static void evergreen_set_global_binding(struct pipe_context *ctx,
unsigned first, unsigned n,
struct pipe_resource **resources,
uint32_t **handles)
{
struct r600_context *rctx = (struct r600_context *)ctx;
struct compute_memory_pool *pool = rctx->screen->global_pool;
struct r600_resource_global **buffers =
(struct r600_resource_global **)resources;
unsigned i;
COMPUTE_DBG(rctx->screen, "*** evergreen_set_global_binding first = %u n = %u\n",
first, n);
if (!resources) {
/* XXX: Unset */
return;
}
/* We mark these items for promotion to the pool if they
* aren't already there */
for (i = first; i < first + n; i++) {
struct compute_memory_item *item = buffers[i]->chunk;
if (!is_item_in_pool(item))
buffers[i]->chunk->status |= ITEM_FOR_PROMOTING;
}
if (compute_memory_finalize_pending(pool, ctx) == -1) {
/* XXX: Unset */
return;
}
for (i = first; i < first + n; i++)
{
uint32_t buffer_offset;
uint32_t handle;
assert(resources[i]->target == PIPE_BUFFER);
assert(resources[i]->bind & PIPE_BIND_GLOBAL);
buffer_offset = util_le32_to_cpu(*(handles[i]));
handle = buffer_offset + buffers[i]->chunk->start_in_dw * 4;
*(handles[i]) = util_cpu_to_le32(handle);
}
/* globals for writing */
evergreen_set_rat(rctx->cs_shader_state.shader, 0, pool->bo, 0, pool->size_in_dw * 4);
/* globals for reading */
evergreen_cs_set_vertex_buffer(rctx, 1, 0,
(struct pipe_resource*)pool->bo);
/* constants for reading, LLVM puts them in text segment */
evergreen_cs_set_vertex_buffer(rctx, 2, 0,
(struct pipe_resource*)rctx->cs_shader_state.shader->code_bo);
}
static void evergreen_set_global_binding(
struct pipe_context *ctx_, unsigned first, unsigned n,
struct pipe_resource **resources,
uint32_t **handles)
{
struct r600_context *ctx = (struct r600_context *)ctx_;
struct compute_memory_pool *pool = ctx->screen->global_pool;
struct r600_resource_global **buffers =
(struct r600_resource_global **)resources;
COMPUTE_DBG("*** evergreen_set_global_binding first = %u n = %u\n",
first, n);
if (!resources) {
/* XXX: Unset */
return;
}
compute_memory_finalize_pending(pool, ctx_);
for (int i = 0; i < n; i++)
{
assert(resources[i]->target == PIPE_BUFFER);
assert(resources[i]->bind & PIPE_BIND_GLOBAL);
*(handles[i]) = buffers[i]->chunk->start_in_dw * 4;
}
evergreen_set_rat(ctx->cs_shader_state.shader, 0, pool->bo, 0, pool->size_in_dw * 4);
evergreen_cs_set_vertex_buffer(ctx, 1, 0,
(struct pipe_resource*)pool->bo);
}
static void evergreen_set_compute_resources(struct pipe_context * ctx_,
unsigned start, unsigned count,
struct pipe_surface ** surfaces)
{
struct r600_context *ctx = (struct r600_context *)ctx_;
struct r600_surface **resources = (struct r600_surface **)surfaces;
COMPUTE_DBG("*** evergreen_set_compute_resources: start = %u count = %u\n",
start, count);
for (int i = 0; i < count; i++) {
/* The First two vertex buffers are reserved for parameters and
* global buffers. */
unsigned vtx_id = 2 + i;
if (resources[i]) {
struct r600_resource_global *buffer =
(struct r600_resource_global*)
resources[i]->base.texture;
if (resources[i]->base.writable) {
assert(i+1 < 12);
evergreen_set_rat(ctx->cs_shader_state.shader, i+1,
(struct r600_resource *)resources[i]->base.texture,
buffer->chunk->start_in_dw*4,
resources[i]->base.texture->width0);
}
evergreen_cs_set_vertex_buffer(ctx, vtx_id,
buffer->chunk->start_in_dw * 4,
resources[i]->base.texture);
}
}
}
/* The kernel parameters are stored a vtx buffer (ID=0), besides the explicit
* kernel parameters there are inplicit parameters that need to be stored
* in the vertex buffer as well. Here is how these parameters are organized in
* the buffer:
*
* DWORDS 0-2: Number of work groups in each dimension (x,y,z)
* DWORDS 3-5: Number of global work items in each dimension (x,y,z)
* DWORDS 6-8: Number of work items within each work group in each dimension
* (x,y,z)
* DWORDS 9+ : Kernel parameters
*/
void evergreen_compute_upload_input(
struct pipe_context *ctx_,
const uint *block_layout,
const uint *grid_layout,
const void *input)
{
struct r600_context *ctx = (struct r600_context *)ctx_;
struct r600_pipe_compute *shader = ctx->cs_shader_state.shader;
int i;
unsigned kernel_parameters_offset_bytes = 36;
uint32_t * num_work_groups_start;
uint32_t * global_size_start;
uint32_t * local_size_start;
uint32_t * kernel_parameters_start;
if (shader->input_size == 0) {
return;
}
if (!shader->kernel_param) {
unsigned buffer_size = shader->input_size;
/* Add space for the grid dimensions */
buffer_size += kernel_parameters_offset_bytes * sizeof(uint);
shader->kernel_param = r600_compute_buffer_alloc_vram(
ctx->screen, buffer_size);
}
num_work_groups_start = ctx->ws->buffer_map(
shader->kernel_param->cs_buf, ctx->cs, PIPE_TRANSFER_WRITE);
global_size_start = num_work_groups_start + (3 * (sizeof(uint) /4));
local_size_start = global_size_start + (3 * (sizeof(uint)) / 4);
kernel_parameters_start = local_size_start + (3 * (sizeof(uint)) / 4);
/* Copy the work group size */
memcpy(num_work_groups_start, grid_layout, 3 * sizeof(uint));
/* Copy the global size */
for (i = 0; i < 3; i++) {
global_size_start[i] = grid_layout[i] * block_layout[i];
}
/* Copy the local dimensions */
memcpy(local_size_start, block_layout, 3 * sizeof(uint));
/* Copy the kernel inputs */
memcpy(kernel_parameters_start, input, shader->input_size);
for (i = 0; i < (kernel_parameters_offset_bytes / 4) +
(shader->input_size / 4); i++) {
COMPUTE_DBG("input %i : %i\n", i,
((unsigned*)num_work_groups_start)[i]);
}
ctx->ws->buffer_unmap(shader->kernel_param->cs_buf);
///ID=0 is reserved for the parameters
evergreen_cs_set_vertex_buffer(ctx, 0, 0,
(struct pipe_resource*)shader->kernel_param);
///ID=0 is reserved for parameters
evergreen_set_const_cache(shader, 0, shader->kernel_param,
shader->input_size, 0);
}
