image2d_t refImg,
__global short2 *prediction_motion_vector_buffer,
__global short2 *motion_vector_buffer,
__global ushort *residuals_buffer,
__global uchar2 *shapes_buffer,
int iterations,
int partitionMask) {
int gid_0 = get_group_id(0);
int gid_1 = 0;
sampler_t vme_samp = 0;
for (int i = 0; i < iterations; i++, gid_1++) {
ushort2 srcCoord = 0;
short2 refCoord = 0;
short2 predMV = 0;
srcCoord.x = gid_0 * 16 + get_global_offset(0);
srcCoord.y = gid_1 * 16 + get_global_offset(1);
if (prediction_motion_vector_buffer != NULL) {
predMV = prediction_motion_vector_buffer[gid_0 + gid_1 * get_num_groups(0)];
refCoord.x = predMV.x / 4;
refCoord.y = predMV.y / 4;
refCoord.y = refCoord.y & 0xFFFE;
}
uchar partition_mask = (uchar)partitionMask;
uchar sad_adjustment = CLK_AVC_ME_SAD_ADJUST_MODE_NONE_INTEL;
uchar pixel_mode = CLK_AVC_ME_SUBPIXEL_MODE_QPEL_INTEL;
intel_sub_group_avc_ime_payload_t payload = intel_sub_group_avc_ime_initialize(srcCoord, partition_mask, sad_adjustment);
payload = intel_sub_group_avc_ime_set_single_reference(refCoord, CLK_AVC_ME_SEARCH_WINDOW_16x12_RADIUS_INTEL, payload);
intel_sub_group_avc_ime_result_t result = intel_sub_group_avc_ime_evaluate_with_single_reference(srcImg, refImg, vme_samp, payload);
// Process Results
long mvs = intel_sub_group_avc_ime_get_motion_vectors(result);
ushort sads = intel_sub_group_avc_ime_get_inter_distortions(result);
uchar major_shape = intel_sub_group_avc_ime_get_inter_major_shape(result);
uchar minor_shapes = intel_sub_group_avc_ime_get_inter_minor_shapes(result);
uchar2 shapes = {major_shape, minor_shapes};
uchar directions = intel_sub_group_avc_ime_get_inter_directions(result);
// Perform FME for non-Integer Pixel mode
if (pixel_mode != CLK_AVC_ME_SUBPIXEL_MODE_INTEGER_INTEL) {
intel_sub_group_avc_ref_payload_t payload = intel_sub_group_avc_fme_initialize(srcCoord, mvs, major_shape, minor_shapes, directions, pixel_mode, sad_adjustment);
intel_sub_group_avc_ref_result_t result = intel_sub_group_avc_ref_evaluate_with_single_reference(srcImg, refImg, vme_samp, payload);
mvs = intel_sub_group_avc_ref_get_motion_vectors(result);
sads = intel_sub_group_avc_ref_get_inter_distortions(result);
}
// Write Out Result
if ((get_local_id(0) % 4) == 0) {
int x = get_local_id(0) % 4;
int y = get_local_id(0) / 4;
int width = get_image_width(srcImg);
int index = (gid_0 * 4 + x) + (gid_1 * width / 4 + y);
int2 bi_mvs = as_int2(mvs);
motion_vector_buffer[index] = as_short2(bi_mvs.s0);
if (residuals_buffer != NULL) {
residuals_buffer[index] = sads;
}
shapes_buffer[gid_0 + gid_1 * get_num_groups(0)] = shapes;
}
}
}
__kernel void non_vme_kernel(__global unsigned int *src, __global unsigned int *dst) {
int id = (int)get_global_id(0);
dst[id] = lgamma((float)src[id]);
dst[id] = src[id];
}