static void si_texture_transfer_unmap(struct pipe_context *ctx,
struct pipe_transfer* transfer)
{
struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
struct r600_context *rctx = (struct r600_context*)ctx;
struct radeon_winsys_cs_handle *buf;
struct pipe_resource *texture = transfer->resource;
struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
if (rtransfer->staging) {
buf = si_resource(rtransfer->staging)->cs_buf;
} else {
buf = si_resource(transfer->resource)->cs_buf;
}
rctx->ws->buffer_unmap(buf);
if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtransfer->staging) {
if (rtex->is_depth) {
ctx->resource_copy_region(ctx, texture, transfer->level,
transfer->box.x, transfer->box.y, transfer->box.z,
&si_resource(rtransfer->staging)->b.b, transfer->level,
&transfer->box);
} else {
r600_copy_from_staging_texture(ctx, rtransfer);
}
}
if (rtransfer->staging)
pipe_resource_reference((struct pipe_resource**)&rtransfer->staging, NULL);
pipe_resource_reference(&transfer->resource, NULL);
FREE(transfer);
}
static void* si_texture_transfer_map(struct pipe_context *ctx,
struct pipe_transfer* transfer)
{
struct r600_context *rctx = (struct r600_context *)ctx;
struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
struct radeon_winsys_cs_handle *buf;
enum pipe_format format = transfer->resource->format;
unsigned offset = 0;
char *map;
if (rtransfer->staging_texture) {
buf = si_resource(rtransfer->staging_texture)->cs_buf;
} else {
struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
if (rtex->flushed_depth_texture)
buf = rtex->flushed_depth_texture->resource.cs_buf;
else
buf = si_resource(transfer->resource)->cs_buf;
offset = rtransfer->offset +
transfer->box.y / util_format_get_blockheight(format) * transfer->stride +
transfer->box.x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
}
if (!(map = rctx->ws->buffer_map(buf, rctx->cs, transfer->usage))) {
return NULL;
}
return map + offset;
}
static void si_test_vmfault(struct si_screen *sscreen)
{
struct pipe_context *ctx = sscreen->aux_context;
struct si_context *sctx = (struct si_context *)ctx;
struct pipe_resource *buf =
pipe_buffer_create_const0(&sscreen->b, 0, PIPE_USAGE_DEFAULT, 64);
if (!buf) {
puts("Buffer allocation failed.");
exit(1);
}
si_resource(buf)->gpu_address = 0; /* cause a VM fault */
if (sscreen->debug_flags & DBG(TEST_VMFAULT_CP)) {
si_cp_dma_copy_buffer(sctx, buf, buf, 0, 4, 4, 0,
SI_COHERENCY_NONE, L2_BYPASS);
ctx->flush(ctx, NULL, 0);
puts("VM fault test: CP - done.");
}
if (sscreen->debug_flags & DBG(TEST_VMFAULT_SDMA)) {
si_sdma_clear_buffer(sctx, buf, 0, 4, 0);
ctx->flush(ctx, NULL, 0);
puts("VM fault test: SDMA - done.");
}
if (sscreen->debug_flags & DBG(TEST_VMFAULT_SHADER)) {
util_test_constant_buffer(ctx, buf);
puts("VM fault test: Shader - done.");
}
exit(0);
}
static void r600_buffer_destroy(struct pipe_screen *screen,
struct pipe_resource *buf)
{
struct r600_screen *rscreen = (struct r600_screen*)screen;
struct si_resource *rbuffer = si_resource(buf);
pb_reference(&rbuffer->buf, NULL);
FREE(rbuffer);
}
static void si_texture_transfer_unmap(struct pipe_context *ctx,
struct pipe_transfer* transfer)
{
struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
struct r600_context *rctx = (struct r600_context*)ctx;
struct radeon_winsys_cs_handle *buf;
if (rtransfer->staging_texture) {
buf = si_resource(rtransfer->staging_texture)->cs_buf;
} else {
struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
if (rtex->flushed_depth_texture) {
buf = rtex->flushed_depth_texture->resource.cs_buf;
} else {
buf = si_resource(transfer->resource)->cs_buf;
}
}
rctx->ws->buffer_unmap(buf);
}
static void *r600_buffer_transfer_map(struct pipe_context *pipe,
struct pipe_transfer *transfer)
{
struct si_resource *rbuffer = si_resource(transfer->resource);
struct r600_context *rctx = (struct r600_context*)pipe;
uint8_t *data;
data = rctx->ws->buffer_map(rbuffer->cs_buf, rctx->cs, transfer->usage);
if (!data)
return NULL;
return (uint8_t*)data + transfer->box.x;
}
static void si_texture_transfer_unmap(struct pipe_context *ctx,
struct pipe_transfer* transfer)
{
struct r600_transfer *rtransfer = (struct r600_transfer*)transfer;
struct r600_context *rctx = (struct r600_context*)ctx;
struct radeon_winsys_cs_handle *buf;
struct pipe_resource *texture = transfer->resource;
struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
if (rtransfer->staging_texture) {
buf = si_resource(rtransfer->staging_texture)->cs_buf;
} else {
struct r600_resource_texture *rtex = (struct r600_resource_texture*)transfer->resource;
if (rtex->flushed_depth_texture) {
buf = rtex->flushed_depth_texture->resource.cs_buf;
} else {
buf = si_resource(transfer->resource)->cs_buf;
}
}
rctx->ws->buffer_unmap(buf);
if (rtransfer->staging_texture) {
if (transfer->usage & PIPE_TRANSFER_WRITE) {
r600_copy_from_staging_texture(ctx, rtransfer);
}
pipe_resource_reference(&rtransfer->staging_texture, NULL);
}
if (rtex->depth && !rtex->is_flushing_texture) {
if ((transfer->usage & PIPE_TRANSFER_WRITE) && rtex->flushed_depth_texture)
r600_blit_push_depth(ctx, rtex);
}
pipe_resource_reference(&transfer->resource, NULL);
FREE(transfer);
}
static void *si_texture_transfer_map(struct pipe_context *ctx,
struct pipe_resource *texture,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **ptransfer)
{
struct r600_context *rctx = (struct r600_context *)ctx;
struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
struct r600_transfer *trans;
boolean use_staging_texture = FALSE;
struct radeon_winsys_cs_handle *buf;
enum pipe_format format = texture->format;
unsigned offset = 0;
char *map;
/* We cannot map a tiled texture directly because the data is
* in a different order, therefore we do detiling using a blit.
*
* Also, use a temporary in GTT memory for read transfers, as
* the CPU is much happier reading out of cached system memory
* than uncached VRAM.
*/
if (rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR_ALIGNED &&
rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR)
use_staging_texture = TRUE;
/* XXX: Use a staging texture for uploads if the underlying BO
* is busy. No interface for checking that currently? so do
* it eagerly whenever the transfer doesn't require a readback
* and might block.
*/
if ((usage & PIPE_TRANSFER_WRITE) &&
!(usage & (PIPE_TRANSFER_READ |
PIPE_TRANSFER_DONTBLOCK |
PIPE_TRANSFER_UNSYNCHRONIZED)))
use_staging_texture = TRUE;
if (texture->flags & R600_RESOURCE_FLAG_TRANSFER)
use_staging_texture = FALSE;
if (use_staging_texture && (usage & PIPE_TRANSFER_MAP_DIRECTLY))
return NULL;
trans = CALLOC_STRUCT(r600_transfer);
if (trans == NULL)
return NULL;
pipe_resource_reference(&trans->transfer.resource, texture);
trans->transfer.level = level;
trans->transfer.usage = usage;
trans->transfer.box = *box;
if (rtex->is_depth) {
/* XXX: only readback the rectangle which is being mapped?
*/
/* XXX: when discard is true, no need to read back from depth texture
*/
struct r600_resource_texture *staging_depth;
if (!r600_init_flushed_depth_texture(ctx, texture, &staging_depth)) {
R600_ERR("failed to create temporary texture to hold untiled copy\n");
pipe_resource_reference(&trans->transfer.resource, NULL);
FREE(trans);
return NULL;
}
si_blit_uncompress_depth(ctx, rtex, staging_depth,
level, level,
box->z, box->z + box->depth - 1);
trans->transfer.stride = staging_depth->surface.level[level].pitch_bytes;
trans->transfer.layer_stride = staging_depth->surface.level[level].slice_size;
trans->offset = r600_texture_get_offset(staging_depth, level, box->z);
trans->staging = &staging_depth->resource.b.b;
} else if (use_staging_texture) {
struct pipe_resource resource;
struct r600_resource_texture *staging;
memset(&resource, 0, sizeof(resource));
resource.format = texture->format;
resource.width0 = box->width;
resource.height0 = box->height;
resource.depth0 = 1;
resource.array_size = 1;
resource.usage = PIPE_USAGE_STAGING;
resource.flags = R600_RESOURCE_FLAG_TRANSFER;
/* We must set the correct texture target and dimensions if needed for a 3D transfer. */
if (box->depth > 1 && util_max_layer(texture, level) > 0)
resource.target = texture->target;
else
resource.target = PIPE_TEXTURE_2D;
switch (resource.target) {
case PIPE_TEXTURE_1D_ARRAY:
case PIPE_TEXTURE_2D_ARRAY:
case PIPE_TEXTURE_CUBE_ARRAY:
resource.array_size = box->depth;
break;
case PIPE_TEXTURE_3D:
resource.depth0 = box->depth;
break;
default:;
}
/* Create the temporary texture. */
staging = (struct r600_resource_texture*)ctx->screen->resource_create(ctx->screen, &resource);
if (staging == NULL) {
R600_ERR("failed to create temporary texture to hold untiled copy\n");
pipe_resource_reference(&trans->transfer.resource, NULL);
FREE(trans);
return NULL;
}
trans->staging = &staging->resource.b.b;
trans->transfer.stride = staging->surface.level[0].pitch_bytes;
trans->transfer.layer_stride = staging->surface.level[0].slice_size;
if (usage & PIPE_TRANSFER_READ) {
r600_copy_to_staging_texture(ctx, trans);
/* Always referenced in the blit. */
radeonsi_flush(ctx, NULL, 0);
}
} else {
trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
trans->transfer.layer_stride = rtex->surface.level[level].slice_size;
trans->offset = r600_texture_get_offset(rtex, level, box->z);
}
if (trans->staging) {
buf = si_resource(trans->staging)->cs_buf;
} else {
buf = rtex->resource.cs_buf;
}
if (rtex->is_depth || !trans->staging)
offset = trans->offset +
box->y / util_format_get_blockheight(format) * trans->transfer.stride +
box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
if (!(map = rctx->ws->buffer_map(buf, rctx->cs, usage))) {
pipe_resource_reference(&trans->staging, NULL);
pipe_resource_reference(&trans->transfer.resource, NULL);
FREE(trans);
return NULL;
}
*ptransfer = &trans->transfer;
return map + offset;
}
static void *si_texture_transfer_map(struct pipe_context *ctx,
struct pipe_resource *texture,
unsigned level,
unsigned usage,
const struct pipe_box *box,
struct pipe_transfer **ptransfer)
{
struct r600_context *rctx = (struct r600_context *)ctx;
struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
struct pipe_resource resource;
struct r600_transfer *trans;
int r;
boolean use_staging_texture = FALSE;
struct radeon_winsys_cs_handle *buf;
enum pipe_format format = texture->format;
unsigned offset = 0;
char *map;
/* We cannot map a tiled texture directly because the data is
* in a different order, therefore we do detiling using a blit.
*
* Also, use a temporary in GTT memory for read transfers, as
* the CPU is much happier reading out of cached system memory
* than uncached VRAM.
*/
if (rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR_ALIGNED &&
rtex->surface.level[level].mode != RADEON_SURF_MODE_LINEAR)
use_staging_texture = TRUE;
/* XXX: Use a staging texture for uploads if the underlying BO
* is busy. No interface for checking that currently? so do
* it eagerly whenever the transfer doesn't require a readback
* and might block.
*/
if ((usage & PIPE_TRANSFER_WRITE) &&
!(usage & (PIPE_TRANSFER_READ |
PIPE_TRANSFER_DONTBLOCK |
PIPE_TRANSFER_UNSYNCHRONIZED)))
use_staging_texture = TRUE;
if (!permit_hardware_blit(ctx->screen, texture) ||
(texture->flags & R600_RESOURCE_FLAG_TRANSFER))
use_staging_texture = FALSE;
if (use_staging_texture && (usage & PIPE_TRANSFER_MAP_DIRECTLY))
return NULL;
trans = CALLOC_STRUCT(r600_transfer);
if (trans == NULL)
return NULL;
pipe_resource_reference(&trans->transfer.resource, texture);
trans->transfer.level = level;
trans->transfer.usage = usage;
trans->transfer.box = *box;
if (rtex->depth) {
/* XXX: only readback the rectangle which is being mapped?
*/
/* XXX: when discard is true, no need to read back from depth texture
*/
r = r600_texture_depth_flush(ctx, texture, FALSE);
if (r < 0) {
R600_ERR("failed to create temporary texture to hold untiled copy\n");
pipe_resource_reference(&trans->transfer.resource, NULL);
FREE(trans);
return NULL;
}
trans->transfer.stride = rtex->flushed_depth_texture->surface.level[level].pitch_bytes;
trans->offset = r600_texture_get_offset(rtex->flushed_depth_texture, level, box->z);
} else if (use_staging_texture) {
resource.target = PIPE_TEXTURE_2D;
resource.format = texture->format;
resource.width0 = box->width;
resource.height0 = box->height;
resource.depth0 = 1;
resource.array_size = 1;
resource.last_level = 0;
resource.nr_samples = 0;
resource.usage = PIPE_USAGE_STAGING;
resource.bind = 0;
resource.flags = R600_RESOURCE_FLAG_TRANSFER;
/* For texture reading, the temporary (detiled) texture is used as
* a render target when blitting from a tiled texture. */
if (usage & PIPE_TRANSFER_READ) {
resource.bind |= PIPE_BIND_RENDER_TARGET;
}
/* For texture writing, the temporary texture is used as a sampler
* when blitting into a tiled texture. */
if (usage & PIPE_TRANSFER_WRITE) {
resource.bind |= PIPE_BIND_SAMPLER_VIEW;
}
/* Create the temporary texture. */
trans->staging_texture = ctx->screen->resource_create(ctx->screen, &resource);
if (trans->staging_texture == NULL) {
R600_ERR("failed to create temporary texture to hold untiled copy\n");
pipe_resource_reference(&trans->transfer.resource, NULL);
FREE(trans);
return NULL;
}
trans->transfer.stride = ((struct r600_resource_texture *)trans->staging_texture)
->surface.level[0].pitch_bytes;
if (usage & PIPE_TRANSFER_READ) {
r600_copy_to_staging_texture(ctx, trans);
/* Always referenced in the blit. */
radeonsi_flush(ctx, NULL, 0);
}
} else {
trans->transfer.stride = rtex->surface.level[level].pitch_bytes;
trans->transfer.layer_stride = rtex->surface.level[level].slice_size;
trans->offset = r600_texture_get_offset(rtex, level, box->z);
}
if (trans->staging_texture) {
buf = si_resource(trans->staging_texture)->cs_buf;
} else {
struct r600_resource_texture *rtex = (struct r600_resource_texture*)texture;
if (rtex->flushed_depth_texture)
buf = rtex->flushed_depth_texture->resource.cs_buf;
else
buf = si_resource(texture)->cs_buf;
offset = trans->offset +
box->y / util_format_get_blockheight(format) * trans->transfer.stride +
box->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
}
if (!(map = rctx->ws->buffer_map(buf, rctx->cs, usage))) {
pipe_resource_reference(&trans->staging_texture, NULL);
pipe_resource_reference(&trans->transfer.resource, NULL);
FREE(trans);
return NULL;
}
*ptransfer = &trans->transfer;
return map + offset;
}
static struct pipe_context *si_create_context(struct pipe_screen *screen,
unsigned flags)
{
struct si_context *sctx = CALLOC_STRUCT(si_context);
struct si_screen* sscreen = (struct si_screen *)screen;
struct radeon_winsys *ws = sscreen->ws;
int shader, i;
bool stop_exec_on_failure = (flags & PIPE_CONTEXT_LOSE_CONTEXT_ON_RESET) != 0;
if (!sctx)
return NULL;
sctx->has_graphics = sscreen->info.chip_class == SI ||
!(flags & PIPE_CONTEXT_COMPUTE_ONLY);
if (flags & PIPE_CONTEXT_DEBUG)
sscreen->record_llvm_ir = true; /* racy but not critical */
sctx->b.screen = screen; /* this must be set first */
sctx->b.priv = NULL;
sctx->b.destroy = si_destroy_context;
sctx->screen = sscreen; /* Easy accessing of screen/winsys. */
sctx->is_debug = (flags & PIPE_CONTEXT_DEBUG) != 0;
slab_create_child(&sctx->pool_transfers, &sscreen->pool_transfers);
slab_create_child(&sctx->pool_transfers_unsync, &sscreen->pool_transfers);
sctx->ws = sscreen->ws;
sctx->family = sscreen->info.family;
sctx->chip_class = sscreen->info.chip_class;
if (sscreen->info.has_gpu_reset_counter_query) {
sctx->gpu_reset_counter =
sctx->ws->query_value(sctx->ws, RADEON_GPU_RESET_COUNTER);
}
if (sctx->chip_class == CIK ||
sctx->chip_class == VI ||
sctx->chip_class == GFX9) {
sctx->eop_bug_scratch = si_resource(
pipe_buffer_create(&sscreen->b, 0, PIPE_USAGE_DEFAULT,
16 * sscreen->info.num_render_backends));
if (!sctx->eop_bug_scratch)
goto fail;
}
/* Initialize context allocators. */
sctx->allocator_zeroed_memory =
u_suballocator_create(&sctx->b, 128 * 1024,
0, PIPE_USAGE_DEFAULT,
SI_RESOURCE_FLAG_UNMAPPABLE |
SI_RESOURCE_FLAG_CLEAR, false);
if (!sctx->allocator_zeroed_memory)
goto fail;
sctx->b.stream_uploader = u_upload_create(&sctx->b, 1024 * 1024,
0, PIPE_USAGE_STREAM,
SI_RESOURCE_FLAG_READ_ONLY);
if (!sctx->b.stream_uploader)
goto fail;
sctx->cached_gtt_allocator = u_upload_create(&sctx->b, 16 * 1024,
0, PIPE_USAGE_STAGING, 0);
if (!sctx->cached_gtt_allocator)
goto fail;
sctx->ctx = sctx->ws->ctx_create(sctx->ws);
if (!sctx->ctx)
goto fail;
if (sscreen->info.num_sdma_rings && !(sscreen->debug_flags & DBG(NO_ASYNC_DMA))) {
sctx->dma_cs = sctx->ws->cs_create(sctx->ctx, RING_DMA,
(void*)si_flush_dma_cs,
sctx, stop_exec_on_failure);
}
bool use_sdma_upload = sscreen->info.has_dedicated_vram && sctx->dma_cs;
sctx->b.const_uploader = u_upload_create(&sctx->b, 256 * 1024,
0, PIPE_USAGE_DEFAULT,
SI_RESOURCE_FLAG_32BIT |
(use_sdma_upload ?
SI_RESOURCE_FLAG_UPLOAD_FLUSH_EXPLICIT_VIA_SDMA :
(sscreen->cpdma_prefetch_writes_memory ?
0 : SI_RESOURCE_FLAG_READ_ONLY)));
if (!sctx->b.const_uploader)
goto fail;
if (use_sdma_upload)
u_upload_enable_flush_explicit(sctx->b.const_uploader);
sctx->gfx_cs = ws->cs_create(sctx->ctx,
sctx->has_graphics ? RING_GFX : RING_COMPUTE,
(void*)si_flush_gfx_cs, sctx, stop_exec_on_failure);
/* Border colors. */
sctx->border_color_table = malloc(SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table));
if (!sctx->border_color_table)
goto fail;
sctx->border_color_buffer = si_resource(
pipe_buffer_create(screen, 0, PIPE_USAGE_DEFAULT,
SI_MAX_BORDER_COLORS *
sizeof(*sctx->border_color_table)));
if (!sctx->border_color_buffer)
goto fail;
sctx->border_color_map =
ws->buffer_map(sctx->border_color_buffer->buf,
NULL, PIPE_TRANSFER_WRITE);
if (!sctx->border_color_map)
goto fail;
/* Initialize context functions used by graphics and compute. */
sctx->b.emit_string_marker = si_emit_string_marker;
sctx->b.set_debug_callback = si_set_debug_callback;
sctx->b.set_log_context = si_set_log_context;
sctx->b.set_context_param = si_set_context_param;
sctx->b.get_device_reset_status = si_get_reset_status;
sctx->b.set_device_reset_callback = si_set_device_reset_callback;
si_init_all_descriptors(sctx);
si_init_buffer_functions(sctx);
si_init_clear_functions(sctx);
si_init_blit_functions(sctx);
si_init_compute_functions(sctx);
si_init_compute_blit_functions(sctx);
si_init_debug_functions(sctx);
si_init_fence_functions(sctx);
si_init_state_compute_functions(sctx);
if (sscreen->debug_flags & DBG(FORCE_DMA))
sctx->b.resource_copy_region = sctx->dma_copy;
/* Initialize graphics-only context functions. */
if (sctx->has_graphics) {
si_init_context_texture_functions(sctx);
si_init_query_functions(sctx);
si_init_msaa_functions(sctx);
si_init_shader_functions(sctx);
si_init_state_functions(sctx);
si_init_streamout_functions(sctx);
si_init_viewport_functions(sctx);
sctx->blitter = util_blitter_create(&sctx->b);
if (sctx->blitter == NULL)
goto fail;
sctx->blitter->skip_viewport_restore = true;
si_init_draw_functions(sctx);
}
/* Initialize SDMA functions. */
if (sctx->chip_class >= CIK)
cik_init_sdma_functions(sctx);
else
si_init_dma_functions(sctx);
sctx->sample_mask = 0xffff;
/* Initialize multimedia functions. */
if (sscreen->info.has_hw_decode) {
sctx->b.create_video_codec = si_uvd_create_decoder;
sctx->b.create_video_buffer = si_video_buffer_create;
} else {
sctx->b.create_video_codec = vl_create_decoder;
sctx->b.create_video_buffer = vl_video_buffer_create;
}
if (sctx->chip_class >= GFX9) {
sctx->wait_mem_scratch = si_resource(
pipe_buffer_create(screen, 0, PIPE_USAGE_DEFAULT, 4));
if (!sctx->wait_mem_scratch)
goto fail;
/* Initialize the memory. */
si_cp_write_data(sctx, sctx->wait_mem_scratch, 0, 4,
V_370_MEM, V_370_ME, &sctx->wait_mem_number);
}
/* CIK cannot unbind a constant buffer (S_BUFFER_LOAD doesn't skip loads
* if NUM_RECORDS == 0). We need to use a dummy buffer instead. */
if (sctx->chip_class == CIK) {
sctx->null_const_buf.buffer =
pipe_aligned_buffer_create(screen,
SI_RESOURCE_FLAG_32BIT,
PIPE_USAGE_DEFAULT, 16,
sctx->screen->info.tcc_cache_line_size);
if (!sctx->null_const_buf.buffer)
goto fail;
sctx->null_const_buf.buffer_size = sctx->null_const_buf.buffer->width0;
unsigned start_shader = sctx->has_graphics ? 0 : PIPE_SHADER_COMPUTE;
for (shader = start_shader; shader < SI_NUM_SHADERS; shader++) {
for (i = 0; i < SI_NUM_CONST_BUFFERS; i++) {
sctx->b.set_constant_buffer(&sctx->b, shader, i,
&sctx->null_const_buf);
}
}
si_set_rw_buffer(sctx, SI_HS_CONST_DEFAULT_TESS_LEVELS,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_VS_CONST_INSTANCE_DIVISORS,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_VS_CONST_CLIP_PLANES,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_PS_CONST_POLY_STIPPLE,
&sctx->null_const_buf);
si_set_rw_buffer(sctx, SI_PS_CONST_SAMPLE_POSITIONS,
&sctx->null_const_buf);
}
uint64_t max_threads_per_block;
screen->get_compute_param(screen, PIPE_SHADER_IR_TGSI,
PIPE_COMPUTE_CAP_MAX_THREADS_PER_BLOCK,
&max_threads_per_block);
/* The maximum number of scratch waves. Scratch space isn't divided
* evenly between CUs. The number is only a function of the number of CUs.
* We can decrease the constant to decrease the scratch buffer size.
*
* sctx->scratch_waves must be >= the maximum posible size of
* 1 threadgroup, so that the hw doesn't hang from being unable
* to start any.
*
* The recommended value is 4 per CU at most. Higher numbers don't
* bring much benefit, but they still occupy chip resources (think
* async compute). I've seen ~2% performance difference between 4 and 32.
*/
sctx->scratch_waves = MAX2(32 * sscreen->info.num_good_compute_units,
max_threads_per_block / 64);
si_init_compiler(sscreen, &sctx->compiler);
/* Bindless handles. */
sctx->tex_handles = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
sctx->img_handles = _mesa_hash_table_create(NULL, _mesa_hash_pointer,
_mesa_key_pointer_equal);
util_dynarray_init(&sctx->resident_tex_handles, NULL);
util_dynarray_init(&sctx->resident_img_handles, NULL);
util_dynarray_init(&sctx->resident_tex_needs_color_decompress, NULL);
util_dynarray_init(&sctx->resident_img_needs_color_decompress, NULL);
util_dynarray_init(&sctx->resident_tex_needs_depth_decompress, NULL);
sctx->sample_pos_buffer =
pipe_buffer_create(sctx->b.screen, 0, PIPE_USAGE_DEFAULT,
sizeof(sctx->sample_positions));
pipe_buffer_write(&sctx->b, sctx->sample_pos_buffer, 0,
sizeof(sctx->sample_positions), &sctx->sample_positions);
/* this must be last */
si_begin_new_gfx_cs(sctx);
if (sctx->chip_class == CIK) {
/* Clear the NULL constant buffer, because loads should return zeros.
* Note that this forces CP DMA to be used, because clover deadlocks
* for some reason when the compute codepath is used.
*/
uint32_t clear_value = 0;
si_clear_buffer(sctx, sctx->null_const_buf.buffer, 0,
sctx->null_const_buf.buffer->width0,
&clear_value, 4, SI_COHERENCY_SHADER, true);
}
return &sctx->b;
fail:
fprintf(stderr, "radeonsi: Failed to create a context.\n");
si_destroy_context(&sctx->b);
return NULL;
}