[Beignet] [PATCH 5/5] add utest compiler_block_motion_estimate_intel for extension cl_intel_device_side_avc_motion_estimation.
xionghu.luo at intel.com
xionghu.luo at intel.com
Tue Jun 13 16:54:16 UTC 2017
From: Luo Xionghu <xionghu.luo at intel.com>
fix build warnings.
Signed-off-by: Chuanbo Weng <chuanbo.weng at intel.com>
Signed-off-by: Xionghu Luo <xionghu.luo at intel.com>
---
kernels/compiler_block_motion_estimate_intel.cl | 76 ++++++++++++
utests/CMakeLists.txt | 1 +
utests/compiler_block_motion_estimate_intel.cpp | 156 ++++++++++++++++++++++++
3 files changed, 233 insertions(+)
create mode 100644 kernels/compiler_block_motion_estimate_intel.cl
create mode 100644 utests/compiler_block_motion_estimate_intel.cpp
diff --git a/kernels/compiler_block_motion_estimate_intel.cl b/kernels/compiler_block_motion_estimate_intel.cl
new file mode 100644
index 0000000..94966a8
--- /dev/null
+++ b/kernels/compiler_block_motion_estimate_intel.cl
@@ -0,0 +1,76 @@
+
+__kernel __attribute__((intel_reqd_sub_group_size(16)))
+void compiler_block_motion_estimate_intel(
+ __read_only image2d_t src_img,
+ __read_only image2d_t ref_img,
+ __global short2* motion_vector_buffer,
+ __global ushort* residuals_buffer,
+ __global uchar* mj_shape_buffer,
+ __global uchar* mn_shapes_buffer,
+ __global uchar* directions_buffer,
+ __global uint* dwo_buffer,
+ __global uint* pld_buffer) {
+
+ int gr_id0 = get_group_id(0);
+ int gr_id1 = get_group_id(1);
+
+ ushort2 src_coord = 0;
+ src_coord.x = gr_id0 * 16;
+ src_coord.y = gr_id1 * 16;
+ uchar partition_mask = CLK_AVC_ME_PARTITION_MASK_16x16_INTEL;
+ uchar sad_adjustment = CLK_AVC_ME_SAD_ADJUST_MODE_NONE_INTEL;
+ intel_sub_group_avc_ime_payload_t payload = intel_sub_group_avc_ime_initialize(src_coord, partition_mask, sad_adjustment);
+ short2 ref_offset = 0;
+ uchar search_window_config = CLK_AVC_ME_SEARCH_WINDOW_EXTRA_TINY_INTEL;
+ payload = intel_sub_group_avc_ime_set_single_reference(ref_offset, search_window_config, payload);
+
+ //mv cost penalty setting
+ ulong packed_cc_delta = 0;
+ uint2 packed_cost_table = intel_sub_group_avc_mce_get_default_medium_penalty_cost_table();
+ uchar cost_precision = CLK_AVC_ME_COST_PRECISION_QPEL_INTEL;
+ payload = intel_sub_group_avc_ime_set_motion_vector_cost_function(
+ packed_cc_delta, packed_cost_table, cost_precision, payload);
+
+ //ime shape penalty
+ ulong packed_shape_cost = (1 << 4 | 2);
+ packed_shape_cost <<= 32;
+ payload = intel_sub_group_avc_ime_set_inter_shape_penalty(packed_shape_cost ,payload);
+
+ sampler_t vs = 0;
+ intel_sub_group_avc_ime_result_t i_result =
+ intel_sub_group_avc_ime_evaluate_with_single_reference(src_img, ref_img, vs, payload);
+
+ //Get ime related result
+ ulong mvs = intel_sub_group_avc_ime_get_motion_vectors(i_result);
+ ushort distortions = intel_sub_group_avc_ime_get_inter_distortions(i_result);
+ uchar major_shape = intel_sub_group_avc_ime_get_inter_major_shape(i_result);
+ uchar minor_shapes = intel_sub_group_avc_ime_get_inter_minor_shapes(i_result);
+ uchar directions = intel_sub_group_avc_ime_get_inter_directions(i_result);
+
+ int lid_x = get_local_id(0);
+ int mb = gr_id0 + gr_id1 * get_num_groups(0);
+ int2 bi_mvs = as_int2(mvs);
+ if(lid_x == 0){
+ motion_vector_buffer[mb] = as_short2(bi_mvs.s0);
+ residuals_buffer[mb] = distortions;
+ mj_shape_buffer[mb] = major_shape;
+ mn_shapes_buffer[mb] = minor_shapes;
+ directions_buffer[mb] = directions;
+ }
+ //fme setting and evaluate
+ uchar pixel_mode = CLK_AVC_ME_SUBPIXEL_MODE_QPEL_INTEL;
+ intel_sub_group_avc_ref_payload_t r_payload =
+ intel_sub_group_avc_fme_initialize(
+ src_coord, mvs, major_shape, minor_shapes,
+ directions, pixel_mode, sad_adjustment);
+ intel_sub_group_avc_ref_result_t r_result =
+ intel_sub_group_avc_ref_evaluate_with_single_reference(src_img, ref_img, vs, r_payload);
+ mvs = intel_sub_group_avc_ref_get_motion_vectors(r_result);
+ distortions = intel_sub_group_avc_ref_get_inter_distortions(r_result);
+
+ dwo_buffer[mb*16*4 + lid_x + 16*0] = i_result.s0;
+ dwo_buffer[mb*16*4 + lid_x + 16*1] = i_result.s1;
+ dwo_buffer[mb*16*4 + lid_x + 16*2] = i_result.s2;
+ dwo_buffer[mb*16*4 + lid_x + 16*3] = i_result.s3;
+
+}
diff --git a/utests/CMakeLists.txt b/utests/CMakeLists.txt
index 61d1326..ee95e00 100644
--- a/utests/CMakeLists.txt
+++ b/utests/CMakeLists.txt
@@ -306,6 +306,7 @@ set (utests_sources
compiler_device_enqueue.cpp
compiler_sqrt_div.cpp
compiler_remove_negative_add.cpp
+ compiler_block_motion_estimate_intel.cpp
compiler_skip_check.cpp
compiler_intra_prediction.cpp)
diff --git a/utests/compiler_block_motion_estimate_intel.cpp b/utests/compiler_block_motion_estimate_intel.cpp
new file mode 100644
index 0000000..d0dbb11
--- /dev/null
+++ b/utests/compiler_block_motion_estimate_intel.cpp
@@ -0,0 +1,156 @@
+#include "utest_helper.hpp"
+#include <string.h>
+
+void compiler_block_motion_estimate_intel(void)
+{
+ if (!cl_check_device_side_avc_motion_estimation()) {
+ return;
+ }
+
+ OCL_CREATE_KERNEL("compiler_block_motion_estimate_intel");
+
+ const size_t w = 80;
+ const size_t h = 48;
+ const size_t mv_w = (w + 15) / 16;
+ const size_t mv_h = (h + 15) / 16;
+
+ cl_image_format format;
+ cl_image_desc desc;
+
+ memset(&desc, 0x0, sizeof(cl_image_desc));
+ memset(&format, 0x0, sizeof(cl_image_format));
+
+ uint8_t* image_data1 = (uint8_t *)malloc(w * h); //src
+ uint8_t* image_data2 = (uint8_t *)malloc(w * h); //ref
+ for (size_t j = 0; j < h; j++) {
+ for (size_t i = 0; i < w; i++) {
+ if (i >= 32 && i <= 47 && j >= 16 && j <= 31)
+ image_data1[w * j + i] = 100;
+ else
+ image_data1[w * j + i] = 0;
+ if (i >= 30 && i <= 45 && j >= 18 && j <= 33)
+ image_data2[w * j + i] = 98;
+ else
+ image_data2[w * j + i] = 0;
+ }
+ }
+
+ format.image_channel_order = CL_R;
+ format.image_channel_data_type = CL_UNORM_INT8;
+ desc.image_type = CL_MEM_OBJECT_IMAGE2D;
+ desc.image_width = w;
+ desc.image_height = h;
+ desc.image_row_pitch = 0;
+ OCL_CREATE_IMAGE(buf[0], CL_MEM_COPY_HOST_PTR, &format, &desc, image_data1); //src
+ OCL_CREATE_IMAGE(buf[1], CL_MEM_COPY_HOST_PTR, &format, &desc, image_data2); //ref
+
+ OCL_CREATE_BUFFER(buf[2], 0, mv_w * mv_h * sizeof(int16_t) * 2, NULL);
+ OCL_CREATE_BUFFER(buf[3], 0, mv_w * mv_h * sizeof(uint16_t), NULL);
+ OCL_CREATE_BUFFER(buf[4], 0, mv_w * mv_h * sizeof(uint8_t), NULL);
+ OCL_CREATE_BUFFER(buf[5], 0, mv_w * mv_h * sizeof(uint8_t), NULL);
+ OCL_CREATE_BUFFER(buf[6], 0, mv_w * mv_h * sizeof(uint8_t), NULL);
+ OCL_CREATE_BUFFER(buf[7], 0, mv_w * mv_h * sizeof(uint32_t) * 16 * 8, NULL);
+ OCL_CREATE_BUFFER(buf[8], 0, mv_w * mv_h * sizeof(uint32_t) * 8 * 8, NULL);
+
+ OCL_SET_ARG(0, sizeof(cl_mem), &buf[0]);
+ OCL_SET_ARG(1, sizeof(cl_mem), &buf[1]);
+ OCL_SET_ARG(2, sizeof(cl_mem), &buf[2]);
+ OCL_SET_ARG(3, sizeof(cl_mem), &buf[3]);
+ OCL_SET_ARG(4, sizeof(cl_mem), &buf[4]);
+ OCL_SET_ARG(5, sizeof(cl_mem), &buf[5]);
+ OCL_SET_ARG(6, sizeof(cl_mem), &buf[6]);
+ OCL_SET_ARG(7, sizeof(cl_mem), &buf[7]);
+ OCL_SET_ARG(8, sizeof(cl_mem), &buf[8]);
+
+ globals[0] = w;
+ globals[1] = h / 16;
+ locals[0] = 16;
+ locals[1] = 1;
+ OCL_NDRANGE(2);
+
+ int16_t expected[] = {-8, -8, //S13.2 fixed point value
+ -8, -8,
+ -8, -8,
+ -8, -8,
+ -8, -8,
+ -8, -8,
+ -8, -8,
+ -8, 4,
+ -8, -8,
+ -8, -8,
+ -8, -8,
+ -8, -8,
+ 4, 4,
+ -8, -8,
+ -8, -8};
+ OCL_MAP_BUFFER(2);
+ OCL_MAP_BUFFER(3);
+ OCL_MAP_BUFFER(4);
+ OCL_MAP_BUFFER(5);
+ OCL_MAP_BUFFER(6);
+ OCL_MAP_BUFFER(7);
+ OCL_MAP_BUFFER(8);
+ int16_t* mv = (int16_t*)buf_data[2];
+#define VME_DEBUG 0
+#if VME_DEBUG
+ uint16_t* residual = (uint16_t*)buf_data[3];
+ uint8_t* major_shape = (uint8_t*)buf_data[4];
+ uint8_t* minor_shape = (uint8_t*)buf_data[5];
+ uint8_t* direction = (uint8_t*)buf_data[6];
+ uint32_t* dwo = (uint32_t*)buf_data[7];
+ uint32_t* pld = (uint32_t*)buf_data[8];
+ std::cout<<std::endl;
+ for (uint32_t j = 0; j <= mv_h - 1; ++j) {
+ for (uint32_t i = 0; i <= mv_w - 1; ++i) {
+ uint32_t mv_num = j * mv_w + i;
+ std::cout<<"******* mv num = "<<mv_num<<": "<<std::endl;
+ std::cout<<"payload register result: "<<std::endl;
+ for(uint32_t row_num = 0; row_num < 8; row_num++){
+ for(int32_t idx = 7; idx >= 0; idx--)
+ printf("%.8x ", pld[mv_num*64 + row_num*8 + idx]);
+ printf("\n");
+ }
+ std::cout<<std::endl;
+ std::cout<<"writeback register result: "<<std::endl;
+ for(uint32_t row_num = 0; row_num < 4; row_num++){
+ for(int32_t wi = 7; wi >= 0; wi--)
+ printf("%.8x ", dwo[mv_num*16*4 + row_num*16 + wi]);
+ printf("\n");
+ for(int32_t wi = 15; wi >= 8; wi--)
+ printf("%.8x ", dwo[mv_num*16*4 + row_num*16 + wi]);
+ printf("\n");
+ }
+ std::cout<<std::endl;
+ std::cout<<"mv: ";
+ std::cout<<"("<<mv[mv_num*2]<<", "<<mv[mv_num*2 + 1]<<") ";
+ std::cout<<std::endl;
+ std::cout<<"residual: ";
+ std::cout<<residual[mv_num]<<" ";
+ std::cout<<std::endl;
+ printf("major shape: %u\n", major_shape[mv_num]);
+ printf("minor shape: %u\n", minor_shape[mv_num]);
+ printf("direction: %u\n", direction[mv_num]);
+ std::cout<<std::endl;
+ }
+ }
+#endif
+ for (uint32_t j = 0; j <= mv_h - 1; ++j) {
+ for (uint32_t i = 0; i <= mv_w - 1; ++i) {
+ uint32_t mv_num = j * mv_w + i;
+ OCL_ASSERT(mv[mv_num*2] == expected[mv_num*2]);
+ OCL_ASSERT(mv[mv_num*2 + 1] == expected[mv_num*2 + 1]);
+ }
+ }
+ OCL_UNMAP_BUFFER(2);
+ OCL_UNMAP_BUFFER(3);
+ OCL_UNMAP_BUFFER(4);
+ OCL_UNMAP_BUFFER(5);
+ OCL_UNMAP_BUFFER(6);
+ OCL_UNMAP_BUFFER(7);
+ OCL_UNMAP_BUFFER(8);
+
+ free(image_data1);
+ free(image_data2);
+}
+
+MAKE_UTEST_FROM_FUNCTION(compiler_block_motion_estimate_intel);
--
2.7.4
More information about the Beignet
mailing list