#include "ac_debug.h"
+enum {
+ RADV_PREFETCH_VBO_DESCRIPTORS = (1 << 0),
+ RADV_PREFETCH_VS = (1 << 1),
+ RADV_PREFETCH_TCS = (1 << 2),
+ RADV_PREFETCH_TES = (1 << 3),
+ RADV_PREFETCH_GS = (1 << 4),
+ RADV_PREFETCH_PS = (1 << 5),
+ RADV_PREFETCH_SHADERS = (RADV_PREFETCH_VS |
+ RADV_PREFETCH_TCS |
+ RADV_PREFETCH_TES |
+ RADV_PREFETCH_GS |
+ RADV_PREFETCH_PS)
+};
+
static void radv_handle_image_transition(struct radv_cmd_buffer *cmd_buffer,
struct radv_image *image,
VkImageLayout src_layout,
},
};
-void
-radv_dynamic_state_copy(struct radv_dynamic_state *dest,
- const struct radv_dynamic_state *src,
- uint32_t copy_mask)
-{
- if (copy_mask & (1 << VK_DYNAMIC_STATE_VIEWPORT)) {
- dest->viewport.count = src->viewport.count;
- typed_memcpy(dest->viewport.viewports, src->viewport.viewports,
- src->viewport.count);
+static void
+radv_bind_dynamic_state(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_dynamic_state *src)
+{
+ struct radv_dynamic_state *dest = &cmd_buffer->state.dynamic;
+ uint32_t copy_mask = src->mask;
+ uint32_t dest_mask = 0;
+
+ /* Make sure to copy the number of viewports/scissors because they can
+ * only be specified at pipeline creation time.
+ */
+ dest->viewport.count = src->viewport.count;
+ dest->scissor.count = src->scissor.count;
+ dest->discard_rectangle.count = src->discard_rectangle.count;
+
+ if (copy_mask & RADV_DYNAMIC_VIEWPORT) {
+ if (memcmp(&dest->viewport.viewports, &src->viewport.viewports,
+ src->viewport.count * sizeof(VkViewport))) {
+ typed_memcpy(dest->viewport.viewports,
+ src->viewport.viewports,
+ src->viewport.count);
+ dest_mask |= RADV_DYNAMIC_VIEWPORT;
+ }
+ }
+
+ if (copy_mask & RADV_DYNAMIC_SCISSOR) {
+ if (memcmp(&dest->scissor.scissors, &src->scissor.scissors,
+ src->scissor.count * sizeof(VkRect2D))) {
+ typed_memcpy(dest->scissor.scissors,
+ src->scissor.scissors, src->scissor.count);
+ dest_mask |= RADV_DYNAMIC_SCISSOR;
+ }
+ }
+
+ if (copy_mask & RADV_DYNAMIC_LINE_WIDTH) {
+ if (dest->line_width != src->line_width) {
+ dest->line_width = src->line_width;
+ dest_mask |= RADV_DYNAMIC_LINE_WIDTH;
+ }
}
- if (copy_mask & (1 << VK_DYNAMIC_STATE_SCISSOR)) {
- dest->scissor.count = src->scissor.count;
- typed_memcpy(dest->scissor.scissors, src->scissor.scissors,
- src->scissor.count);
+ if (copy_mask & RADV_DYNAMIC_DEPTH_BIAS) {
+ if (memcmp(&dest->depth_bias, &src->depth_bias,
+ sizeof(src->depth_bias))) {
+ dest->depth_bias = src->depth_bias;
+ dest_mask |= RADV_DYNAMIC_DEPTH_BIAS;
+ }
}
- if (copy_mask & (1 << VK_DYNAMIC_STATE_LINE_WIDTH))
- dest->line_width = src->line_width;
+ if (copy_mask & RADV_DYNAMIC_BLEND_CONSTANTS) {
+ if (memcmp(&dest->blend_constants, &src->blend_constants,
+ sizeof(src->blend_constants))) {
+ typed_memcpy(dest->blend_constants,
+ src->blend_constants, 4);
+ dest_mask |= RADV_DYNAMIC_BLEND_CONSTANTS;
+ }
+ }
- if (copy_mask & (1 << VK_DYNAMIC_STATE_DEPTH_BIAS))
- dest->depth_bias = src->depth_bias;
+ if (copy_mask & RADV_DYNAMIC_DEPTH_BOUNDS) {
+ if (memcmp(&dest->depth_bounds, &src->depth_bounds,
+ sizeof(src->depth_bounds))) {
+ dest->depth_bounds = src->depth_bounds;
+ dest_mask |= RADV_DYNAMIC_DEPTH_BOUNDS;
+ }
+ }
- if (copy_mask & (1 << VK_DYNAMIC_STATE_BLEND_CONSTANTS))
- typed_memcpy(dest->blend_constants, src->blend_constants, 4);
+ if (copy_mask & RADV_DYNAMIC_STENCIL_COMPARE_MASK) {
+ if (memcmp(&dest->stencil_compare_mask,
+ &src->stencil_compare_mask,
+ sizeof(src->stencil_compare_mask))) {
+ dest->stencil_compare_mask = src->stencil_compare_mask;
+ dest_mask |= RADV_DYNAMIC_STENCIL_COMPARE_MASK;
+ }
+ }
- if (copy_mask & (1 << VK_DYNAMIC_STATE_DEPTH_BOUNDS))
- dest->depth_bounds = src->depth_bounds;
+ if (copy_mask & RADV_DYNAMIC_STENCIL_WRITE_MASK) {
+ if (memcmp(&dest->stencil_write_mask, &src->stencil_write_mask,
+ sizeof(src->stencil_write_mask))) {
+ dest->stencil_write_mask = src->stencil_write_mask;
+ dest_mask |= RADV_DYNAMIC_STENCIL_WRITE_MASK;
+ }
+ }
- if (copy_mask & (1 << VK_DYNAMIC_STATE_STENCIL_COMPARE_MASK))
- dest->stencil_compare_mask = src->stencil_compare_mask;
+ if (copy_mask & RADV_DYNAMIC_STENCIL_REFERENCE) {
+ if (memcmp(&dest->stencil_reference, &src->stencil_reference,
+ sizeof(src->stencil_reference))) {
+ dest->stencil_reference = src->stencil_reference;
+ dest_mask |= RADV_DYNAMIC_STENCIL_REFERENCE;
+ }
+ }
- if (copy_mask & (1 << VK_DYNAMIC_STATE_STENCIL_WRITE_MASK))
- dest->stencil_write_mask = src->stencil_write_mask;
+ if (copy_mask & RADV_DYNAMIC_DISCARD_RECTANGLE) {
+ if (memcmp(&dest->discard_rectangle.rectangles, &src->discard_rectangle.rectangles,
+ src->discard_rectangle.count * sizeof(VkRect2D))) {
+ typed_memcpy(dest->discard_rectangle.rectangles,
+ src->discard_rectangle.rectangles,
+ src->discard_rectangle.count);
+ dest_mask |= RADV_DYNAMIC_DISCARD_RECTANGLE;
+ }
+ }
- if (copy_mask & (1 << VK_DYNAMIC_STATE_STENCIL_REFERENCE))
- dest->stencil_reference = src->stencil_reference;
+ cmd_buffer->state.dirty |= dest_mask;
}
bool radv_cmd_buffer_uses_mec(struct radv_cmd_buffer *cmd_buffer)
{
struct radv_cmd_buffer *cmd_buffer;
unsigned ring;
- cmd_buffer = vk_alloc(&pool->alloc, sizeof(*cmd_buffer), 8,
- VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ cmd_buffer = vk_zalloc(&pool->alloc, sizeof(*cmd_buffer), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
if (cmd_buffer == NULL)
return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
- memset(cmd_buffer, 0, sizeof(*cmd_buffer));
cmd_buffer->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
cmd_buffer->device = device;
cmd_buffer->pool = pool;
cmd_buffer->queue_family_index = pool->queue_family_index;
} else {
- /* Init the pool_link so we can safefly call list_del when we destroy
+ /* Init the pool_link so we can safely call list_del when we destroy
* the command buffer
*/
list_inithead(&cmd_buffer->pool_link);
*pCommandBuffer = radv_cmd_buffer_to_handle(cmd_buffer);
- cmd_buffer->upload.offset = 0;
- cmd_buffer->upload.size = 0;
list_inithead(&cmd_buffer->upload.list);
return VK_SUCCESS;
if (cmd_buffer->upload.upload_bo)
cmd_buffer->device->ws->buffer_destroy(cmd_buffer->upload.upload_bo);
cmd_buffer->device->ws->cs_destroy(cmd_buffer->cs);
- free(cmd_buffer->push_descriptors.set.mapped_ptr);
+
+ for (unsigned i = 0; i < VK_PIPELINE_BIND_POINT_RANGE_SIZE; i++)
+ free(cmd_buffer->descriptors[i].push_set.set.mapped_ptr);
+
vk_free(&cmd_buffer->pool->alloc, cmd_buffer);
}
cmd_buffer->sample_positions_needed = false;
if (cmd_buffer->upload.upload_bo)
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs,
- cmd_buffer->upload.upload_bo, 8);
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs,
+ cmd_buffer->upload.upload_bo, 8);
cmd_buffer->upload.offset = 0;
cmd_buffer->record_result = VK_SUCCESS;
cmd_buffer->ring_offsets_idx = -1;
+ for (unsigned i = 0; i < VK_PIPELINE_BIND_POINT_RANGE_SIZE; i++) {
+ cmd_buffer->descriptors[i].dirty = 0;
+ cmd_buffer->descriptors[i].valid = 0;
+ cmd_buffer->descriptors[i].push_dirty = false;
+ }
+
if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
void *fence_ptr;
radv_cmd_buffer_upload_alloc(cmd_buffer, 8, 0,
cmd_buffer->gfx9_fence_bo = cmd_buffer->upload.upload_bo;
}
+ cmd_buffer->status = RADV_CMD_BUFFER_STATUS_INITIAL;
+
return cmd_buffer->record_result;
}
bo = device->ws->buffer_create(device->ws,
new_size, 4096,
RADEON_DOMAIN_GTT,
- RADEON_FLAG_CPU_ACCESS);
+ RADEON_FLAG_CPU_ACCESS|
+ RADEON_FLAG_NO_INTERPROCESS_SHARING |
+ RADEON_FLAG_32BIT);
if (!bo) {
cmd_buffer->record_result = VK_ERROR_OUT_OF_DEVICE_MEMORY;
return false;
}
- device->ws->cs_add_buffer(cmd_buffer->cs, bo, 8);
+ radv_cs_add_buffer(device->ws, cmd_buffer->cs, bo, 8);
if (cmd_buffer->upload.upload_bo) {
upload = malloc(sizeof(*upload));
struct radeon_winsys_cs *cs = cmd_buffer->cs;
uint64_t va;
- if (!device->trace_bo)
- return;
-
- va = device->ws->buffer_get_va(device->trace_bo);
+ va = radv_buffer_get_va(device->trace_bo);
if (cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_SECONDARY)
va += 4;
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 7);
++cmd_buffer->state.trace_id;
- device->ws->cs_add_buffer(cs, device->trace_bo, 8);
+ radv_cs_add_buffer(device->ws, cs, device->trace_bo, 8);
radv_emit_write_data_packet(cs, va, 1, &cmd_buffer->state.trace_id);
radeon_emit(cs, PKT3(PKT3_NOP, 0, 0));
radeon_emit(cs, AC_ENCODE_TRACE_POINT(cmd_buffer->state.trace_id));
}
static void
-radv_cmd_buffer_after_draw(struct radv_cmd_buffer *cmd_buffer)
+radv_cmd_buffer_after_draw(struct radv_cmd_buffer *cmd_buffer,
+ enum radv_cmd_flush_bits flags)
{
- if (cmd_buffer->device->debug_flags & RADV_DEBUG_SYNC_SHADERS) {
- enum radv_cmd_flush_bits flags;
+ if (cmd_buffer->device->instance->debug_flags & RADV_DEBUG_SYNC_SHADERS) {
+ uint32_t *ptr = NULL;
+ uint64_t va = 0;
+
+ assert(flags & (RADV_CMD_FLAG_PS_PARTIAL_FLUSH |
+ RADV_CMD_FLAG_CS_PARTIAL_FLUSH));
- /* Force wait for graphics/compute engines to be idle. */
- flags = RADV_CMD_FLAG_PS_PARTIAL_FLUSH |
- RADV_CMD_FLAG_CS_PARTIAL_FLUSH;
+ if (cmd_buffer->device->physical_device->rad_info.chip_class == GFX9) {
+ va = radv_buffer_get_va(cmd_buffer->gfx9_fence_bo) +
+ cmd_buffer->gfx9_fence_offset;
+ ptr = &cmd_buffer->gfx9_fence_idx;
+ }
- si_cs_emit_cache_flush(cmd_buffer->cs, false,
+ /* Force wait for graphics or compute engines to be idle. */
+ si_cs_emit_cache_flush(cmd_buffer->cs,
cmd_buffer->device->physical_device->rad_info.chip_class,
- NULL, 0,
+ ptr, va,
radv_cmd_buffer_uses_mec(cmd_buffer),
flags);
}
- radv_cmd_buffer_trace_emit(cmd_buffer);
+ if (unlikely(cmd_buffer->device->trace_bo))
+ radv_cmd_buffer_trace_emit(cmd_buffer);
}
static void
uint32_t data[2];
uint64_t va;
- if (!device->trace_bo)
- return;
-
- va = device->ws->buffer_get_va(device->trace_bo);
+ va = radv_buffer_get_va(device->trace_bo);
switch (ring) {
case RING_GFX:
data[0] = (uintptr_t)pipeline;
data[1] = (uintptr_t)pipeline >> 32;
- device->ws->cs_add_buffer(cs, device->trace_bo, 8);
+ radv_cs_add_buffer(device->ws, cs, device->trace_bo, 8);
radv_emit_write_data_packet(cs, va, 2, data);
}
-static void
-radv_emit_graphics_blend_state(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
+void radv_set_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
+ VkPipelineBindPoint bind_point,
+ struct radv_descriptor_set *set,
+ unsigned idx)
{
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028780_CB_BLEND0_CONTROL, 8);
- radeon_emit_array(cmd_buffer->cs, pipeline->graphics.blend.cb_blend_control,
- 8);
- radeon_set_context_reg(cmd_buffer->cs, R_028808_CB_COLOR_CONTROL, pipeline->graphics.blend.cb_color_control);
- radeon_set_context_reg(cmd_buffer->cs, R_028B70_DB_ALPHA_TO_MASK, pipeline->graphics.blend.db_alpha_to_mask);
-
- if (cmd_buffer->device->physical_device->has_rbplus) {
-
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028760_SX_MRT0_BLEND_OPT, 8);
- radeon_emit_array(cmd_buffer->cs, pipeline->graphics.blend.sx_mrt_blend_opt, 8);
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, bind_point);
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028754_SX_PS_DOWNCONVERT, 3);
- radeon_emit(cmd_buffer->cs, 0); /* R_028754_SX_PS_DOWNCONVERT */
- radeon_emit(cmd_buffer->cs, 0); /* R_028758_SX_BLEND_OPT_EPSILON */
- radeon_emit(cmd_buffer->cs, 0); /* R_02875C_SX_BLEND_OPT_CONTROL */
- }
+ descriptors_state->sets[idx] = set;
+ if (set)
+ descriptors_state->valid |= (1u << idx);
+ else
+ descriptors_state->valid &= ~(1u << idx);
+ descriptors_state->dirty |= (1u << idx);
}
static void
-radv_emit_graphics_depth_stencil_state(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
+radv_save_descriptors(struct radv_cmd_buffer *cmd_buffer,
+ VkPipelineBindPoint bind_point)
{
- struct radv_depth_stencil_state *ds = &pipeline->graphics.ds;
- radeon_set_context_reg(cmd_buffer->cs, R_028800_DB_DEPTH_CONTROL, ds->db_depth_control);
- radeon_set_context_reg(cmd_buffer->cs, R_02842C_DB_STENCIL_CONTROL, ds->db_stencil_control);
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, bind_point);
+ struct radv_device *device = cmd_buffer->device;
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
+ uint32_t data[MAX_SETS * 2] = {};
+ uint64_t va;
+ unsigned i;
+ va = radv_buffer_get_va(device->trace_bo) + 24;
- radeon_set_context_reg(cmd_buffer->cs, R_028000_DB_RENDER_CONTROL, ds->db_render_control);
- radeon_set_context_reg(cmd_buffer->cs, R_028010_DB_RENDER_OVERRIDE2, ds->db_render_override2);
-}
+ MAYBE_UNUSED unsigned cdw_max = radeon_check_space(device->ws,
+ cmd_buffer->cs, 4 + MAX_SETS * 2);
-/* 12.4 fixed-point */
-static unsigned radv_pack_float_12p4(float x)
-{
- return x <= 0 ? 0 :
- x >= 4096 ? 0xffff : x * 16;
+ for_each_bit(i, descriptors_state->valid) {
+ struct radv_descriptor_set *set = descriptors_state->sets[i];
+ data[i * 2] = (uintptr_t)set;
+ data[i * 2 + 1] = (uintptr_t)set >> 32;
+ }
+
+ radv_cs_add_buffer(device->ws, cs, device->trace_bo, 8);
+ radv_emit_write_data_packet(cs, va, MAX_SETS * 2, data);
}
-struct ac_userdata_info *
+struct radv_userdata_info *
radv_lookup_user_sgpr(struct radv_pipeline *pipeline,
gl_shader_stage stage,
int idx)
{
+ if (stage == MESA_SHADER_VERTEX) {
+ if (pipeline->shaders[MESA_SHADER_VERTEX])
+ return &pipeline->shaders[MESA_SHADER_VERTEX]->info.user_sgprs_locs.shader_data[idx];
+ if (pipeline->shaders[MESA_SHADER_TESS_CTRL])
+ return &pipeline->shaders[MESA_SHADER_TESS_CTRL]->info.user_sgprs_locs.shader_data[idx];
+ if (pipeline->shaders[MESA_SHADER_GEOMETRY])
+ return &pipeline->shaders[MESA_SHADER_GEOMETRY]->info.user_sgprs_locs.shader_data[idx];
+ } else if (stage == MESA_SHADER_TESS_EVAL) {
+ if (pipeline->shaders[MESA_SHADER_TESS_EVAL])
+ return &pipeline->shaders[MESA_SHADER_TESS_EVAL]->info.user_sgprs_locs.shader_data[idx];
+ if (pipeline->shaders[MESA_SHADER_GEOMETRY])
+ return &pipeline->shaders[MESA_SHADER_GEOMETRY]->info.user_sgprs_locs.shader_data[idx];
+ }
return &pipeline->shaders[stage]->info.user_sgprs_locs.shader_data[idx];
}
gl_shader_stage stage,
int idx, uint64_t va)
{
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(pipeline, stage, idx);
- uint32_t base_reg = radv_shader_stage_to_user_data_0(stage, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
+ struct radv_userdata_info *loc = radv_lookup_user_sgpr(pipeline, stage, idx);
+ uint32_t base_reg = pipeline->user_data_0[stage];
if (loc->sgpr_idx == -1)
return;
assert(loc->num_sgprs == 2);
assert(!loc->indirect);
- radeon_set_sh_reg_seq(cmd_buffer->cs, base_reg + loc->sgpr_idx * 4, 2);
- radeon_emit(cmd_buffer->cs, va);
- radeon_emit(cmd_buffer->cs, va >> 32);
+
+ radv_emit_shader_pointer(cmd_buffer->cs,
+ base_reg + loc->sgpr_idx * 4, va);
}
static void
struct radv_multisample_state *ms = &pipeline->graphics.ms;
struct radv_pipeline *old_pipeline = cmd_buffer->state.emitted_pipeline;
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028C38_PA_SC_AA_MASK_X0Y0_X1Y0, 2);
- radeon_emit(cmd_buffer->cs, ms->pa_sc_aa_mask[0]);
- radeon_emit(cmd_buffer->cs, ms->pa_sc_aa_mask[1]);
-
- radeon_set_context_reg(cmd_buffer->cs, CM_R_028804_DB_EQAA, ms->db_eqaa);
- radeon_set_context_reg(cmd_buffer->cs, EG_R_028A4C_PA_SC_MODE_CNTL_1, ms->pa_sc_mode_cntl_1);
+ if (pipeline->shaders[MESA_SHADER_FRAGMENT]->info.info.ps.needs_sample_positions)
+ cmd_buffer->sample_positions_needed = true;
if (old_pipeline && num_samples == old_pipeline->graphics.ms.num_samples)
return;
- radeon_set_context_reg_seq(cmd_buffer->cs, CM_R_028BDC_PA_SC_LINE_CNTL, 2);
+ radeon_set_context_reg_seq(cmd_buffer->cs, R_028BDC_PA_SC_LINE_CNTL, 2);
radeon_emit(cmd_buffer->cs, ms->pa_sc_line_cntl);
radeon_emit(cmd_buffer->cs, ms->pa_sc_aa_config);
+ radeon_set_context_reg(cmd_buffer->cs, R_028A48_PA_SC_MODE_CNTL_0, ms->pa_sc_mode_cntl_0);
+
radv_cayman_emit_msaa_sample_locs(cmd_buffer->cs, num_samples);
/* GFX9: Flush DFSM when the AA mode changes. */
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
+ if (cmd_buffer->device->dfsm_allowed) {
radeon_emit(cmd_buffer->cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
radeon_emit(cmd_buffer->cs, EVENT_TYPE(V_028A90_FLUSH_DFSM) | EVENT_INDEX(0));
}
- if (pipeline->shaders[MESA_SHADER_FRAGMENT]->info.info.ps.needs_sample_positions) {
- uint32_t offset;
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_FRAGMENT, AC_UD_PS_SAMPLE_POS_OFFSET);
- uint32_t base_reg = radv_shader_stage_to_user_data_0(MESA_SHADER_FRAGMENT, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
- if (loc->sgpr_idx == -1)
- return;
- assert(loc->num_sgprs == 1);
- assert(!loc->indirect);
- switch (num_samples) {
- default:
- offset = 0;
- break;
- case 2:
- offset = 1;
- break;
- case 4:
- offset = 3;
- break;
- case 8:
- offset = 7;
- break;
- case 16:
- offset = 15;
- break;
- }
-
- radeon_set_sh_reg(cmd_buffer->cs, base_reg + loc->sgpr_idx * 4, offset);
- cmd_buffer->sample_positions_needed = true;
- }
-}
-
-static void
-radv_emit_graphics_raster_state(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
-{
- struct radv_raster_state *raster = &pipeline->graphics.raster;
-
- radeon_set_context_reg(cmd_buffer->cs, R_028810_PA_CL_CLIP_CNTL,
- raster->pa_cl_clip_cntl);
-
- radeon_set_context_reg(cmd_buffer->cs, R_0286D4_SPI_INTERP_CONTROL_0,
- raster->spi_interp_control);
-
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028A00_PA_SU_POINT_SIZE, 2);
- unsigned tmp = (unsigned)(1.0 * 8.0);
- radeon_emit(cmd_buffer->cs, S_028A00_HEIGHT(tmp) | S_028A00_WIDTH(tmp));
- radeon_emit(cmd_buffer->cs, S_028A04_MIN_SIZE(radv_pack_float_12p4(0)) |
- S_028A04_MAX_SIZE(radv_pack_float_12p4(8192/2))); /* R_028A04_PA_SU_POINT_MINMAX */
-
- radeon_set_context_reg(cmd_buffer->cs, R_028BE4_PA_SU_VTX_CNTL,
- raster->pa_su_vtx_cntl);
-
- radeon_set_context_reg(cmd_buffer->cs, R_028814_PA_SU_SC_MODE_CNTL,
- raster->pa_su_sc_mode_cntl);
-}
-
-static inline void
-radv_emit_prefetch(struct radv_cmd_buffer *cmd_buffer, uint64_t va,
- unsigned size)
-{
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK)
- si_cp_dma_prefetch(cmd_buffer, va, size);
-}
-
-static void
-radv_emit_hw_vs(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline,
- struct radv_shader_variant *shader,
- struct ac_vs_output_info *outinfo)
-{
- struct radeon_winsys *ws = cmd_buffer->device->ws;
- uint64_t va = ws->buffer_get_va(shader->bo) + shader->bo_offset;
- unsigned export_count;
-
- ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
- radv_emit_prefetch(cmd_buffer, va, shader->code_size);
-
- export_count = MAX2(1, outinfo->param_exports);
- radeon_set_context_reg(cmd_buffer->cs, R_0286C4_SPI_VS_OUT_CONFIG,
- S_0286C4_VS_EXPORT_COUNT(export_count - 1));
-
- radeon_set_context_reg(cmd_buffer->cs, R_02870C_SPI_SHADER_POS_FORMAT,
- S_02870C_POS0_EXPORT_FORMAT(V_02870C_SPI_SHADER_4COMP) |
- S_02870C_POS1_EXPORT_FORMAT(outinfo->pos_exports > 1 ?
- V_02870C_SPI_SHADER_4COMP :
- V_02870C_SPI_SHADER_NONE) |
- S_02870C_POS2_EXPORT_FORMAT(outinfo->pos_exports > 2 ?
- V_02870C_SPI_SHADER_4COMP :
- V_02870C_SPI_SHADER_NONE) |
- S_02870C_POS3_EXPORT_FORMAT(outinfo->pos_exports > 3 ?
- V_02870C_SPI_SHADER_4COMP :
- V_02870C_SPI_SHADER_NONE));
-
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B120_SPI_SHADER_PGM_LO_VS, 4);
- radeon_emit(cmd_buffer->cs, va >> 8);
- radeon_emit(cmd_buffer->cs, va >> 40);
- radeon_emit(cmd_buffer->cs, shader->rsrc1);
- radeon_emit(cmd_buffer->cs, shader->rsrc2);
-
- radeon_set_context_reg(cmd_buffer->cs, R_028818_PA_CL_VTE_CNTL,
- S_028818_VTX_W0_FMT(1) |
- S_028818_VPORT_X_SCALE_ENA(1) | S_028818_VPORT_X_OFFSET_ENA(1) |
- S_028818_VPORT_Y_SCALE_ENA(1) | S_028818_VPORT_Y_OFFSET_ENA(1) |
- S_028818_VPORT_Z_SCALE_ENA(1) | S_028818_VPORT_Z_OFFSET_ENA(1));
-
-
- radeon_set_context_reg(cmd_buffer->cs, R_02881C_PA_CL_VS_OUT_CNTL,
- pipeline->graphics.pa_cl_vs_out_cntl);
-
- if (cmd_buffer->device->physical_device->rad_info.chip_class <= VI)
- radeon_set_context_reg(cmd_buffer->cs, R_028AB4_VGT_REUSE_OFF,
- S_028AB4_REUSE_OFF(outinfo->writes_viewport_index));
-}
-
-static void
-radv_emit_hw_es(struct radv_cmd_buffer *cmd_buffer,
- struct radv_shader_variant *shader,
- struct ac_es_output_info *outinfo)
-{
- struct radeon_winsys *ws = cmd_buffer->device->ws;
- uint64_t va = ws->buffer_get_va(shader->bo) + shader->bo_offset;
-
- ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
- radv_emit_prefetch(cmd_buffer, va, shader->code_size);
-
- radeon_set_context_reg(cmd_buffer->cs, R_028AAC_VGT_ESGS_RING_ITEMSIZE,
- outinfo->esgs_itemsize / 4);
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B320_SPI_SHADER_PGM_LO_ES, 4);
- radeon_emit(cmd_buffer->cs, va >> 8);
- radeon_emit(cmd_buffer->cs, va >> 40);
- radeon_emit(cmd_buffer->cs, shader->rsrc1);
- radeon_emit(cmd_buffer->cs, shader->rsrc2);
-}
-
-static void
-radv_emit_hw_ls(struct radv_cmd_buffer *cmd_buffer,
- struct radv_shader_variant *shader)
-{
- struct radeon_winsys *ws = cmd_buffer->device->ws;
- uint64_t va = ws->buffer_get_va(shader->bo) + shader->bo_offset;
- uint32_t rsrc2 = shader->rsrc2;
-
- ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
- radv_emit_prefetch(cmd_buffer, va, shader->code_size);
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B520_SPI_SHADER_PGM_LO_LS, 2);
- radeon_emit(cmd_buffer->cs, va >> 8);
- radeon_emit(cmd_buffer->cs, va >> 40);
-
- rsrc2 |= S_00B52C_LDS_SIZE(cmd_buffer->state.pipeline->graphics.tess.lds_size);
- if (cmd_buffer->device->physical_device->rad_info.chip_class == CIK &&
- cmd_buffer->device->physical_device->rad_info.family != CHIP_HAWAII)
- radeon_set_sh_reg(cmd_buffer->cs, R_00B52C_SPI_SHADER_PGM_RSRC2_LS, rsrc2);
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B528_SPI_SHADER_PGM_RSRC1_LS, 2);
- radeon_emit(cmd_buffer->cs, shader->rsrc1);
- radeon_emit(cmd_buffer->cs, rsrc2);
-}
-
-static void
-radv_emit_hw_hs(struct radv_cmd_buffer *cmd_buffer,
- struct radv_shader_variant *shader)
-{
- struct radeon_winsys *ws = cmd_buffer->device->ws;
- uint64_t va = ws->buffer_get_va(shader->bo) + shader->bo_offset;
-
- ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
- radv_emit_prefetch(cmd_buffer, va, shader->code_size);
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B420_SPI_SHADER_PGM_LO_HS, 4);
- radeon_emit(cmd_buffer->cs, va >> 8);
- radeon_emit(cmd_buffer->cs, va >> 40);
- radeon_emit(cmd_buffer->cs, shader->rsrc1);
- radeon_emit(cmd_buffer->cs, shader->rsrc2);
}
static void
-radv_emit_vertex_shader(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
+radv_emit_shader_prefetch(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_shader_variant *shader)
{
- struct radv_shader_variant *vs;
-
- assert (pipeline->shaders[MESA_SHADER_VERTEX]);
+ uint64_t va;
- vs = pipeline->shaders[MESA_SHADER_VERTEX];
+ if (!shader)
+ return;
- if (vs->info.vs.as_ls)
- radv_emit_hw_ls(cmd_buffer, vs);
- else if (vs->info.vs.as_es)
- radv_emit_hw_es(cmd_buffer, vs, &vs->info.vs.es_info);
- else
- radv_emit_hw_vs(cmd_buffer, pipeline, vs, &vs->info.vs.outinfo);
+ va = radv_buffer_get_va(shader->bo) + shader->bo_offset;
- radeon_set_context_reg(cmd_buffer->cs, R_028A84_VGT_PRIMITIVEID_EN, pipeline->graphics.vgt_primitiveid_en);
+ si_cp_dma_prefetch(cmd_buffer, va, shader->code_size);
}
-
static void
-radv_emit_tess_shaders(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
+radv_emit_prefetch_L2(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_pipeline *pipeline,
+ bool vertex_stage_only)
{
- if (!radv_pipeline_has_tess(pipeline))
- return;
-
- struct radv_shader_variant *tes, *tcs;
-
- tcs = pipeline->shaders[MESA_SHADER_TESS_CTRL];
- tes = pipeline->shaders[MESA_SHADER_TESS_EVAL];
-
- if (tes->info.tes.as_es)
- radv_emit_hw_es(cmd_buffer, tes, &tes->info.tes.es_info);
- else
- radv_emit_hw_vs(cmd_buffer, pipeline, tes, &tes->info.tes.outinfo);
-
- radv_emit_hw_hs(cmd_buffer, tcs);
+ struct radv_cmd_state *state = &cmd_buffer->state;
+ uint32_t mask = state->prefetch_L2_mask;
- radeon_set_context_reg(cmd_buffer->cs, R_028B6C_VGT_TF_PARAM,
- pipeline->graphics.tess.tf_param);
+ if (vertex_stage_only) {
+ /* Fast prefetch path for starting draws as soon as possible.
+ */
+ mask = state->prefetch_L2_mask & (RADV_PREFETCH_VS |
+ RADV_PREFETCH_VBO_DESCRIPTORS);
+ }
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK)
- radeon_set_context_reg_idx(cmd_buffer->cs, R_028B58_VGT_LS_HS_CONFIG, 2,
- pipeline->graphics.tess.ls_hs_config);
- else
- radeon_set_context_reg(cmd_buffer->cs, R_028B58_VGT_LS_HS_CONFIG,
- pipeline->graphics.tess.ls_hs_config);
+ if (mask & RADV_PREFETCH_VS)
+ radv_emit_shader_prefetch(cmd_buffer,
+ pipeline->shaders[MESA_SHADER_VERTEX]);
- struct ac_userdata_info *loc;
+ if (mask & RADV_PREFETCH_VBO_DESCRIPTORS)
+ si_cp_dma_prefetch(cmd_buffer, state->vb_va, state->vb_size);
- loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_TESS_CTRL, AC_UD_TCS_OFFCHIP_LAYOUT);
- if (loc->sgpr_idx != -1) {
- uint32_t base_reg = radv_shader_stage_to_user_data_0(MESA_SHADER_TESS_CTRL, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
- assert(loc->num_sgprs == 4);
- assert(!loc->indirect);
- radeon_set_sh_reg_seq(cmd_buffer->cs, base_reg + loc->sgpr_idx * 4, 4);
- radeon_emit(cmd_buffer->cs, pipeline->graphics.tess.offchip_layout);
- radeon_emit(cmd_buffer->cs, pipeline->graphics.tess.tcs_out_offsets);
- radeon_emit(cmd_buffer->cs, pipeline->graphics.tess.tcs_out_layout |
- pipeline->graphics.tess.num_tcs_input_cp << 26);
- radeon_emit(cmd_buffer->cs, pipeline->graphics.tess.tcs_in_layout);
- }
+ if (mask & RADV_PREFETCH_TCS)
+ radv_emit_shader_prefetch(cmd_buffer,
+ pipeline->shaders[MESA_SHADER_TESS_CTRL]);
- loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_TESS_EVAL, AC_UD_TES_OFFCHIP_LAYOUT);
- if (loc->sgpr_idx != -1) {
- uint32_t base_reg = radv_shader_stage_to_user_data_0(MESA_SHADER_TESS_EVAL, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
- assert(loc->num_sgprs == 1);
- assert(!loc->indirect);
+ if (mask & RADV_PREFETCH_TES)
+ radv_emit_shader_prefetch(cmd_buffer,
+ pipeline->shaders[MESA_SHADER_TESS_EVAL]);
- radeon_set_sh_reg(cmd_buffer->cs, base_reg + loc->sgpr_idx * 4,
- pipeline->graphics.tess.offchip_layout);
+ if (mask & RADV_PREFETCH_GS) {
+ radv_emit_shader_prefetch(cmd_buffer,
+ pipeline->shaders[MESA_SHADER_GEOMETRY]);
+ radv_emit_shader_prefetch(cmd_buffer, pipeline->gs_copy_shader);
}
- loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_VERTEX, AC_UD_VS_LS_TCS_IN_LAYOUT);
- if (loc->sgpr_idx != -1) {
- uint32_t base_reg = radv_shader_stage_to_user_data_0(MESA_SHADER_VERTEX, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
- assert(loc->num_sgprs == 1);
- assert(!loc->indirect);
+ if (mask & RADV_PREFETCH_PS)
+ radv_emit_shader_prefetch(cmd_buffer,
+ pipeline->shaders[MESA_SHADER_FRAGMENT]);
- radeon_set_sh_reg(cmd_buffer->cs, base_reg + loc->sgpr_idx * 4,
- pipeline->graphics.tess.tcs_in_layout);
- }
+ state->prefetch_L2_mask &= ~mask;
}
static void
-radv_emit_geometry_shader(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
+radv_emit_rbplus_state(struct radv_cmd_buffer *cmd_buffer)
{
- struct radeon_winsys *ws = cmd_buffer->device->ws;
- struct radv_shader_variant *gs;
- uint64_t va;
-
- radeon_set_context_reg(cmd_buffer->cs, R_028A40_VGT_GS_MODE, pipeline->graphics.vgt_gs_mode);
-
- gs = pipeline->shaders[MESA_SHADER_GEOMETRY];
- if (!gs)
+ if (!cmd_buffer->device->physical_device->rbplus_allowed)
return;
- uint32_t gsvs_itemsize = gs->info.gs.max_gsvs_emit_size >> 2;
-
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028A60_VGT_GSVS_RING_OFFSET_1, 3);
- radeon_emit(cmd_buffer->cs, gsvs_itemsize);
- radeon_emit(cmd_buffer->cs, gsvs_itemsize);
- radeon_emit(cmd_buffer->cs, gsvs_itemsize);
-
- radeon_set_context_reg(cmd_buffer->cs, R_028AB0_VGT_GSVS_RING_ITEMSIZE, gsvs_itemsize);
-
- radeon_set_context_reg(cmd_buffer->cs, R_028B38_VGT_GS_MAX_VERT_OUT, gs->info.gs.vertices_out);
-
- uint32_t gs_vert_itemsize = gs->info.gs.gsvs_vertex_size;
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028B5C_VGT_GS_VERT_ITEMSIZE, 4);
- radeon_emit(cmd_buffer->cs, gs_vert_itemsize >> 2);
- radeon_emit(cmd_buffer->cs, 0);
- radeon_emit(cmd_buffer->cs, 0);
- radeon_emit(cmd_buffer->cs, 0);
-
- uint32_t gs_num_invocations = gs->info.gs.invocations;
- radeon_set_context_reg(cmd_buffer->cs, R_028B90_VGT_GS_INSTANCE_CNT,
- S_028B90_CNT(MIN2(gs_num_invocations, 127)) |
- S_028B90_ENABLE(gs_num_invocations > 0));
-
- va = ws->buffer_get_va(gs->bo) + gs->bo_offset;
- ws->cs_add_buffer(cmd_buffer->cs, gs->bo, 8);
- radv_emit_prefetch(cmd_buffer, va, gs->code_size);
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B220_SPI_SHADER_PGM_LO_GS, 4);
- radeon_emit(cmd_buffer->cs, va >> 8);
- radeon_emit(cmd_buffer->cs, va >> 40);
- radeon_emit(cmd_buffer->cs, gs->rsrc1);
- radeon_emit(cmd_buffer->cs, gs->rsrc2);
-
- radv_emit_hw_vs(cmd_buffer, pipeline, pipeline->gs_copy_shader, &pipeline->gs_copy_shader->info.vs.outinfo);
+ struct radv_pipeline *pipeline = cmd_buffer->state.pipeline;
+ struct radv_framebuffer *framebuffer = cmd_buffer->state.framebuffer;
+ const struct radv_subpass *subpass = cmd_buffer->state.subpass;
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(cmd_buffer->state.pipeline, MESA_SHADER_GEOMETRY,
- AC_UD_GS_VS_RING_STRIDE_ENTRIES);
- if (loc->sgpr_idx != -1) {
- uint32_t stride = gs->info.gs.max_gsvs_emit_size;
- uint32_t num_entries = 64;
- bool is_vi = cmd_buffer->device->physical_device->rad_info.chip_class >= VI;
+ unsigned sx_ps_downconvert = 0;
+ unsigned sx_blend_opt_epsilon = 0;
+ unsigned sx_blend_opt_control = 0;
- if (is_vi)
- num_entries *= stride;
+ for (unsigned i = 0; i < subpass->color_count; ++i) {
+ if (subpass->color_attachments[i].attachment == VK_ATTACHMENT_UNUSED)
+ continue;
- stride = S_008F04_STRIDE(stride);
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B230_SPI_SHADER_USER_DATA_GS_0 + loc->sgpr_idx * 4, 2);
- radeon_emit(cmd_buffer->cs, stride);
- radeon_emit(cmd_buffer->cs, num_entries);
- }
-}
+ int idx = subpass->color_attachments[i].attachment;
+ struct radv_color_buffer_info *cb = &framebuffer->attachments[idx].cb;
-static void
-radv_emit_fragment_shader(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
-{
- struct radeon_winsys *ws = cmd_buffer->device->ws;
- struct radv_shader_variant *ps;
- uint64_t va;
- unsigned spi_baryc_cntl = S_0286E0_FRONT_FACE_ALL_BITS(1);
- struct radv_blend_state *blend = &pipeline->graphics.blend;
- assert (pipeline->shaders[MESA_SHADER_FRAGMENT]);
+ unsigned format = G_028C70_FORMAT(cb->cb_color_info);
+ unsigned swap = G_028C70_COMP_SWAP(cb->cb_color_info);
+ uint32_t spi_format = (pipeline->graphics.col_format >> (i * 4)) & 0xf;
+ uint32_t colormask = (pipeline->graphics.cb_target_mask >> (i * 4)) & 0xf;
- ps = pipeline->shaders[MESA_SHADER_FRAGMENT];
- va = ws->buffer_get_va(ps->bo) + ps->bo_offset;
- ws->cs_add_buffer(cmd_buffer->cs, ps->bo, 8);
- radv_emit_prefetch(cmd_buffer, va, ps->code_size);
+ bool has_alpha, has_rgb;
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B020_SPI_SHADER_PGM_LO_PS, 4);
- radeon_emit(cmd_buffer->cs, va >> 8);
- radeon_emit(cmd_buffer->cs, va >> 40);
- radeon_emit(cmd_buffer->cs, ps->rsrc1);
- radeon_emit(cmd_buffer->cs, ps->rsrc2);
+ /* Set if RGB and A are present. */
+ has_alpha = !G_028C74_FORCE_DST_ALPHA_1(cb->cb_color_attrib);
- radeon_set_context_reg(cmd_buffer->cs, R_02880C_DB_SHADER_CONTROL,
- pipeline->graphics.db_shader_control);
+ if (format == V_028C70_COLOR_8 ||
+ format == V_028C70_COLOR_16 ||
+ format == V_028C70_COLOR_32)
+ has_rgb = !has_alpha;
+ else
+ has_rgb = true;
- radeon_set_context_reg(cmd_buffer->cs, R_0286CC_SPI_PS_INPUT_ENA,
- ps->config.spi_ps_input_ena);
+ /* Check the colormask and export format. */
+ if (!(colormask & 0x7))
+ has_rgb = false;
+ if (!(colormask & 0x8))
+ has_alpha = false;
- radeon_set_context_reg(cmd_buffer->cs, R_0286D0_SPI_PS_INPUT_ADDR,
- ps->config.spi_ps_input_addr);
+ if (spi_format == V_028714_SPI_SHADER_ZERO) {
+ has_rgb = false;
+ has_alpha = false;
+ }
- if (ps->info.info.ps.force_persample)
- spi_baryc_cntl |= S_0286E0_POS_FLOAT_LOCATION(2);
+ /* Disable value checking for disabled channels. */
+ if (!has_rgb)
+ sx_blend_opt_control |= S_02875C_MRT0_COLOR_OPT_DISABLE(1) << (i * 4);
+ if (!has_alpha)
+ sx_blend_opt_control |= S_02875C_MRT0_ALPHA_OPT_DISABLE(1) << (i * 4);
+
+ /* Enable down-conversion for 32bpp and smaller formats. */
+ switch (format) {
+ case V_028C70_COLOR_8:
+ case V_028C70_COLOR_8_8:
+ case V_028C70_COLOR_8_8_8_8:
+ /* For 1 and 2-channel formats, use the superset thereof. */
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_UINT16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_SINT16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_8_8_8_8 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_8BIT_FORMAT << (i * 4);
+ }
+ break;
- radeon_set_context_reg(cmd_buffer->cs, R_0286D8_SPI_PS_IN_CONTROL,
- S_0286D8_NUM_INTERP(ps->info.fs.num_interp));
+ case V_028C70_COLOR_5_6_5:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_5_6_5 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_6BIT_FORMAT << (i * 4);
+ }
+ break;
- radeon_set_context_reg(cmd_buffer->cs, R_0286E0_SPI_BARYC_CNTL, spi_baryc_cntl);
+ case V_028C70_COLOR_1_5_5_5:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_1_5_5_5 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_5BIT_FORMAT << (i * 4);
+ }
+ break;
- radeon_set_context_reg(cmd_buffer->cs, R_028710_SPI_SHADER_Z_FORMAT,
- pipeline->graphics.shader_z_format);
+ case V_028C70_COLOR_4_4_4_4:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_4_4_4_4 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_4BIT_FORMAT << (i * 4);
+ }
+ break;
- radeon_set_context_reg(cmd_buffer->cs, R_028714_SPI_SHADER_COL_FORMAT, blend->spi_shader_col_format);
+ case V_028C70_COLOR_32:
+ if (swap == V_028C70_SWAP_STD &&
+ spi_format == V_028714_SPI_SHADER_32_R)
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_32_R << (i * 4);
+ else if (swap == V_028C70_SWAP_ALT_REV &&
+ spi_format == V_028714_SPI_SHADER_32_AR)
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_32_A << (i * 4);
+ break;
- radeon_set_context_reg(cmd_buffer->cs, R_028238_CB_TARGET_MASK, blend->cb_target_mask);
- radeon_set_context_reg(cmd_buffer->cs, R_02823C_CB_SHADER_MASK, blend->cb_shader_mask);
+ case V_028C70_COLOR_16:
+ case V_028C70_COLOR_16_16:
+ /* For 1-channel formats, use the superset thereof. */
+ if (spi_format == V_028714_SPI_SHADER_UNORM16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_SNORM16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_UINT16_ABGR ||
+ spi_format == V_028714_SPI_SHADER_SINT16_ABGR) {
+ if (swap == V_028C70_SWAP_STD ||
+ swap == V_028C70_SWAP_STD_REV)
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_16_16_GR << (i * 4);
+ else
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_16_16_AR << (i * 4);
+ }
+ break;
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
- /* optimise this? */
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
- radeon_emit(cmd_buffer->cs, EVENT_TYPE(V_028A90_FLUSH_DFSM) | EVENT_INDEX(0));
- }
+ case V_028C70_COLOR_10_11_11:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_10_11_11 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_11BIT_FORMAT << (i * 4);
+ }
+ break;
- if (pipeline->graphics.ps_input_cntl_num) {
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028644_SPI_PS_INPUT_CNTL_0, pipeline->graphics.ps_input_cntl_num);
- for (unsigned i = 0; i < pipeline->graphics.ps_input_cntl_num; i++) {
- radeon_emit(cmd_buffer->cs, pipeline->graphics.ps_input_cntl[i]);
+ case V_028C70_COLOR_2_10_10_10:
+ if (spi_format == V_028714_SPI_SHADER_FP16_ABGR) {
+ sx_ps_downconvert |= V_028754_SX_RT_EXPORT_2_10_10_10 << (i * 4);
+ sx_blend_opt_epsilon |= V_028758_10BIT_FORMAT << (i * 4);
+ }
+ break;
}
}
-}
-
-static void polaris_set_vgt_vertex_reuse(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
-{
- uint32_t vtx_reuse_depth = 30;
- if (cmd_buffer->device->physical_device->rad_info.family < CHIP_POLARIS10)
- return;
- if (pipeline->shaders[MESA_SHADER_TESS_EVAL]) {
- if (pipeline->shaders[MESA_SHADER_TESS_EVAL]->info.tes.spacing == TESS_SPACING_FRACTIONAL_ODD)
- vtx_reuse_depth = 14;
- }
- radeon_set_context_reg(cmd_buffer->cs, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL,
- vtx_reuse_depth);
+ radeon_set_context_reg_seq(cmd_buffer->cs, R_028754_SX_PS_DOWNCONVERT, 3);
+ radeon_emit(cmd_buffer->cs, sx_ps_downconvert);
+ radeon_emit(cmd_buffer->cs, sx_blend_opt_epsilon);
+ radeon_emit(cmd_buffer->cs, sx_blend_opt_control);
}
static void
if (!pipeline || cmd_buffer->state.emitted_pipeline == pipeline)
return;
- radv_emit_graphics_depth_stencil_state(cmd_buffer, pipeline);
- radv_emit_graphics_blend_state(cmd_buffer, pipeline);
- radv_emit_graphics_raster_state(cmd_buffer, pipeline);
radv_update_multisample_state(cmd_buffer, pipeline);
- radv_emit_vertex_shader(cmd_buffer, pipeline);
- radv_emit_tess_shaders(cmd_buffer, pipeline);
- radv_emit_geometry_shader(cmd_buffer, pipeline);
- radv_emit_fragment_shader(cmd_buffer, pipeline);
- polaris_set_vgt_vertex_reuse(cmd_buffer, pipeline);
cmd_buffer->scratch_size_needed =
MAX2(cmd_buffer->scratch_size_needed,
pipeline->max_waves * pipeline->scratch_bytes_per_wave);
- radeon_set_context_reg(cmd_buffer->cs, R_0286E8_SPI_TMPRING_SIZE,
- S_0286E8_WAVES(pipeline->max_waves) |
- S_0286E8_WAVESIZE(pipeline->scratch_bytes_per_wave >> 10));
-
if (!cmd_buffer->state.emitted_pipeline ||
cmd_buffer->state.emitted_pipeline->graphics.can_use_guardband !=
pipeline->graphics.can_use_guardband)
cmd_buffer->state.dirty |= RADV_CMD_DIRTY_DYNAMIC_SCISSOR;
- radeon_set_context_reg(cmd_buffer->cs, R_028B54_VGT_SHADER_STAGES_EN, pipeline->graphics.vgt_shader_stages_en);
+ radeon_emit_array(cmd_buffer->cs, pipeline->cs.buf, pipeline->cs.cdw);
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK) {
- radeon_set_uconfig_reg_idx(cmd_buffer->cs, R_030908_VGT_PRIMITIVE_TYPE, 1, pipeline->graphics.prim);
- } else {
- radeon_set_config_reg(cmd_buffer->cs, R_008958_VGT_PRIMITIVE_TYPE, pipeline->graphics.prim);
+ for (unsigned i = 0; i < MESA_SHADER_COMPUTE; i++) {
+ if (!pipeline->shaders[i])
+ continue;
+
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs,
+ pipeline->shaders[i]->bo, 8);
}
- radeon_set_context_reg(cmd_buffer->cs, R_028A6C_VGT_GS_OUT_PRIM_TYPE, pipeline->graphics.gs_out);
- radv_save_pipeline(cmd_buffer, pipeline, RING_GFX);
+ if (radv_pipeline_has_gs(pipeline))
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs,
+ pipeline->gs_copy_shader->bo, 8);
+
+ if (unlikely(cmd_buffer->device->trace_bo))
+ radv_save_pipeline(cmd_buffer, pipeline, RING_GFX);
cmd_buffer->state.emitted_pipeline = pipeline;
+
+ cmd_buffer->state.dirty &= ~RADV_CMD_DIRTY_PIPELINE;
}
static void
radv_emit_scissor(struct radv_cmd_buffer *cmd_buffer)
{
uint32_t count = cmd_buffer->state.dynamic.scissor.count;
+
+ /* Vega10/Raven scissor bug workaround. This must be done before VPORT
+ * scissor registers are changed. There is also a more efficient but
+ * more involved alternative workaround.
+ */
+ if (cmd_buffer->device->physical_device->has_scissor_bug) {
+ cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_PS_PARTIAL_FLUSH;
+ si_emit_cache_flush(cmd_buffer);
+ }
si_write_scissors(cmd_buffer->cs, 0, count,
cmd_buffer->state.dynamic.scissor.scissors,
cmd_buffer->state.dynamic.viewport.viewports,
cmd_buffer->state.emitted_pipeline->graphics.can_use_guardband);
- radeon_set_context_reg(cmd_buffer->cs, R_028A48_PA_SC_MODE_CNTL_0,
- cmd_buffer->state.pipeline->graphics.ms.pa_sc_mode_cntl_0 | S_028A48_VPORT_SCISSOR_ENABLE(count ? 1 : 0));
+}
+
+static void
+radv_emit_discard_rectangle(struct radv_cmd_buffer *cmd_buffer)
+{
+ if (!cmd_buffer->state.dynamic.discard_rectangle.count)
+ return;
+
+ radeon_set_context_reg_seq(cmd_buffer->cs, R_028210_PA_SC_CLIPRECT_0_TL,
+ cmd_buffer->state.dynamic.discard_rectangle.count * 2);
+ for (unsigned i = 0; i < cmd_buffer->state.dynamic.discard_rectangle.count; ++i) {
+ VkRect2D rect = cmd_buffer->state.dynamic.discard_rectangle.rectangles[i];
+ radeon_emit(cmd_buffer->cs, S_028210_TL_X(rect.offset.x) | S_028210_TL_Y(rect.offset.y));
+ radeon_emit(cmd_buffer->cs, S_028214_BR_X(rect.offset.x + rect.extent.width) |
+ S_028214_BR_Y(rect.offset.y + rect.extent.height));
+ }
}
static void
}
static void
-radv_emit_depth_biais(struct radv_cmd_buffer *cmd_buffer)
+radv_emit_depth_bias(struct radv_cmd_buffer *cmd_buffer)
{
- struct radv_raster_state *raster = &cmd_buffer->state.pipeline->graphics.raster;
struct radv_dynamic_state *d = &cmd_buffer->state.dynamic;
unsigned slope = fui(d->depth_bias.slope * 16.0f);
unsigned bias = fui(d->depth_bias.bias * cmd_buffer->state.offset_scale);
- if (G_028814_POLY_OFFSET_FRONT_ENABLE(raster->pa_su_sc_mode_cntl)) {
- radeon_set_context_reg_seq(cmd_buffer->cs,
- R_028B7C_PA_SU_POLY_OFFSET_CLAMP, 5);
- radeon_emit(cmd_buffer->cs, fui(d->depth_bias.clamp)); /* CLAMP */
- radeon_emit(cmd_buffer->cs, slope); /* FRONT SCALE */
- radeon_emit(cmd_buffer->cs, bias); /* FRONT OFFSET */
- radeon_emit(cmd_buffer->cs, slope); /* BACK SCALE */
- radeon_emit(cmd_buffer->cs, bias); /* BACK OFFSET */
- }
+
+ radeon_set_context_reg_seq(cmd_buffer->cs,
+ R_028B7C_PA_SU_POLY_OFFSET_CLAMP, 5);
+ radeon_emit(cmd_buffer->cs, fui(d->depth_bias.clamp)); /* CLAMP */
+ radeon_emit(cmd_buffer->cs, slope); /* FRONT SCALE */
+ radeon_emit(cmd_buffer->cs, bias); /* FRONT OFFSET */
+ radeon_emit(cmd_buffer->cs, slope); /* BACK SCALE */
+ radeon_emit(cmd_buffer->cs, bias); /* BACK OFFSET */
}
static void
radv_emit_fb_color_state(struct radv_cmd_buffer *cmd_buffer,
int index,
- struct radv_color_buffer_info *cb)
+ struct radv_attachment_info *att,
+ struct radv_image *image,
+ VkImageLayout layout)
{
bool is_vi = cmd_buffer->device->physical_device->rad_info.chip_class >= VI;
+ struct radv_color_buffer_info *cb = &att->cb;
+ uint32_t cb_color_info = cb->cb_color_info;
+
+ if (!radv_layout_dcc_compressed(image, layout,
+ radv_image_queue_family_mask(image,
+ cmd_buffer->queue_family_index,
+ cmd_buffer->queue_family_index))) {
+ cb_color_info &= C_028C70_DCC_ENABLE;
+ }
if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
radeon_set_context_reg_seq(cmd_buffer->cs, R_028C60_CB_COLOR0_BASE + index * 0x3c, 11);
radeon_emit(cmd_buffer->cs, cb->cb_color_base);
- radeon_emit(cmd_buffer->cs, cb->cb_color_base >> 32);
+ radeon_emit(cmd_buffer->cs, S_028C64_BASE_256B(cb->cb_color_base >> 32));
radeon_emit(cmd_buffer->cs, cb->cb_color_attrib2);
radeon_emit(cmd_buffer->cs, cb->cb_color_view);
- radeon_emit(cmd_buffer->cs, cb->cb_color_info);
+ radeon_emit(cmd_buffer->cs, cb_color_info);
radeon_emit(cmd_buffer->cs, cb->cb_color_attrib);
radeon_emit(cmd_buffer->cs, cb->cb_dcc_control);
radeon_emit(cmd_buffer->cs, cb->cb_color_cmask);
- radeon_emit(cmd_buffer->cs, cb->cb_color_cmask >> 32);
+ radeon_emit(cmd_buffer->cs, S_028C80_BASE_256B(cb->cb_color_cmask >> 32));
radeon_emit(cmd_buffer->cs, cb->cb_color_fmask);
- radeon_emit(cmd_buffer->cs, cb->cb_color_fmask >> 32);
+ radeon_emit(cmd_buffer->cs, S_028C88_BASE_256B(cb->cb_color_fmask >> 32));
radeon_set_context_reg_seq(cmd_buffer->cs, R_028C94_CB_COLOR0_DCC_BASE + index * 0x3c, 2);
radeon_emit(cmd_buffer->cs, cb->cb_dcc_base);
- radeon_emit(cmd_buffer->cs, cb->cb_dcc_base >> 32);
+ radeon_emit(cmd_buffer->cs, S_028C98_BASE_256B(cb->cb_dcc_base >> 32));
radeon_set_context_reg(cmd_buffer->cs, R_0287A0_CB_MRT0_EPITCH + index * 4,
- cb->gfx9_epitch);
+ S_0287A0_EPITCH(att->attachment->image->surface.u.gfx9.surf.epitch));
} else {
radeon_set_context_reg_seq(cmd_buffer->cs, R_028C60_CB_COLOR0_BASE + index * 0x3c, 11);
radeon_emit(cmd_buffer->cs, cb->cb_color_base);
radeon_emit(cmd_buffer->cs, cb->cb_color_pitch);
radeon_emit(cmd_buffer->cs, cb->cb_color_slice);
radeon_emit(cmd_buffer->cs, cb->cb_color_view);
- radeon_emit(cmd_buffer->cs, cb->cb_color_info);
+ radeon_emit(cmd_buffer->cs, cb_color_info);
radeon_emit(cmd_buffer->cs, cb->cb_color_attrib);
radeon_emit(cmd_buffer->cs, cb->cb_dcc_control);
radeon_emit(cmd_buffer->cs, cb->cb_color_cmask);
if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
radeon_set_context_reg_seq(cmd_buffer->cs, R_028014_DB_HTILE_DATA_BASE, 3);
radeon_emit(cmd_buffer->cs, ds->db_htile_data_base);
- radeon_emit(cmd_buffer->cs, ds->db_htile_data_base >> 32);
+ radeon_emit(cmd_buffer->cs, S_028018_BASE_HI(ds->db_htile_data_base >> 32));
radeon_emit(cmd_buffer->cs, ds->db_depth_size);
radeon_set_context_reg_seq(cmd_buffer->cs, R_028038_DB_Z_INFO, 10);
radeon_emit(cmd_buffer->cs, db_z_info); /* DB_Z_INFO */
radeon_emit(cmd_buffer->cs, db_stencil_info); /* DB_STENCIL_INFO */
radeon_emit(cmd_buffer->cs, ds->db_z_read_base); /* DB_Z_READ_BASE */
- radeon_emit(cmd_buffer->cs, ds->db_z_read_base >> 32); /* DB_Z_READ_BASE_HI */
+ radeon_emit(cmd_buffer->cs, S_028044_BASE_HI(ds->db_z_read_base >> 32)); /* DB_Z_READ_BASE_HI */
radeon_emit(cmd_buffer->cs, ds->db_stencil_read_base); /* DB_STENCIL_READ_BASE */
- radeon_emit(cmd_buffer->cs, ds->db_stencil_read_base >> 32); /* DB_STENCIL_READ_BASE_HI */
+ radeon_emit(cmd_buffer->cs, S_02804C_BASE_HI(ds->db_stencil_read_base >> 32)); /* DB_STENCIL_READ_BASE_HI */
radeon_emit(cmd_buffer->cs, ds->db_z_write_base); /* DB_Z_WRITE_BASE */
- radeon_emit(cmd_buffer->cs, ds->db_z_write_base >> 32); /* DB_Z_WRITE_BASE_HI */
+ radeon_emit(cmd_buffer->cs, S_028054_BASE_HI(ds->db_z_write_base >> 32)); /* DB_Z_WRITE_BASE_HI */
radeon_emit(cmd_buffer->cs, ds->db_stencil_write_base); /* DB_STENCIL_WRITE_BASE */
- radeon_emit(cmd_buffer->cs, ds->db_stencil_write_base >> 32); /* DB_STENCIL_WRITE_BASE_HI */
+ radeon_emit(cmd_buffer->cs, S_02805C_BASE_HI(ds->db_stencil_write_base >> 32)); /* DB_STENCIL_WRITE_BASE_HI */
radeon_set_context_reg_seq(cmd_buffer->cs, R_028068_DB_Z_INFO2, 2);
radeon_emit(cmd_buffer->cs, ds->db_z_info2);
VkClearDepthStencilValue ds_clear_value,
VkImageAspectFlags aspects)
{
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(image->bo);
+ uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->clear_value_offset;
unsigned reg_offset = 0, reg_count = 0;
- if (!image->surface.htile_size || !aspects)
- return;
+ assert(radv_image_has_htile(image));
if (aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
++reg_count;
if (aspects & VK_IMAGE_ASPECT_DEPTH_BIT)
++reg_count;
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, image->bo, 8);
-
radeon_emit(cmd_buffer->cs, PKT3(PKT3_WRITE_DATA, 2 + reg_count, 0));
radeon_emit(cmd_buffer->cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
S_370_WR_CONFIRM(1) |
radv_load_depth_clear_regs(struct radv_cmd_buffer *cmd_buffer,
struct radv_image *image)
{
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(image->bo);
+ VkImageAspectFlags aspects = vk_format_aspects(image->vk_format);
+ uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->clear_value_offset;
+ unsigned reg_offset = 0, reg_count = 0;
- if (!image->surface.htile_size)
+ if (!radv_image_has_htile(image))
return;
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, image->bo, 8);
+ if (aspects & VK_IMAGE_ASPECT_STENCIL_BIT) {
+ ++reg_count;
+ } else {
+ ++reg_offset;
+ va += 4;
+ }
+ if (aspects & VK_IMAGE_ASPECT_DEPTH_BIT)
+ ++reg_count;
radeon_emit(cmd_buffer->cs, PKT3(PKT3_COPY_DATA, 4, 0));
radeon_emit(cmd_buffer->cs, COPY_DATA_SRC_SEL(COPY_DATA_MEM) |
COPY_DATA_DST_SEL(COPY_DATA_REG) |
- COPY_DATA_COUNT_SEL);
+ (reg_count == 2 ? COPY_DATA_COUNT_SEL : 0));
radeon_emit(cmd_buffer->cs, va);
radeon_emit(cmd_buffer->cs, va >> 32);
- radeon_emit(cmd_buffer->cs, R_028028_DB_STENCIL_CLEAR >> 2);
+ radeon_emit(cmd_buffer->cs, (R_028028_DB_STENCIL_CLEAR + 4 * reg_offset) >> 2);
radeon_emit(cmd_buffer->cs, 0);
radeon_emit(cmd_buffer->cs, PKT3(PKT3_PFP_SYNC_ME, 0, 0));
}
/*
- *with DCC some colors don't require CMASK elimiation before being
+ * With DCC some colors don't require CMASK elimination before being
* used as a texture. This sets a predicate value to determine if the
* cmask eliminate is required.
*/
bool value)
{
uint64_t pred_val = value;
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(image->bo);
+ uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->dcc_pred_offset;
- if (!image->surface.dcc_size)
- return;
-
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, image->bo, 8);
+ assert(radv_image_has_dcc(image));
radeon_emit(cmd_buffer->cs, PKT3(PKT3_WRITE_DATA, 4, 0));
radeon_emit(cmd_buffer->cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
int idx,
uint32_t color_values[2])
{
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(image->bo);
+ uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->clear_value_offset;
- if (!image->cmask.size && !image->surface.dcc_size)
- return;
-
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, image->bo, 8);
+ assert(radv_image_has_cmask(image) || radv_image_has_dcc(image));
radeon_emit(cmd_buffer->cs, PKT3(PKT3_WRITE_DATA, 4, 0));
radeon_emit(cmd_buffer->cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
struct radv_image *image,
int idx)
{
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(image->bo);
+ uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->clear_value_offset;
- if (!image->cmask.size && !image->surface.dcc_size)
+ if (!radv_image_has_cmask(image) && !radv_image_has_dcc(image))
return;
uint32_t reg = R_028C8C_CB_COLOR0_CLEAR_WORD0 + idx * 0x3c;
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, image->bo, 8);
radeon_emit(cmd_buffer->cs, PKT3(PKT3_COPY_DATA, 4, cmd_buffer->state.predicating));
radeon_emit(cmd_buffer->cs, COPY_DATA_SRC_SEL(COPY_DATA_MEM) |
radeon_emit(cmd_buffer->cs, 0);
}
-void
+static void
radv_emit_framebuffer_state(struct radv_cmd_buffer *cmd_buffer)
{
int i;
int idx = subpass->color_attachments[i].attachment;
struct radv_attachment_info *att = &framebuffer->attachments[idx];
+ struct radv_image *image = att->attachment->image;
+ VkImageLayout layout = subpass->color_attachments[i].layout;
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, att->attachment->bo, 8);
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs, att->attachment->bo, 8);
assert(att->attachment->aspect_mask & VK_IMAGE_ASPECT_COLOR_BIT);
- radv_emit_fb_color_state(cmd_buffer, i, &att->cb);
+ radv_emit_fb_color_state(cmd_buffer, i, att, image, layout);
- radv_load_color_clear_regs(cmd_buffer, att->attachment->image, i);
+ radv_load_color_clear_regs(cmd_buffer, image, i);
}
if(subpass->depth_stencil_attachment.attachment != VK_ATTACHMENT_UNUSED) {
VkImageLayout layout = subpass->depth_stencil_attachment.layout;
struct radv_attachment_info *att = &framebuffer->attachments[idx];
struct radv_image *image = att->attachment->image;
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, att->attachment->bo, 8);
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs, att->attachment->bo, 8);
MAYBE_UNUSED uint32_t queue_mask = radv_image_queue_family_mask(image,
cmd_buffer->queue_family_index,
cmd_buffer->queue_family_index);
S_028208_BR_X(framebuffer->width) |
S_028208_BR_Y(framebuffer->height));
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
+ if (cmd_buffer->device->dfsm_allowed) {
radeon_emit(cmd_buffer->cs, PKT3(PKT3_EVENT_WRITE, 0, 0));
radeon_emit(cmd_buffer->cs, EVENT_TYPE(V_028A90_BREAK_BATCH) | EVENT_INDEX(0));
}
+
+ cmd_buffer->state.dirty &= ~RADV_CMD_DIRTY_FRAMEBUFFER;
+}
+
+static void
+radv_emit_index_buffer(struct radv_cmd_buffer *cmd_buffer)
+{
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
+ struct radv_cmd_state *state = &cmd_buffer->state;
+
+ if (state->index_type != state->last_index_type) {
+ if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
+ radeon_set_uconfig_reg_idx(cs, R_03090C_VGT_INDEX_TYPE,
+ 2, state->index_type);
+ } else {
+ radeon_emit(cs, PKT3(PKT3_INDEX_TYPE, 0, 0));
+ radeon_emit(cs, state->index_type);
+ }
+
+ state->last_index_type = state->index_type;
+ }
+
+ radeon_emit(cs, PKT3(PKT3_INDEX_BASE, 1, 0));
+ radeon_emit(cs, state->index_va);
+ radeon_emit(cs, state->index_va >> 32);
+
+ radeon_emit(cs, PKT3(PKT3_INDEX_BUFFER_SIZE, 0, 0));
+ radeon_emit(cs, state->max_index_count);
+
+ cmd_buffer->state.dirty &= ~RADV_CMD_DIRTY_INDEX_BUFFER;
}
void radv_set_db_count_control(struct radv_cmd_buffer *cmd_buffer)
{
+ bool has_perfect_queries = cmd_buffer->state.perfect_occlusion_queries_enabled;
+ struct radv_pipeline *pipeline = cmd_buffer->state.pipeline;
+ uint32_t pa_sc_mode_cntl_1 =
+ pipeline ? pipeline->graphics.ms.pa_sc_mode_cntl_1 : 0;
uint32_t db_count_control;
if(!cmd_buffer->state.active_occlusion_queries) {
if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK) {
+ if (G_028A4C_OUT_OF_ORDER_PRIMITIVE_ENABLE(pa_sc_mode_cntl_1) &&
+ pipeline->graphics.disable_out_of_order_rast_for_occlusion &&
+ has_perfect_queries) {
+ /* Re-enable out-of-order rasterization if the
+ * bound pipeline supports it and if it's has
+ * been disabled before starting any perfect
+ * occlusion queries.
+ */
+ radeon_set_context_reg(cmd_buffer->cs,
+ R_028A4C_PA_SC_MODE_CNTL_1,
+ pa_sc_mode_cntl_1);
+ }
db_count_control = 0;
} else {
db_count_control = S_028004_ZPASS_INCREMENT_DISABLE(1);
}
} else {
+ const struct radv_subpass *subpass = cmd_buffer->state.subpass;
+ uint32_t sample_rate = subpass ? util_logbase2(subpass->max_sample_count) : 0;
+
if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK) {
- db_count_control = S_028004_PERFECT_ZPASS_COUNTS(1) |
- S_028004_SAMPLE_RATE(0) | /* TODO: set this to the number of samples of the current framebuffer */
+ db_count_control =
+ S_028004_PERFECT_ZPASS_COUNTS(has_perfect_queries) |
+ S_028004_SAMPLE_RATE(sample_rate) |
S_028004_ZPASS_ENABLE(1) |
S_028004_SLICE_EVEN_ENABLE(1) |
S_028004_SLICE_ODD_ENABLE(1);
+
+ if (G_028A4C_OUT_OF_ORDER_PRIMITIVE_ENABLE(pa_sc_mode_cntl_1) &&
+ pipeline->graphics.disable_out_of_order_rast_for_occlusion &&
+ has_perfect_queries) {
+ /* If the bound pipeline has enabled
+ * out-of-order rasterization, we should
+ * disable it before starting any perfect
+ * occlusion queries.
+ */
+ pa_sc_mode_cntl_1 &= C_028A4C_OUT_OF_ORDER_PRIMITIVE_ENABLE;
+
+ radeon_set_context_reg(cmd_buffer->cs,
+ R_028A4C_PA_SC_MODE_CNTL_1,
+ pa_sc_mode_cntl_1);
+ }
} else {
db_count_control = S_028004_PERFECT_ZPASS_COUNTS(1) |
- S_028004_SAMPLE_RATE(0); /* TODO: set this to the number of samples of the current framebuffer */
+ S_028004_SAMPLE_RATE(sample_rate);
}
}
static void
radv_cmd_buffer_flush_dynamic_state(struct radv_cmd_buffer *cmd_buffer)
{
- if (G_028810_DX_RASTERIZATION_KILL(cmd_buffer->state.pipeline->graphics.raster.pa_cl_clip_cntl))
- return;
+ uint32_t states = cmd_buffer->state.dirty & cmd_buffer->state.emitted_pipeline->graphics.needed_dynamic_state;
- if (cmd_buffer->state.dirty & (RADV_CMD_DIRTY_DYNAMIC_VIEWPORT))
+ if (states & (RADV_CMD_DIRTY_DYNAMIC_VIEWPORT))
radv_emit_viewport(cmd_buffer);
- if (cmd_buffer->state.dirty & (RADV_CMD_DIRTY_DYNAMIC_SCISSOR | RADV_CMD_DIRTY_DYNAMIC_VIEWPORT))
+ if (states & (RADV_CMD_DIRTY_DYNAMIC_SCISSOR | RADV_CMD_DIRTY_DYNAMIC_VIEWPORT))
radv_emit_scissor(cmd_buffer);
- if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)
+ if (states & RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)
radv_emit_line_width(cmd_buffer);
- if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS)
+ if (states & RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS)
radv_emit_blend_constants(cmd_buffer);
- if (cmd_buffer->state.dirty & (RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE |
+ if (states & (RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE |
RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK |
RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK))
radv_emit_stencil(cmd_buffer);
- if (cmd_buffer->state.dirty & (RADV_CMD_DIRTY_PIPELINE |
- RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS))
+ if (states & RADV_CMD_DIRTY_DYNAMIC_DEPTH_BOUNDS)
radv_emit_depth_bounds(cmd_buffer);
- if (cmd_buffer->state.dirty & (RADV_CMD_DIRTY_PIPELINE |
- RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS))
- radv_emit_depth_biais(cmd_buffer);
+ if (states & RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS)
+ radv_emit_depth_bias(cmd_buffer);
- cmd_buffer->state.dirty = 0;
+ if (states & RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE)
+ radv_emit_discard_rectangle(cmd_buffer);
+
+ cmd_buffer->state.dirty &= ~states;
}
static void
uint64_t va,
gl_shader_stage stage)
{
- struct ac_userdata_info *desc_set_loc = &pipeline->shaders[stage]->info.user_sgprs_locs.descriptor_sets[idx];
- uint32_t base_reg = radv_shader_stage_to_user_data_0(stage, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
+ struct radv_userdata_info *desc_set_loc = &pipeline->shaders[stage]->info.user_sgprs_locs.descriptor_sets[idx];
+ uint32_t base_reg = pipeline->user_data_0[stage];
if (desc_set_loc->sgpr_idx == -1 || desc_set_loc->indirect)
return;
assert(!desc_set_loc->indirect);
assert(desc_set_loc->num_sgprs == 2);
- radeon_set_sh_reg_seq(cmd_buffer->cs,
- base_reg + desc_set_loc->sgpr_idx * 4, 2);
- radeon_emit(cmd_buffer->cs, va);
- radeon_emit(cmd_buffer->cs, va >> 32);
+
+ radv_emit_shader_pointer(cmd_buffer->cs,
+ base_reg + desc_set_loc->sgpr_idx * 4, va);
}
static void
}
static void
-radv_flush_push_descriptors(struct radv_cmd_buffer *cmd_buffer)
+radv_flush_push_descriptors(struct radv_cmd_buffer *cmd_buffer,
+ VkPipelineBindPoint bind_point)
{
- struct radv_descriptor_set *set = &cmd_buffer->push_descriptors.set;
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, bind_point);
+ struct radv_descriptor_set *set = &descriptors_state->push_set.set;
unsigned bo_offset;
if (!radv_cmd_buffer_upload_data(cmd_buffer, set->size, 32,
&bo_offset))
return;
- set->va = cmd_buffer->device->ws->buffer_get_va(cmd_buffer->upload.upload_bo);
+ set->va = radv_buffer_get_va(cmd_buffer->upload.upload_bo);
set->va += bo_offset;
}
static void
-radv_flush_indirect_descriptor_sets(struct radv_cmd_buffer *cmd_buffer)
+radv_flush_indirect_descriptor_sets(struct radv_cmd_buffer *cmd_buffer,
+ VkPipelineBindPoint bind_point)
{
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, bind_point);
uint32_t size = MAX_SETS * 2 * 4;
uint32_t offset;
void *ptr;
for (unsigned i = 0; i < MAX_SETS; i++) {
uint32_t *uptr = ((uint32_t *)ptr) + i * 2;
uint64_t set_va = 0;
- struct radv_descriptor_set *set = cmd_buffer->state.descriptors[i];
- if (set)
+ struct radv_descriptor_set *set = descriptors_state->sets[i];
+ if (descriptors_state->valid & (1u << i))
set_va = set->va;
uptr[0] = set_va & 0xffffffff;
uptr[1] = set_va >> 32;
}
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(cmd_buffer->upload.upload_bo);
+ uint64_t va = radv_buffer_get_va(cmd_buffer->upload.upload_bo);
va += offset;
if (cmd_buffer->state.pipeline) {
radv_flush_descriptors(struct radv_cmd_buffer *cmd_buffer,
VkShaderStageFlags stages)
{
+ VkPipelineBindPoint bind_point = stages & VK_SHADER_STAGE_COMPUTE_BIT ?
+ VK_PIPELINE_BIND_POINT_COMPUTE :
+ VK_PIPELINE_BIND_POINT_GRAPHICS;
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, bind_point);
unsigned i;
- if (!cmd_buffer->state.descriptors_dirty)
+ if (!descriptors_state->dirty)
return;
- if (cmd_buffer->state.push_descriptors_dirty)
- radv_flush_push_descriptors(cmd_buffer);
+ if (descriptors_state->push_dirty)
+ radv_flush_push_descriptors(cmd_buffer, bind_point);
if ((cmd_buffer->state.pipeline && cmd_buffer->state.pipeline->need_indirect_descriptor_sets) ||
(cmd_buffer->state.compute_pipeline && cmd_buffer->state.compute_pipeline->need_indirect_descriptor_sets)) {
- radv_flush_indirect_descriptor_sets(cmd_buffer);
+ radv_flush_indirect_descriptor_sets(cmd_buffer, bind_point);
}
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
cmd_buffer->cs,
MAX_SETS * MESA_SHADER_STAGES * 4);
- for (i = 0; i < MAX_SETS; i++) {
- if (!(cmd_buffer->state.descriptors_dirty & (1u << i)))
- continue;
- struct radv_descriptor_set *set = cmd_buffer->state.descriptors[i];
- if (!set)
+ for_each_bit(i, descriptors_state->dirty) {
+ struct radv_descriptor_set *set = descriptors_state->sets[i];
+ if (!(descriptors_state->valid & (1u << i)))
continue;
radv_emit_descriptor_set_userdata(cmd_buffer, stages, set, i);
}
- cmd_buffer->state.descriptors_dirty = 0;
- cmd_buffer->state.push_descriptors_dirty = false;
+ descriptors_state->dirty = 0;
+ descriptors_state->push_dirty = false;
+
+ if (unlikely(cmd_buffer->device->trace_bo))
+ radv_save_descriptors(cmd_buffer, bind_point);
+
assert(cmd_buffer->cs->cdw <= cdw_max);
}
static void
radv_flush_constants(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline,
VkShaderStageFlags stages)
{
+ struct radv_pipeline *pipeline = stages & VK_SHADER_STAGE_COMPUTE_BIT
+ ? cmd_buffer->state.compute_pipeline
+ : cmd_buffer->state.pipeline;
struct radv_pipeline_layout *layout = pipeline->layout;
unsigned offset;
void *ptr;
uint64_t va;
stages &= cmd_buffer->push_constant_stages;
- if (!stages || !layout || (!layout->push_constant_size && !layout->dynamic_offset_count))
+ if (!stages ||
+ (!layout->push_constant_size && !layout->dynamic_offset_count))
return;
if (!radv_cmd_buffer_upload_alloc(cmd_buffer, layout->push_constant_size +
memcpy((char*)ptr + layout->push_constant_size, cmd_buffer->dynamic_buffers,
16 * layout->dynamic_offset_count);
- va = cmd_buffer->device->ws->buffer_get_va(cmd_buffer->upload.upload_bo);
+ va = radv_buffer_get_va(cmd_buffer->upload.upload_bo);
va += offset;
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
assert(cmd_buffer->cs->cdw <= cdw_max);
}
-static void radv_emit_primitive_reset_state(struct radv_cmd_buffer *cmd_buffer,
- bool indexed_draw)
-{
- int32_t primitive_reset_en = indexed_draw && cmd_buffer->state.pipeline->graphics.prim_restart_enable;
-
- if (primitive_reset_en != cmd_buffer->state.last_primitive_reset_en) {
- cmd_buffer->state.last_primitive_reset_en = primitive_reset_en;
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
- radeon_set_uconfig_reg(cmd_buffer->cs, R_03092C_VGT_MULTI_PRIM_IB_RESET_EN,
- primitive_reset_en);
- } else {
- radeon_set_context_reg(cmd_buffer->cs, R_028A94_VGT_MULTI_PRIM_IB_RESET_EN,
- primitive_reset_en);
- }
- }
-
- if (primitive_reset_en) {
- uint32_t primitive_reset_index = cmd_buffer->state.index_type ? 0xffffffffu : 0xffffu;
-
- if (primitive_reset_index != cmd_buffer->state.last_primitive_reset_index) {
- cmd_buffer->state.last_primitive_reset_index = primitive_reset_index;
- radeon_set_context_reg(cmd_buffer->cs, R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX,
- primitive_reset_index);
- }
- }
-}
-
-static bool
-radv_cmd_buffer_update_vertex_descriptors(struct radv_cmd_buffer *cmd_buffer)
+static void
+radv_flush_vertex_descriptors(struct radv_cmd_buffer *cmd_buffer,
+ bool pipeline_is_dirty)
{
- struct radv_device *device = cmd_buffer->device;
-
- if ((cmd_buffer->state.pipeline != cmd_buffer->state.emitted_pipeline || cmd_buffer->state.vb_dirty) &&
+ if ((pipeline_is_dirty ||
+ (cmd_buffer->state.dirty & RADV_CMD_DIRTY_VERTEX_BUFFER)) &&
cmd_buffer->state.pipeline->vertex_elements.count &&
- cmd_buffer->state.pipeline->shaders[MESA_SHADER_VERTEX]->info.info.vs.has_vertex_buffers) {
+ radv_get_vertex_shader(cmd_buffer->state.pipeline)->info.info.vs.has_vertex_buffers) {
struct radv_vertex_elements_info *velems = &cmd_buffer->state.pipeline->vertex_elements;
unsigned vb_offset;
void *vb_ptr;
/* allocate some descriptor state for vertex buffers */
if (!radv_cmd_buffer_upload_alloc(cmd_buffer, count * 16, 256,
&vb_offset, &vb_ptr))
- return false;
+ return;
for (i = 0; i < count; i++) {
uint32_t *desc = &((uint32_t *)vb_ptr)[i * 4];
uint32_t offset;
int vb = velems->binding[i];
- struct radv_buffer *buffer = cmd_buffer->state.vertex_bindings[vb].buffer;
+ struct radv_buffer *buffer = cmd_buffer->vertex_bindings[vb].buffer;
uint32_t stride = cmd_buffer->state.pipeline->binding_stride[vb];
- device->ws->cs_add_buffer(cmd_buffer->cs, buffer->bo, 8);
- va = device->ws->buffer_get_va(buffer->bo);
+ va = radv_buffer_get_va(buffer->bo);
- offset = cmd_buffer->state.vertex_bindings[vb].offset + velems->offset[i];
+ offset = cmd_buffer->vertex_bindings[vb].offset + velems->offset[i];
va += offset + buffer->offset;
desc[0] = va;
desc[1] = S_008F04_BASE_ADDRESS_HI(va >> 32) | S_008F04_STRIDE(stride);
desc[3] = velems->rsrc_word3[i];
}
- va = device->ws->buffer_get_va(cmd_buffer->upload.upload_bo);
+ va = radv_buffer_get_va(cmd_buffer->upload.upload_bo);
va += vb_offset;
radv_emit_userdata_address(cmd_buffer, cmd_buffer->state.pipeline, MESA_SHADER_VERTEX,
AC_UD_VS_VERTEX_BUFFERS, va);
+
+ cmd_buffer->state.vb_va = va;
+ cmd_buffer->state.vb_size = count * 16;
+ cmd_buffer->state.prefetch_L2_mask |= RADV_PREFETCH_VBO_DESCRIPTORS;
}
- cmd_buffer->state.vb_dirty = false;
+ cmd_buffer->state.dirty &= ~RADV_CMD_DIRTY_VERTEX_BUFFER;
+}
- return true;
+static void
+radv_upload_graphics_shader_descriptors(struct radv_cmd_buffer *cmd_buffer, bool pipeline_is_dirty)
+{
+ radv_flush_vertex_descriptors(cmd_buffer, pipeline_is_dirty);
+ radv_flush_descriptors(cmd_buffer, VK_SHADER_STAGE_ALL_GRAPHICS);
+ radv_flush_constants(cmd_buffer, VK_SHADER_STAGE_ALL_GRAPHICS);
}
static void
-radv_cmd_buffer_flush_state(struct radv_cmd_buffer *cmd_buffer,
- bool indexed_draw, bool instanced_draw,
- bool indirect_draw,
- uint32_t draw_vertex_count)
+radv_emit_draw_registers(struct radv_cmd_buffer *cmd_buffer, bool indexed_draw,
+ bool instanced_draw, bool indirect_draw,
+ uint32_t draw_vertex_count)
{
+ struct radeon_info *info = &cmd_buffer->device->physical_device->rad_info;
+ struct radv_cmd_state *state = &cmd_buffer->state;
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
uint32_t ia_multi_vgt_param;
-
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
- cmd_buffer->cs, 4096);
-
- if (!radv_cmd_buffer_update_vertex_descriptors(cmd_buffer))
- return;
-
- if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_PIPELINE)
- radv_emit_graphics_pipeline(cmd_buffer);
-
- if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_RENDER_TARGETS)
- radv_emit_framebuffer_state(cmd_buffer);
-
- ia_multi_vgt_param = si_get_ia_multi_vgt_param(cmd_buffer, instanced_draw, indirect_draw, draw_vertex_count);
- if (cmd_buffer->state.last_ia_multi_vgt_param != ia_multi_vgt_param) {
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9)
- radeon_set_uconfig_reg_idx(cmd_buffer->cs, R_030960_IA_MULTI_VGT_PARAM, 4, ia_multi_vgt_param);
- else if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK)
- radeon_set_context_reg_idx(cmd_buffer->cs, R_028AA8_IA_MULTI_VGT_PARAM, 1, ia_multi_vgt_param);
- else
- radeon_set_context_reg(cmd_buffer->cs, R_028AA8_IA_MULTI_VGT_PARAM, ia_multi_vgt_param);
- cmd_buffer->state.last_ia_multi_vgt_param = ia_multi_vgt_param;
+ int32_t primitive_reset_en;
+
+ /* Draw state. */
+ ia_multi_vgt_param =
+ si_get_ia_multi_vgt_param(cmd_buffer, instanced_draw,
+ indirect_draw, draw_vertex_count);
+
+ if (state->last_ia_multi_vgt_param != ia_multi_vgt_param) {
+ if (info->chip_class >= GFX9) {
+ radeon_set_uconfig_reg_idx(cs,
+ R_030960_IA_MULTI_VGT_PARAM,
+ 4, ia_multi_vgt_param);
+ } else if (info->chip_class >= CIK) {
+ radeon_set_context_reg_idx(cs,
+ R_028AA8_IA_MULTI_VGT_PARAM,
+ 1, ia_multi_vgt_param);
+ } else {
+ radeon_set_context_reg(cs, R_028AA8_IA_MULTI_VGT_PARAM,
+ ia_multi_vgt_param);
+ }
+ state->last_ia_multi_vgt_param = ia_multi_vgt_param;
}
- radv_cmd_buffer_flush_dynamic_state(cmd_buffer);
-
- radv_emit_primitive_reset_state(cmd_buffer, indexed_draw);
+ /* Primitive restart. */
+ primitive_reset_en =
+ indexed_draw && state->pipeline->graphics.prim_restart_enable;
- radv_flush_descriptors(cmd_buffer, VK_SHADER_STAGE_ALL_GRAPHICS);
- radv_flush_constants(cmd_buffer, cmd_buffer->state.pipeline,
- VK_SHADER_STAGE_ALL_GRAPHICS);
+ if (primitive_reset_en != state->last_primitive_reset_en) {
+ state->last_primitive_reset_en = primitive_reset_en;
+ if (info->chip_class >= GFX9) {
+ radeon_set_uconfig_reg(cs,
+ R_03092C_VGT_MULTI_PRIM_IB_RESET_EN,
+ primitive_reset_en);
+ } else {
+ radeon_set_context_reg(cs,
+ R_028A94_VGT_MULTI_PRIM_IB_RESET_EN,
+ primitive_reset_en);
+ }
+ }
- assert(cmd_buffer->cs->cdw <= cdw_max);
+ if (primitive_reset_en) {
+ uint32_t primitive_reset_index =
+ state->index_type ? 0xffffffffu : 0xffffu;
- si_emit_cache_flush(cmd_buffer);
+ if (primitive_reset_index != state->last_primitive_reset_index) {
+ radeon_set_context_reg(cs,
+ R_02840C_VGT_MULTI_PRIM_IB_RESET_INDX,
+ primitive_reset_index);
+ state->last_primitive_reset_index = primitive_reset_index;
+ }
+ }
}
static void radv_stage_flush(struct radv_cmd_buffer *cmd_buffer,
VK_PIPELINE_STAGE_ALL_GRAPHICS_BIT |
VK_PIPELINE_STAGE_ALL_COMMANDS_BIT)) {
cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_PS_PARTIAL_FLUSH;
- } else if (src_stage_mask & (VK_PIPELINE_STAGE_TOP_OF_PIPE_BIT |
- VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT |
+ } else if (src_stage_mask & (VK_PIPELINE_STAGE_DRAW_INDIRECT_BIT |
VK_PIPELINE_STAGE_VERTEX_INPUT_BIT |
VK_PIPELINE_STAGE_VERTEX_SHADER_BIT)) {
cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_VS_PARTIAL_FLUSH;
switch ((VkAccessFlagBits)(1 << b)) {
case VK_ACCESS_INDIRECT_COMMAND_READ_BIT:
case VK_ACCESS_INDEX_READ_BIT:
- case VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT:
break;
case VK_ACCESS_UNIFORM_READ_BIT:
flush_bits |= RADV_CMD_FLAG_INV_VMEM_L1 | RADV_CMD_FLAG_INV_SMEM_L1;
break;
+ case VK_ACCESS_VERTEX_ATTRIBUTE_READ_BIT:
case VK_ACCESS_SHADER_READ_BIT:
case VK_ACCESS_TRANSFER_READ_BIT:
case VK_ACCESS_INPUT_ATTACHMENT_READ_BIT:
cmd_buffer->state.subpass = subpass;
- cmd_buffer->state.dirty |= RADV_CMD_DIRTY_RENDER_TARGETS;
+ cmd_buffer->state.dirty |= RADV_CMD_DIRTY_FRAMEBUFFER;
}
static VkResult
VkResult result = VK_SUCCESS;
uint32_t i;
- memset(pCommandBuffers, 0,
- sizeof(*pCommandBuffers)*pAllocateInfo->commandBufferCount);
-
for (i = 0; i < pAllocateInfo->commandBufferCount; i++) {
if (!list_empty(&pool->free_cmd_buffers)) {
break;
}
- if (result != VK_SUCCESS)
+ if (result != VK_SUCCESS) {
radv_FreeCommandBuffers(_device, pAllocateInfo->commandPool,
i, pCommandBuffers);
+ /* From the Vulkan 1.0.66 spec:
+ *
+ * "vkAllocateCommandBuffers can be used to create multiple
+ * command buffers. If the creation of any of those command
+ * buffers fails, the implementation must destroy all
+ * successfully created command buffer objects from this
+ * command, set all entries of the pCommandBuffers array to
+ * NULL and return the error."
+ */
+ memset(pCommandBuffers, 0,
+ sizeof(*pCommandBuffers) * pAllocateInfo->commandBufferCount);
+ }
+
return result;
}
{
struct radv_device *device = cmd_buffer->device;
if (device->gfx_init) {
- uint64_t va = device->ws->buffer_get_va(device->gfx_init);
- device->ws->cs_add_buffer(cmd_buffer->cs, device->gfx_init, 8);
+ uint64_t va = radv_buffer_get_va(device->gfx_init);
+ radv_cs_add_buffer(device->ws, cmd_buffer->cs, device->gfx_init, 8);
radeon_emit(cmd_buffer->cs, PKT3(PKT3_INDIRECT_BUFFER_CIK, 2, 0));
radeon_emit(cmd_buffer->cs, va);
radeon_emit(cmd_buffer->cs, va >> 32);
const VkCommandBufferBeginInfo *pBeginInfo)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
- VkResult result;
+ VkResult result = VK_SUCCESS;
- result = radv_reset_cmd_buffer(cmd_buffer);
- if (result != VK_SUCCESS)
- return result;
+ if (cmd_buffer->status != RADV_CMD_BUFFER_STATUS_INITIAL) {
+ /* If the command buffer has already been resetted with
+ * vkResetCommandBuffer, no need to do it again.
+ */
+ result = radv_reset_cmd_buffer(cmd_buffer);
+ if (result != VK_SUCCESS)
+ return result;
+ }
memset(&cmd_buffer->state, 0, sizeof(cmd_buffer->state));
cmd_buffer->state.last_primitive_reset_en = -1;
+ cmd_buffer->state.last_index_type = -1;
+ cmd_buffer->state.last_num_instances = -1;
+ cmd_buffer->state.last_vertex_offset = -1;
+ cmd_buffer->state.last_first_instance = -1;
cmd_buffer->usage_flags = pBeginInfo->flags;
/* setup initial configuration into command buffer */
switch (cmd_buffer->queue_family_index) {
case RADV_QUEUE_GENERAL:
emit_gfx_buffer_state(cmd_buffer);
- radv_set_db_count_control(cmd_buffer);
break;
case RADV_QUEUE_COMPUTE:
si_init_compute(cmd_buffer);
}
if (pBeginInfo->flags & VK_COMMAND_BUFFER_USAGE_RENDER_PASS_CONTINUE_BIT) {
+ assert(pBeginInfo->pInheritanceInfo);
cmd_buffer->state.framebuffer = radv_framebuffer_from_handle(pBeginInfo->pInheritanceInfo->framebuffer);
cmd_buffer->state.pass = radv_render_pass_from_handle(pBeginInfo->pInheritanceInfo->renderPass);
radv_cmd_buffer_set_subpass(cmd_buffer, subpass, false);
}
- radv_cmd_buffer_trace_emit(cmd_buffer);
+ if (unlikely(cmd_buffer->device->trace_bo))
+ radv_cmd_buffer_trace_emit(cmd_buffer);
+
+ cmd_buffer->status = RADV_CMD_BUFFER_STATUS_RECORDING;
+
return result;
}
const VkDeviceSize* pOffsets)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
- struct radv_vertex_binding *vb = cmd_buffer->state.vertex_bindings;
+ struct radv_vertex_binding *vb = cmd_buffer->vertex_bindings;
+ bool changed = false;
/* We have to defer setting up vertex buffer since we need the buffer
* stride from the pipeline. */
assert(firstBinding + bindingCount <= MAX_VBS);
for (uint32_t i = 0; i < bindingCount; i++) {
- vb[firstBinding + i].buffer = radv_buffer_from_handle(pBuffers[i]);
- vb[firstBinding + i].offset = pOffsets[i];
+ uint32_t idx = firstBinding + i;
+
+ if (!changed &&
+ (vb[idx].buffer != radv_buffer_from_handle(pBuffers[i]) ||
+ vb[idx].offset != pOffsets[i])) {
+ changed = true;
+ }
+
+ vb[idx].buffer = radv_buffer_from_handle(pBuffers[i]);
+ vb[idx].offset = pOffsets[i];
+
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs,
+ vb[idx].buffer->bo, 8);
+ }
+
+ if (!changed) {
+ /* No state changes. */
+ return;
}
- cmd_buffer->state.vb_dirty = true;
+ cmd_buffer->state.dirty |= RADV_CMD_DIRTY_VERTEX_BUFFER;
}
void radv_CmdBindIndexBuffer(
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
RADV_FROM_HANDLE(radv_buffer, index_buffer, buffer);
+ if (cmd_buffer->state.index_buffer == index_buffer &&
+ cmd_buffer->state.index_offset == offset &&
+ cmd_buffer->state.index_type == indexType) {
+ /* No state changes. */
+ return;
+ }
+
+ cmd_buffer->state.index_buffer = index_buffer;
+ cmd_buffer->state.index_offset = offset;
cmd_buffer->state.index_type = indexType; /* vk matches hw */
- cmd_buffer->state.index_va = cmd_buffer->device->ws->buffer_get_va(index_buffer->bo);
+ cmd_buffer->state.index_va = radv_buffer_get_va(index_buffer->bo);
cmd_buffer->state.index_va += index_buffer->offset + offset;
int index_size_shift = cmd_buffer->state.index_type ? 2 : 1;
cmd_buffer->state.max_index_count = (index_buffer->size - offset) >> index_size_shift;
cmd_buffer->state.dirty |= RADV_CMD_DIRTY_INDEX_BUFFER;
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, index_buffer->bo, 8);
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs, index_buffer->bo, 8);
}
-void radv_bind_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
- struct radv_descriptor_set *set,
- unsigned idx)
+static void
+radv_bind_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
+ VkPipelineBindPoint bind_point,
+ struct radv_descriptor_set *set, unsigned idx)
{
struct radeon_winsys *ws = cmd_buffer->device->ws;
- cmd_buffer->state.descriptors[idx] = set;
- cmd_buffer->state.descriptors_dirty |= (1u << idx);
+ radv_set_descriptor_set(cmd_buffer, bind_point, set, idx);
if (!set)
return;
assert(!(set->layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR));
- for (unsigned j = 0; j < set->layout->buffer_count; ++j)
- if (set->descriptors[j])
- ws->cs_add_buffer(cmd_buffer->cs, set->descriptors[j], 7);
+ if (!cmd_buffer->device->use_global_bo_list) {
+ for (unsigned j = 0; j < set->layout->buffer_count; ++j)
+ if (set->descriptors[j])
+ radv_cs_add_buffer(ws, cmd_buffer->cs, set->descriptors[j], 7);
+ }
if(set->bo)
- ws->cs_add_buffer(cmd_buffer->cs, set->bo, 8);
+ radv_cs_add_buffer(ws, cmd_buffer->cs, set->bo, 8);
}
void radv_CmdBindDescriptorSets(
RADV_FROM_HANDLE(radv_pipeline_layout, layout, _layout);
unsigned dyn_idx = 0;
+ const bool no_dynamic_bounds = cmd_buffer->device->instance->debug_flags & RADV_DEBUG_NO_DYNAMIC_BOUNDS;
+
for (unsigned i = 0; i < descriptorSetCount; ++i) {
unsigned idx = i + firstSet;
RADV_FROM_HANDLE(radv_descriptor_set, set, pDescriptorSets[i]);
- radv_bind_descriptor_set(cmd_buffer, set, idx);
+ radv_bind_descriptor_set(cmd_buffer, pipelineBindPoint, set, idx);
for(unsigned j = 0; j < set->layout->dynamic_offset_count; ++j, ++dyn_idx) {
unsigned idx = j + layout->set[i + firstSet].dynamic_offset_start;
uint64_t va = range->va + pDynamicOffsets[dyn_idx];
dst[0] = va;
dst[1] = S_008F04_BASE_ADDRESS_HI(va >> 32);
- dst[2] = range->size;
+ dst[2] = no_dynamic_bounds ? 0xffffffffu : range->size;
dst[3] = S_008F0C_DST_SEL_X(V_008F0C_SQ_SEL_X) |
S_008F0C_DST_SEL_Y(V_008F0C_SQ_SEL_Y) |
S_008F0C_DST_SEL_Z(V_008F0C_SQ_SEL_Z) |
static bool radv_init_push_descriptor_set(struct radv_cmd_buffer *cmd_buffer,
struct radv_descriptor_set *set,
- struct radv_descriptor_set_layout *layout)
+ struct radv_descriptor_set_layout *layout,
+ VkPipelineBindPoint bind_point)
{
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, bind_point);
set->size = layout->size;
set->layout = layout;
- if (cmd_buffer->push_descriptors.capacity < set->size) {
+ if (descriptors_state->push_set.capacity < set->size) {
size_t new_size = MAX2(set->size, 1024);
- new_size = MAX2(new_size, 2 * cmd_buffer->push_descriptors.capacity);
+ new_size = MAX2(new_size, 2 * descriptors_state->push_set.capacity);
new_size = MIN2(new_size, 96 * MAX_PUSH_DESCRIPTORS);
free(set->mapped_ptr);
set->mapped_ptr = malloc(new_size);
if (!set->mapped_ptr) {
- cmd_buffer->push_descriptors.capacity = 0;
+ descriptors_state->push_set.capacity = 0;
cmd_buffer->record_result = VK_ERROR_OUT_OF_HOST_MEMORY;
return false;
}
- cmd_buffer->push_descriptors.capacity = new_size;
+ descriptors_state->push_set.capacity = new_size;
}
return true;
(void**) &push_set->mapped_ptr))
return;
- push_set->va = cmd_buffer->device->ws->buffer_get_va(cmd_buffer->upload.upload_bo);
+ push_set->va = radv_buffer_get_va(cmd_buffer->upload.upload_bo);
push_set->va += bo_offset;
radv_update_descriptor_sets(cmd_buffer->device, cmd_buffer,
radv_descriptor_set_to_handle(push_set),
descriptorWriteCount, pDescriptorWrites, 0, NULL);
- cmd_buffer->state.descriptors[set] = push_set;
- cmd_buffer->state.descriptors_dirty |= (1u << set);
+ radv_set_descriptor_set(cmd_buffer, pipelineBindPoint, push_set, set);
}
void radv_CmdPushDescriptorSetKHR(
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
RADV_FROM_HANDLE(radv_pipeline_layout, layout, _layout);
- struct radv_descriptor_set *push_set = &cmd_buffer->push_descriptors.set;
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, pipelineBindPoint);
+ struct radv_descriptor_set *push_set = &descriptors_state->push_set.set;
assert(layout->set[set].layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR);
- if (!radv_init_push_descriptor_set(cmd_buffer, push_set, layout->set[set].layout))
+ if (!radv_init_push_descriptor_set(cmd_buffer, push_set,
+ layout->set[set].layout,
+ pipelineBindPoint))
return;
radv_update_descriptor_sets(cmd_buffer->device, cmd_buffer,
radv_descriptor_set_to_handle(push_set),
descriptorWriteCount, pDescriptorWrites, 0, NULL);
- cmd_buffer->state.descriptors[set] = push_set;
- cmd_buffer->state.descriptors_dirty |= (1u << set);
- cmd_buffer->state.push_descriptors_dirty = true;
+ radv_set_descriptor_set(cmd_buffer, pipelineBindPoint, push_set, set);
+ descriptors_state->push_dirty = true;
}
void radv_CmdPushDescriptorSetWithTemplateKHR(
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
RADV_FROM_HANDLE(radv_pipeline_layout, layout, _layout);
- struct radv_descriptor_set *push_set = &cmd_buffer->push_descriptors.set;
+ RADV_FROM_HANDLE(radv_descriptor_update_template, templ, descriptorUpdateTemplate);
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, templ->bind_point);
+ struct radv_descriptor_set *push_set = &descriptors_state->push_set.set;
assert(layout->set[set].layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR);
- if (!radv_init_push_descriptor_set(cmd_buffer, push_set, layout->set[set].layout))
+ if (!radv_init_push_descriptor_set(cmd_buffer, push_set,
+ layout->set[set].layout,
+ templ->bind_point))
return;
radv_update_descriptor_set_with_template(cmd_buffer->device, cmd_buffer, push_set,
descriptorUpdateTemplate, pData);
- cmd_buffer->state.descriptors[set] = push_set;
- cmd_buffer->state.descriptors_dirty |= (1u << set);
- cmd_buffer->state.push_descriptors_dirty = true;
+ radv_set_descriptor_set(cmd_buffer, templ->bind_point, push_set, set);
+ descriptors_state->push_dirty = true;
}
void radv_CmdPushConstants(VkCommandBuffer commandBuffer,
si_emit_cache_flush(cmd_buffer);
}
+ vk_free(&cmd_buffer->pool->alloc, cmd_buffer->state.attachments);
+
if (!cmd_buffer->device->ws->cs_finalize(cmd_buffer->cs))
- return VK_ERROR_OUT_OF_DEVICE_MEMORY;
+ return vk_error(VK_ERROR_OUT_OF_DEVICE_MEMORY);
+
+ cmd_buffer->status = RADV_CMD_BUFFER_STATUS_EXECUTABLE;
return cmd_buffer->record_result;
}
static void
radv_emit_compute_pipeline(struct radv_cmd_buffer *cmd_buffer)
{
- struct radeon_winsys *ws = cmd_buffer->device->ws;
- struct radv_shader_variant *compute_shader;
struct radv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
- uint64_t va;
if (!pipeline || pipeline == cmd_buffer->state.emitted_compute_pipeline)
return;
cmd_buffer->state.emitted_compute_pipeline = pipeline;
- compute_shader = pipeline->shaders[MESA_SHADER_COMPUTE];
- va = ws->buffer_get_va(compute_shader->bo) + compute_shader->bo_offset;
-
- ws->cs_add_buffer(cmd_buffer->cs, compute_shader->bo, 8);
- radv_emit_prefetch(cmd_buffer, va, compute_shader->code_size);
-
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
- cmd_buffer->cs, 16);
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B830_COMPUTE_PGM_LO, 2);
- radeon_emit(cmd_buffer->cs, va >> 8);
- radeon_emit(cmd_buffer->cs, va >> 40);
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B848_COMPUTE_PGM_RSRC1, 2);
- radeon_emit(cmd_buffer->cs, compute_shader->rsrc1);
- radeon_emit(cmd_buffer->cs, compute_shader->rsrc2);
-
+ radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, pipeline->cs.cdw);
+ radeon_emit_array(cmd_buffer->cs, pipeline->cs.buf, pipeline->cs.cdw);
cmd_buffer->compute_scratch_size_needed =
MAX2(cmd_buffer->compute_scratch_size_needed,
pipeline->max_waves * pipeline->scratch_bytes_per_wave);
- /* change these once we have scratch support */
- radeon_set_sh_reg(cmd_buffer->cs, R_00B860_COMPUTE_TMPRING_SIZE,
- S_00B860_WAVES(pipeline->max_waves) |
- S_00B860_WAVESIZE(pipeline->scratch_bytes_per_wave >> 10));
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B81C_COMPUTE_NUM_THREAD_X, 3);
- radeon_emit(cmd_buffer->cs,
- S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[0]));
- radeon_emit(cmd_buffer->cs,
- S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[1]));
- radeon_emit(cmd_buffer->cs,
- S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[2]));
+ radv_cs_add_buffer(cmd_buffer->device->ws, cmd_buffer->cs,
+ pipeline->shaders[MESA_SHADER_COMPUTE]->bo, 8);
- assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_save_pipeline(cmd_buffer, pipeline, RING_COMPUTE);
+ if (unlikely(cmd_buffer->device->trace_bo))
+ radv_save_pipeline(cmd_buffer, pipeline, RING_COMPUTE);
}
-static void radv_mark_descriptor_sets_dirty(struct radv_cmd_buffer *cmd_buffer)
+static void radv_mark_descriptor_sets_dirty(struct radv_cmd_buffer *cmd_buffer,
+ VkPipelineBindPoint bind_point)
{
- for (unsigned i = 0; i < MAX_SETS; i++) {
- if (cmd_buffer->state.descriptors[i])
- cmd_buffer->state.descriptors_dirty |= (1u << i);
- }
+ struct radv_descriptor_state *descriptors_state =
+ radv_get_descriptors_state(cmd_buffer, bind_point);
+
+ descriptors_state->dirty |= descriptors_state->valid;
}
void radv_CmdBindPipeline(
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
RADV_FROM_HANDLE(radv_pipeline, pipeline, _pipeline);
- radv_mark_descriptor_sets_dirty(cmd_buffer);
-
switch (pipelineBindPoint) {
case VK_PIPELINE_BIND_POINT_COMPUTE:
+ if (cmd_buffer->state.compute_pipeline == pipeline)
+ return;
+ radv_mark_descriptor_sets_dirty(cmd_buffer, pipelineBindPoint);
+
cmd_buffer->state.compute_pipeline = pipeline;
cmd_buffer->push_constant_stages |= VK_SHADER_STAGE_COMPUTE_BIT;
break;
case VK_PIPELINE_BIND_POINT_GRAPHICS:
+ if (cmd_buffer->state.pipeline == pipeline)
+ return;
+ radv_mark_descriptor_sets_dirty(cmd_buffer, pipelineBindPoint);
+
cmd_buffer->state.pipeline = pipeline;
if (!pipeline)
break;
cmd_buffer->state.dirty |= RADV_CMD_DIRTY_PIPELINE;
cmd_buffer->push_constant_stages |= pipeline->active_stages;
- /* Apply the dynamic state from the pipeline */
- cmd_buffer->state.dirty |= pipeline->dynamic_state_mask;
- radv_dynamic_state_copy(&cmd_buffer->state.dynamic,
- &pipeline->dynamic_state,
- pipeline->dynamic_state_mask);
+ /* the new vertex shader might not have the same user regs */
+ cmd_buffer->state.last_first_instance = -1;
+ cmd_buffer->state.last_vertex_offset = -1;
+
+ /* Prefetch all pipeline shaders at first draw time. */
+ cmd_buffer->state.prefetch_L2_mask |= RADV_PREFETCH_SHADERS;
+
+ radv_bind_dynamic_state(cmd_buffer, &pipeline->dynamic_state);
if (pipeline->graphics.esgs_ring_size > cmd_buffer->esgs_ring_size_needed)
cmd_buffer->esgs_ring_size_needed = pipeline->graphics.esgs_ring_size;
cmd_buffer->tess_rings_needed = true;
if (radv_pipeline_has_gs(pipeline)) {
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(cmd_buffer->state.pipeline, MESA_SHADER_GEOMETRY,
+ struct radv_userdata_info *loc = radv_lookup_user_sgpr(cmd_buffer->state.pipeline, MESA_SHADER_GEOMETRY,
AC_UD_SCRATCH_RING_OFFSETS);
if (cmd_buffer->ring_offsets_idx == -1)
cmd_buffer->ring_offsets_idx = loc->sgpr_idx;
const VkViewport* pViewports)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ struct radv_cmd_state *state = &cmd_buffer->state;
+ MAYBE_UNUSED const uint32_t total_count = firstViewport + viewportCount;
- const uint32_t total_count = firstViewport + viewportCount;
- if (cmd_buffer->state.dynamic.viewport.count < total_count)
- cmd_buffer->state.dynamic.viewport.count = total_count;
+ assert(firstViewport < MAX_VIEWPORTS);
+ assert(total_count >= 1 && total_count <= MAX_VIEWPORTS);
- memcpy(cmd_buffer->state.dynamic.viewport.viewports + firstViewport,
- pViewports, viewportCount * sizeof(*pViewports));
+ if (cmd_buffer->device->physical_device->has_scissor_bug) {
+ /* Try to skip unnecessary PS partial flushes when the viewports
+ * don't change.
+ */
+ if (!(state->dirty & (RADV_CMD_DIRTY_DYNAMIC_VIEWPORT |
+ RADV_CMD_DIRTY_DYNAMIC_SCISSOR)) &&
+ !memcmp(state->dynamic.viewport.viewports + firstViewport,
+ pViewports, viewportCount * sizeof(*pViewports))) {
+ return;
+ }
+ }
+
+ memcpy(state->dynamic.viewport.viewports + firstViewport, pViewports,
+ viewportCount * sizeof(*pViewports));
- cmd_buffer->state.dirty |= RADV_CMD_DIRTY_DYNAMIC_VIEWPORT;
+ state->dirty |= RADV_CMD_DIRTY_DYNAMIC_VIEWPORT;
}
void radv_CmdSetScissor(
const VkRect2D* pScissors)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ struct radv_cmd_state *state = &cmd_buffer->state;
+ MAYBE_UNUSED const uint32_t total_count = firstScissor + scissorCount;
- const uint32_t total_count = firstScissor + scissorCount;
- if (cmd_buffer->state.dynamic.scissor.count < total_count)
- cmd_buffer->state.dynamic.scissor.count = total_count;
+ assert(firstScissor < MAX_SCISSORS);
+ assert(total_count >= 1 && total_count <= MAX_SCISSORS);
- memcpy(cmd_buffer->state.dynamic.scissor.scissors + firstScissor,
- pScissors, scissorCount * sizeof(*pScissors));
- cmd_buffer->state.dirty |= RADV_CMD_DIRTY_DYNAMIC_SCISSOR;
+ if (cmd_buffer->device->physical_device->has_scissor_bug) {
+ /* Try to skip unnecessary PS partial flushes when the scissors
+ * don't change.
+ */
+ if (!(state->dirty & (RADV_CMD_DIRTY_DYNAMIC_VIEWPORT |
+ RADV_CMD_DIRTY_DYNAMIC_SCISSOR)) &&
+ !memcmp(state->dynamic.scissor.scissors + firstScissor,
+ pScissors, scissorCount * sizeof(*pScissors))) {
+ return;
+ }
+ }
+
+ memcpy(state->dynamic.scissor.scissors + firstScissor, pScissors,
+ scissorCount * sizeof(*pScissors));
+
+ state->dirty |= RADV_CMD_DIRTY_DYNAMIC_SCISSOR;
}
void radv_CmdSetLineWidth(
cmd_buffer->state.dirty |= RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE;
}
+void radv_CmdSetDiscardRectangleEXT(
+ VkCommandBuffer commandBuffer,
+ uint32_t firstDiscardRectangle,
+ uint32_t discardRectangleCount,
+ const VkRect2D* pDiscardRectangles)
+{
+ RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ struct radv_cmd_state *state = &cmd_buffer->state;
+ MAYBE_UNUSED const uint32_t total_count = firstDiscardRectangle + discardRectangleCount;
+
+ assert(firstDiscardRectangle < MAX_DISCARD_RECTANGLES);
+ assert(total_count >= 1 && total_count <= MAX_DISCARD_RECTANGLES);
+
+ typed_memcpy(&state->dynamic.discard_rectangle.rectangles[firstDiscardRectangle],
+ pDiscardRectangles, discardRectangleCount);
+
+ state->dirty |= RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE;
+}
+
void radv_CmdExecuteCommands(
VkCommandBuffer commandBuffer,
uint32_t commandBufferCount,
{
RADV_FROM_HANDLE(radv_cmd_buffer, primary, commandBuffer);
+ assert(commandBufferCount > 0);
+
/* Emit pending flushes on primary prior to executing secondary */
si_emit_cache_flush(primary);
assert(secondary->ring_offsets_idx == primary->ring_offsets_idx);
}
primary->device->ws->cs_execute_secondary(primary->cs, secondary->cs);
- }
- /* if we execute secondary we need to re-emit out pipelines */
- if (commandBufferCount) {
- primary->state.emitted_pipeline = NULL;
- primary->state.emitted_compute_pipeline = NULL;
- primary->state.dirty |= RADV_CMD_DIRTY_PIPELINE;
- primary->state.dirty |= RADV_CMD_DIRTY_DYNAMIC_ALL;
- primary->state.last_primitive_reset_en = -1;
- primary->state.last_primitive_reset_index = 0;
- radv_mark_descriptor_sets_dirty(primary);
+
+ /* When the secondary command buffer is compute only we don't
+ * need to re-emit the current graphics pipeline.
+ */
+ if (secondary->state.emitted_pipeline) {
+ primary->state.emitted_pipeline =
+ secondary->state.emitted_pipeline;
+ }
+
+ /* When the secondary command buffer is graphics only we don't
+ * need to re-emit the current compute pipeline.
+ */
+ if (secondary->state.emitted_compute_pipeline) {
+ primary->state.emitted_compute_pipeline =
+ secondary->state.emitted_compute_pipeline;
+ }
+
+ /* Only re-emit the draw packets when needed. */
+ if (secondary->state.last_primitive_reset_en != -1) {
+ primary->state.last_primitive_reset_en =
+ secondary->state.last_primitive_reset_en;
+ }
+
+ if (secondary->state.last_primitive_reset_index) {
+ primary->state.last_primitive_reset_index =
+ secondary->state.last_primitive_reset_index;
+ }
+
+ if (secondary->state.last_ia_multi_vgt_param) {
+ primary->state.last_ia_multi_vgt_param =
+ secondary->state.last_ia_multi_vgt_param;
+ }
+
+ primary->state.last_first_instance = secondary->state.last_first_instance;
+ primary->state.last_num_instances = secondary->state.last_num_instances;
+ primary->state.last_vertex_offset = secondary->state.last_vertex_offset;
+
+ if (secondary->state.last_index_type != -1) {
+ primary->state.last_index_type =
+ secondary->state.last_index_type;
+ }
}
+
+ /* After executing commands from secondary buffers we have to dirty
+ * some states.
+ */
+ primary->state.dirty |= RADV_CMD_DIRTY_PIPELINE |
+ RADV_CMD_DIRTY_INDEX_BUFFER |
+ RADV_CMD_DIRTY_DYNAMIC_ALL;
+ radv_mark_descriptor_sets_dirty(primary, VK_PIPELINE_BIND_POINT_GRAPHICS);
+ radv_mark_descriptor_sets_dirty(primary, VK_PIPELINE_BIND_POINT_COMPUTE);
}
VkResult radv_CreateCommandPool(
return VK_SUCCESS;
}
-void radv_TrimCommandPoolKHR(
+void radv_TrimCommandPool(
VkDevice device,
VkCommandPool commandPool,
VkCommandPoolTrimFlagsKHR flags)
cmd_buffer->state.framebuffer = framebuffer;
cmd_buffer->state.pass = pass;
cmd_buffer->state.render_area = pRenderPassBegin->renderArea;
+
result = radv_cmd_state_setup_attachments(cmd_buffer, pass, pRenderPassBegin);
if (result != VK_SUCCESS)
- cmd_buffer->record_result = result;
+ return;
radv_cmd_buffer_set_subpass(cmd_buffer, pass->subpasses, true);
assert(cmd_buffer->cs->cdw <= cdw_max);
for (unsigned stage = 0; stage < MESA_SHADER_STAGES; ++stage) {
if (!pipeline->shaders[stage])
continue;
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(pipeline, stage, AC_UD_VIEW_INDEX);
+ struct radv_userdata_info *loc = radv_lookup_user_sgpr(pipeline, stage, AC_UD_VIEW_INDEX);
if (loc->sgpr_idx == -1)
continue;
- uint32_t base_reg = radv_shader_stage_to_user_data_0(stage, radv_pipeline_has_gs(pipeline), radv_pipeline_has_tess(pipeline));
+ uint32_t base_reg = pipeline->user_data_0[stage];
radeon_set_sh_reg(cmd_buffer->cs, base_reg + loc->sgpr_idx * 4, index);
}
if (pipeline->gs_copy_shader) {
- struct ac_userdata_info *loc = &pipeline->gs_copy_shader->info.user_sgprs_locs.shader_data[AC_UD_VIEW_INDEX];
+ struct radv_userdata_info *loc = &pipeline->gs_copy_shader->info.user_sgprs_locs.shader_data[AC_UD_VIEW_INDEX];
if (loc->sgpr_idx != -1) {
uint32_t base_reg = R_00B130_SPI_SHADER_USER_DATA_VS_0;
radeon_set_sh_reg(cmd_buffer->cs, base_reg + loc->sgpr_idx * 4, index);
S_0287F0_USE_OPAQUE(0));
}
-void radv_CmdDraw(
- VkCommandBuffer commandBuffer,
- uint32_t vertexCount,
- uint32_t instanceCount,
- uint32_t firstVertex,
- uint32_t firstInstance)
-{
- RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
-
- radv_cmd_buffer_flush_state(cmd_buffer, false, (instanceCount > 1), false, vertexCount);
-
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 20 * MAX_VIEWS);
-
- assert(cmd_buffer->state.pipeline->graphics.vtx_base_sgpr);
- radeon_set_sh_reg_seq(cmd_buffer->cs, cmd_buffer->state.pipeline->graphics.vtx_base_sgpr,
- cmd_buffer->state.pipeline->graphics.vtx_emit_num);
- radeon_emit(cmd_buffer->cs, firstVertex);
- radeon_emit(cmd_buffer->cs, firstInstance);
- if (cmd_buffer->state.pipeline->graphics.vtx_emit_num == 3)
- radeon_emit(cmd_buffer->cs, 0);
-
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_NUM_INSTANCES, 0, cmd_buffer->state.predicating));
- radeon_emit(cmd_buffer->cs, instanceCount);
-
- if (!cmd_buffer->state.subpass->view_mask) {
- radv_cs_emit_draw_packet(cmd_buffer, vertexCount);
- } else {
- unsigned i;
- for_each_bit(i, cmd_buffer->state.subpass->view_mask) {
- radv_emit_view_index(cmd_buffer, i);
-
- radv_cs_emit_draw_packet(cmd_buffer, vertexCount);
- }
- }
-
- assert(cmd_buffer->cs->cdw <= cdw_max);
-
- radv_cmd_buffer_after_draw(cmd_buffer);
-}
-
-
static void
radv_cs_emit_draw_indexed_packet(struct radv_cmd_buffer *cmd_buffer,
uint64_t index_va,
radeon_emit(cmd_buffer->cs, PKT3(PKT3_DRAW_INDEX_2, 4, false));
radeon_emit(cmd_buffer->cs, cmd_buffer->state.max_index_count);
radeon_emit(cmd_buffer->cs, index_va);
- radeon_emit(cmd_buffer->cs, (index_va >> 32UL) & 0xFF);
+ radeon_emit(cmd_buffer->cs, index_va >> 32);
radeon_emit(cmd_buffer->cs, index_count);
radeon_emit(cmd_buffer->cs, V_0287F0_DI_SRC_SEL_DMA);
}
-void radv_CmdDrawIndexed(
- VkCommandBuffer commandBuffer,
- uint32_t indexCount,
- uint32_t instanceCount,
- uint32_t firstIndex,
- int32_t vertexOffset,
- uint32_t firstInstance)
-{
- RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
- int index_size = cmd_buffer->state.index_type ? 4 : 2;
- uint64_t index_va;
-
- radv_cmd_buffer_flush_state(cmd_buffer, true, (instanceCount > 1), false, indexCount);
-
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 26 * MAX_VIEWS);
-
- if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
- radeon_set_uconfig_reg_idx(cmd_buffer->cs, R_03090C_VGT_INDEX_TYPE,
- 2, cmd_buffer->state.index_type);
- } else {
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_INDEX_TYPE, 0, 0));
- radeon_emit(cmd_buffer->cs, cmd_buffer->state.index_type);
- }
-
- assert(cmd_buffer->state.pipeline->graphics.vtx_base_sgpr);
- radeon_set_sh_reg_seq(cmd_buffer->cs, cmd_buffer->state.pipeline->graphics.vtx_base_sgpr,
- cmd_buffer->state.pipeline->graphics.vtx_emit_num);
- radeon_emit(cmd_buffer->cs, vertexOffset);
- radeon_emit(cmd_buffer->cs, firstInstance);
- if (cmd_buffer->state.pipeline->graphics.vtx_emit_num == 3)
- radeon_emit(cmd_buffer->cs, 0);
-
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_NUM_INSTANCES, 0, 0));
- radeon_emit(cmd_buffer->cs, instanceCount);
-
- index_va = cmd_buffer->state.index_va;
- index_va += firstIndex * index_size;
- if (!cmd_buffer->state.subpass->view_mask) {
- radv_cs_emit_draw_indexed_packet(cmd_buffer, index_va, indexCount);
- } else {
- unsigned i;
- for_each_bit(i, cmd_buffer->state.subpass->view_mask) {
- radv_emit_view_index(cmd_buffer, i);
-
- radv_cs_emit_draw_indexed_packet(cmd_buffer, index_va, indexCount);
- }
- }
-
- assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_after_draw(cmd_buffer);
-}
-
static void
radv_cs_emit_indirect_draw_packet(struct radv_cmd_buffer *cmd_buffer,
bool indexed,
struct radeon_winsys_cs *cs = cmd_buffer->cs;
unsigned di_src_sel = indexed ? V_0287F0_DI_SRC_SEL_DMA
: V_0287F0_DI_SRC_SEL_AUTO_INDEX;
- bool draw_id_enable = cmd_buffer->state.pipeline->shaders[MESA_SHADER_VERTEX]->info.info.vs.needs_draw_id;
+ bool draw_id_enable = radv_get_vertex_shader(cmd_buffer->state.pipeline)->info.info.vs.needs_draw_id;
uint32_t base_reg = cmd_buffer->state.pipeline->graphics.vtx_base_sgpr;
assert(base_reg);
+ /* just reset draw state for vertex data */
+ cmd_buffer->state.last_first_instance = -1;
+ cmd_buffer->state.last_num_instances = -1;
+ cmd_buffer->state.last_vertex_offset = -1;
+
if (draw_count == 1 && !count_va && !draw_id_enable) {
radeon_emit(cs, PKT3(indexed ? PKT3_DRAW_INDEX_INDIRECT :
PKT3_DRAW_INDIRECT, 3, false));
}
}
+struct radv_draw_info {
+ /**
+ * Number of vertices.
+ */
+ uint32_t count;
+
+ /**
+ * Index of the first vertex.
+ */
+ int32_t vertex_offset;
+
+ /**
+ * First instance id.
+ */
+ uint32_t first_instance;
+
+ /**
+ * Number of instances.
+ */
+ uint32_t instance_count;
+
+ /**
+ * First index (indexed draws only).
+ */
+ uint32_t first_index;
+
+ /**
+ * Whether it's an indexed draw.
+ */
+ bool indexed;
+
+ /**
+ * Indirect draw parameters resource.
+ */
+ struct radv_buffer *indirect;
+ uint64_t indirect_offset;
+ uint32_t stride;
+
+ /**
+ * Draw count parameters resource.
+ */
+ struct radv_buffer *count_buffer;
+ uint64_t count_buffer_offset;
+};
+
static void
-radv_emit_indirect_draw(struct radv_cmd_buffer *cmd_buffer,
- VkBuffer _buffer,
- VkDeviceSize offset,
- VkBuffer _count_buffer,
- VkDeviceSize count_offset,
- uint32_t draw_count,
- uint32_t stride,
- bool indexed)
+radv_emit_draw_packets(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_draw_info *info)
{
- RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
- RADV_FROM_HANDLE(radv_buffer, count_buffer, _count_buffer);
+ struct radv_cmd_state *state = &cmd_buffer->state;
+ struct radeon_winsys *ws = cmd_buffer->device->ws;
struct radeon_winsys_cs *cs = cmd_buffer->cs;
- uint64_t indirect_va = cmd_buffer->device->ws->buffer_get_va(buffer->bo);
- indirect_va += offset + buffer->offset;
- uint64_t count_va = 0;
+ if (info->indirect) {
+ uint64_t va = radv_buffer_get_va(info->indirect->bo);
+ uint64_t count_va = 0;
- if (count_buffer) {
- count_va = cmd_buffer->device->ws->buffer_get_va(count_buffer->bo);
- count_va += count_offset + count_buffer->offset;
- }
+ va += info->indirect->offset + info->indirect_offset;
- if (!draw_count)
- return;
+ radv_cs_add_buffer(ws, cs, info->indirect->bo, 8);
+
+ radeon_emit(cs, PKT3(PKT3_SET_BASE, 2, 0));
+ radeon_emit(cs, 1);
+ radeon_emit(cs, va);
+ radeon_emit(cs, va >> 32);
- cmd_buffer->device->ws->cs_add_buffer(cs, buffer->bo, 8);
+ if (info->count_buffer) {
+ count_va = radv_buffer_get_va(info->count_buffer->bo);
+ count_va += info->count_buffer->offset +
+ info->count_buffer_offset;
- radeon_emit(cs, PKT3(PKT3_SET_BASE, 2, 0));
- radeon_emit(cs, 1);
- radeon_emit(cs, indirect_va);
- radeon_emit(cs, indirect_va >> 32);
+ radv_cs_add_buffer(ws, cs, info->count_buffer->bo, 8);
+ }
- if (!cmd_buffer->state.subpass->view_mask) {
- radv_cs_emit_indirect_draw_packet(cmd_buffer, indexed, draw_count, count_va, stride);
+ if (!state->subpass->view_mask) {
+ radv_cs_emit_indirect_draw_packet(cmd_buffer,
+ info->indexed,
+ info->count,
+ count_va,
+ info->stride);
+ } else {
+ unsigned i;
+ for_each_bit(i, state->subpass->view_mask) {
+ radv_emit_view_index(cmd_buffer, i);
+
+ radv_cs_emit_indirect_draw_packet(cmd_buffer,
+ info->indexed,
+ info->count,
+ count_va,
+ info->stride);
+ }
+ }
} else {
- unsigned i;
- for_each_bit(i, cmd_buffer->state.subpass->view_mask) {
- radv_emit_view_index(cmd_buffer, i);
+ assert(state->pipeline->graphics.vtx_base_sgpr);
+
+ if (info->vertex_offset != state->last_vertex_offset ||
+ info->first_instance != state->last_first_instance) {
+ radeon_set_sh_reg_seq(cs, state->pipeline->graphics.vtx_base_sgpr,
+ state->pipeline->graphics.vtx_emit_num);
+
+ radeon_emit(cs, info->vertex_offset);
+ radeon_emit(cs, info->first_instance);
+ if (state->pipeline->graphics.vtx_emit_num == 3)
+ radeon_emit(cs, 0);
+ state->last_first_instance = info->first_instance;
+ state->last_vertex_offset = info->vertex_offset;
+ }
- radv_cs_emit_indirect_draw_packet(cmd_buffer, indexed, draw_count, count_va, stride);
+ if (state->last_num_instances != info->instance_count) {
+ radeon_emit(cs, PKT3(PKT3_NUM_INSTANCES, 0, false));
+ radeon_emit(cs, info->instance_count);
+ state->last_num_instances = info->instance_count;
+ }
+
+ if (info->indexed) {
+ int index_size = state->index_type ? 4 : 2;
+ uint64_t index_va;
+
+ index_va = state->index_va;
+ index_va += info->first_index * index_size;
+
+ if (!state->subpass->view_mask) {
+ radv_cs_emit_draw_indexed_packet(cmd_buffer,
+ index_va,
+ info->count);
+ } else {
+ unsigned i;
+ for_each_bit(i, state->subpass->view_mask) {
+ radv_emit_view_index(cmd_buffer, i);
+
+ radv_cs_emit_draw_indexed_packet(cmd_buffer,
+ index_va,
+ info->count);
+ }
+ }
+ } else {
+ if (!state->subpass->view_mask) {
+ radv_cs_emit_draw_packet(cmd_buffer, info->count);
+ } else {
+ unsigned i;
+ for_each_bit(i, state->subpass->view_mask) {
+ radv_emit_view_index(cmd_buffer, i);
+
+ radv_cs_emit_draw_packet(cmd_buffer,
+ info->count);
+ }
+ }
}
}
- radv_cmd_buffer_after_draw(cmd_buffer);
}
static void
-radv_cmd_draw_indirect_count(VkCommandBuffer commandBuffer,
- VkBuffer buffer,
- VkDeviceSize offset,
- VkBuffer countBuffer,
- VkDeviceSize countBufferOffset,
- uint32_t maxDrawCount,
- uint32_t stride)
+radv_emit_all_graphics_states(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_draw_info *info)
{
- RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
- radv_cmd_buffer_flush_state(cmd_buffer, false, false, true, 0);
+ if ((cmd_buffer->state.dirty & RADV_CMD_DIRTY_FRAMEBUFFER) ||
+ cmd_buffer->state.emitted_pipeline != cmd_buffer->state.pipeline)
+ radv_emit_rbplus_state(cmd_buffer);
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
- cmd_buffer->cs, 24 * MAX_VIEWS);
+ if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_PIPELINE)
+ radv_emit_graphics_pipeline(cmd_buffer);
- radv_emit_indirect_draw(cmd_buffer, buffer, offset,
- countBuffer, countBufferOffset, maxDrawCount, stride, false);
+ if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_FRAMEBUFFER)
+ radv_emit_framebuffer_state(cmd_buffer);
- assert(cmd_buffer->cs->cdw <= cdw_max);
+ if (info->indexed) {
+ if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_INDEX_BUFFER)
+ radv_emit_index_buffer(cmd_buffer);
+ } else {
+ /* On CI and later, non-indexed draws overwrite VGT_INDEX_TYPE,
+ * so the state must be re-emitted before the next indexed
+ * draw.
+ */
+ if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK) {
+ cmd_buffer->state.last_index_type = -1;
+ cmd_buffer->state.dirty |= RADV_CMD_DIRTY_INDEX_BUFFER;
+ }
+ }
+
+ radv_cmd_buffer_flush_dynamic_state(cmd_buffer);
+
+ radv_emit_draw_registers(cmd_buffer, info->indexed,
+ info->instance_count > 1, info->indirect,
+ info->indirect ? 0 : info->count);
}
static void
-radv_cmd_draw_indexed_indirect_count(
- VkCommandBuffer commandBuffer,
- VkBuffer buffer,
- VkDeviceSize offset,
- VkBuffer countBuffer,
- VkDeviceSize countBufferOffset,
- uint32_t maxDrawCount,
- uint32_t stride)
-{
- RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
- uint64_t index_va;
- radv_cmd_buffer_flush_state(cmd_buffer, true, false, true, 0);
+radv_draw(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_draw_info *info)
+{
+ bool has_prefetch =
+ cmd_buffer->device->physical_device->rad_info.chip_class >= CIK;
+ bool pipeline_is_dirty =
+ (cmd_buffer->state.dirty & RADV_CMD_DIRTY_PIPELINE) &&
+ cmd_buffer->state.pipeline &&
+ cmd_buffer->state.pipeline != cmd_buffer->state.emitted_pipeline;
+
+ MAYBE_UNUSED unsigned cdw_max =
+ radeon_check_space(cmd_buffer->device->ws,
+ cmd_buffer->cs, 4096);
+
+ /* Use optimal packet order based on whether we need to sync the
+ * pipeline.
+ */
+ if (cmd_buffer->state.flush_bits & (RADV_CMD_FLAG_FLUSH_AND_INV_CB |
+ RADV_CMD_FLAG_FLUSH_AND_INV_DB |
+ RADV_CMD_FLAG_PS_PARTIAL_FLUSH |
+ RADV_CMD_FLAG_CS_PARTIAL_FLUSH)) {
+ /* If we have to wait for idle, set all states first, so that
+ * all SET packets are processed in parallel with previous draw
+ * calls. Then upload descriptors, set shader pointers, and
+ * draw, and prefetch at the end. This ensures that the time
+ * the CUs are idle is very short. (there are only SET_SH
+ * packets between the wait and the draw)
+ */
+ radv_emit_all_graphics_states(cmd_buffer, info);
+ si_emit_cache_flush(cmd_buffer);
+ /* <-- CUs are idle here --> */
- index_va = cmd_buffer->state.index_va;
+ radv_upload_graphics_shader_descriptors(cmd_buffer, pipeline_is_dirty);
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 31 * MAX_VIEWS);
+ radv_emit_draw_packets(cmd_buffer, info);
+ /* <-- CUs are busy here --> */
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_INDEX_TYPE, 0, 0));
- radeon_emit(cmd_buffer->cs, cmd_buffer->state.index_type);
+ /* Start prefetches after the draw has been started. Both will
+ * run in parallel, but starting the draw first is more
+ * important.
+ */
+ if (has_prefetch && cmd_buffer->state.prefetch_L2_mask) {
+ radv_emit_prefetch_L2(cmd_buffer,
+ cmd_buffer->state.pipeline, false);
+ }
+ } else {
+ /* If we don't wait for idle, start prefetches first, then set
+ * states, and draw at the end.
+ */
+ si_emit_cache_flush(cmd_buffer);
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_INDEX_BASE, 1, 0));
- radeon_emit(cmd_buffer->cs, index_va);
- radeon_emit(cmd_buffer->cs, index_va >> 32);
+ if (has_prefetch && cmd_buffer->state.prefetch_L2_mask) {
+ /* Only prefetch the vertex shader and VBO descriptors
+ * in order to start the draw as soon as possible.
+ */
+ radv_emit_prefetch_L2(cmd_buffer,
+ cmd_buffer->state.pipeline, true);
+ }
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_INDEX_BUFFER_SIZE, 0, 0));
- radeon_emit(cmd_buffer->cs, cmd_buffer->state.max_index_count);
+ radv_upload_graphics_shader_descriptors(cmd_buffer, pipeline_is_dirty);
+
+ radv_emit_all_graphics_states(cmd_buffer, info);
+ radv_emit_draw_packets(cmd_buffer, info);
- radv_emit_indirect_draw(cmd_buffer, buffer, offset,
- countBuffer, countBufferOffset, maxDrawCount, stride, true);
+ /* Prefetch the remaining shaders after the draw has been
+ * started.
+ */
+ if (has_prefetch && cmd_buffer->state.prefetch_L2_mask) {
+ radv_emit_prefetch_L2(cmd_buffer,
+ cmd_buffer->state.pipeline, false);
+ }
+ }
assert(cmd_buffer->cs->cdw <= cdw_max);
+ radv_cmd_buffer_after_draw(cmd_buffer, RADV_CMD_FLAG_PS_PARTIAL_FLUSH);
+}
+
+void radv_CmdDraw(
+ VkCommandBuffer commandBuffer,
+ uint32_t vertexCount,
+ uint32_t instanceCount,
+ uint32_t firstVertex,
+ uint32_t firstInstance)
+{
+ RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ struct radv_draw_info info = {};
+
+ info.count = vertexCount;
+ info.instance_count = instanceCount;
+ info.first_instance = firstInstance;
+ info.vertex_offset = firstVertex;
+
+ radv_draw(cmd_buffer, &info);
+}
+
+void radv_CmdDrawIndexed(
+ VkCommandBuffer commandBuffer,
+ uint32_t indexCount,
+ uint32_t instanceCount,
+ uint32_t firstIndex,
+ int32_t vertexOffset,
+ uint32_t firstInstance)
+{
+ RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ struct radv_draw_info info = {};
+
+ info.indexed = true;
+ info.count = indexCount;
+ info.instance_count = instanceCount;
+ info.first_index = firstIndex;
+ info.vertex_offset = vertexOffset;
+ info.first_instance = firstInstance;
+
+ radv_draw(cmd_buffer, &info);
}
void radv_CmdDrawIndirect(
VkCommandBuffer commandBuffer,
- VkBuffer buffer,
+ VkBuffer _buffer,
VkDeviceSize offset,
uint32_t drawCount,
uint32_t stride)
{
- radv_cmd_draw_indirect_count(commandBuffer, buffer, offset,
- VK_NULL_HANDLE, 0, drawCount, stride);
+ RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
+ struct radv_draw_info info = {};
+
+ info.count = drawCount;
+ info.indirect = buffer;
+ info.indirect_offset = offset;
+ info.stride = stride;
+
+ radv_draw(cmd_buffer, &info);
}
void radv_CmdDrawIndexedIndirect(
VkCommandBuffer commandBuffer,
- VkBuffer buffer,
+ VkBuffer _buffer,
VkDeviceSize offset,
uint32_t drawCount,
uint32_t stride)
{
- radv_cmd_draw_indexed_indirect_count(commandBuffer, buffer, offset,
- VK_NULL_HANDLE, 0, drawCount, stride);
+ RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
+ struct radv_draw_info info = {};
+
+ info.indexed = true;
+ info.count = drawCount;
+ info.indirect = buffer;
+ info.indirect_offset = offset;
+ info.stride = stride;
+
+ radv_draw(cmd_buffer, &info);
}
void radv_CmdDrawIndirectCountAMD(
VkCommandBuffer commandBuffer,
- VkBuffer buffer,
+ VkBuffer _buffer,
VkDeviceSize offset,
- VkBuffer countBuffer,
+ VkBuffer _countBuffer,
VkDeviceSize countBufferOffset,
uint32_t maxDrawCount,
uint32_t stride)
{
- radv_cmd_draw_indirect_count(commandBuffer, buffer, offset,
- countBuffer, countBufferOffset,
- maxDrawCount, stride);
+ RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
+ RADV_FROM_HANDLE(radv_buffer, count_buffer, _countBuffer);
+ struct radv_draw_info info = {};
+
+ info.count = maxDrawCount;
+ info.indirect = buffer;
+ info.indirect_offset = offset;
+ info.count_buffer = count_buffer;
+ info.count_buffer_offset = countBufferOffset;
+ info.stride = stride;
+
+ radv_draw(cmd_buffer, &info);
}
void radv_CmdDrawIndexedIndirectCountAMD(
VkCommandBuffer commandBuffer,
- VkBuffer buffer,
+ VkBuffer _buffer,
VkDeviceSize offset,
- VkBuffer countBuffer,
+ VkBuffer _countBuffer,
VkDeviceSize countBufferOffset,
uint32_t maxDrawCount,
uint32_t stride)
{
- radv_cmd_draw_indexed_indirect_count(commandBuffer, buffer, offset,
- countBuffer, countBufferOffset,
- maxDrawCount, stride);
+ RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
+ RADV_FROM_HANDLE(radv_buffer, count_buffer, _countBuffer);
+ struct radv_draw_info info = {};
+
+ info.indexed = true;
+ info.count = maxDrawCount;
+ info.indirect = buffer;
+ info.indirect_offset = offset;
+ info.count_buffer = count_buffer;
+ info.count_buffer_offset = countBufferOffset;
+ info.stride = stride;
+
+ radv_draw(cmd_buffer, &info);
+}
+
+struct radv_dispatch_info {
+ /**
+ * Determine the layout of the grid (in block units) to be used.
+ */
+ uint32_t blocks[3];
+
+ /**
+ * A starting offset for the grid. If unaligned is set, the offset
+ * must still be aligned.
+ */
+ uint32_t offsets[3];
+ /**
+ * Whether it's an unaligned compute dispatch.
+ */
+ bool unaligned;
+
+ /**
+ * Indirect compute parameters resource.
+ */
+ struct radv_buffer *indirect;
+ uint64_t indirect_offset;
+};
+
+static void
+radv_emit_dispatch_packets(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_dispatch_info *info)
+{
+ struct radv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
+ struct radv_shader_variant *compute_shader = pipeline->shaders[MESA_SHADER_COMPUTE];
+ unsigned dispatch_initiator = cmd_buffer->device->dispatch_initiator;
+ struct radeon_winsys *ws = cmd_buffer->device->ws;
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
+ struct radv_userdata_info *loc;
+
+ loc = radv_lookup_user_sgpr(pipeline, MESA_SHADER_COMPUTE,
+ AC_UD_CS_GRID_SIZE);
+
+ MAYBE_UNUSED unsigned cdw_max = radeon_check_space(ws, cs, 25);
+
+ if (info->indirect) {
+ uint64_t va = radv_buffer_get_va(info->indirect->bo);
+
+ va += info->indirect->offset + info->indirect_offset;
+
+ radv_cs_add_buffer(ws, cs, info->indirect->bo, 8);
+
+ if (loc->sgpr_idx != -1) {
+ for (unsigned i = 0; i < 3; ++i) {
+ radeon_emit(cs, PKT3(PKT3_COPY_DATA, 4, 0));
+ radeon_emit(cs, COPY_DATA_SRC_SEL(COPY_DATA_MEM) |
+ COPY_DATA_DST_SEL(COPY_DATA_REG));
+ radeon_emit(cs, (va + 4 * i));
+ radeon_emit(cs, (va + 4 * i) >> 32);
+ radeon_emit(cs, ((R_00B900_COMPUTE_USER_DATA_0
+ + loc->sgpr_idx * 4) >> 2) + i);
+ radeon_emit(cs, 0);
+ }
+ }
+
+ if (radv_cmd_buffer_uses_mec(cmd_buffer)) {
+ radeon_emit(cs, PKT3(PKT3_DISPATCH_INDIRECT, 2, 0) |
+ PKT3_SHADER_TYPE_S(1));
+ radeon_emit(cs, va);
+ radeon_emit(cs, va >> 32);
+ radeon_emit(cs, dispatch_initiator);
+ } else {
+ radeon_emit(cs, PKT3(PKT3_SET_BASE, 2, 0) |
+ PKT3_SHADER_TYPE_S(1));
+ radeon_emit(cs, 1);
+ radeon_emit(cs, va);
+ radeon_emit(cs, va >> 32);
+
+ radeon_emit(cs, PKT3(PKT3_DISPATCH_INDIRECT, 1, 0) |
+ PKT3_SHADER_TYPE_S(1));
+ radeon_emit(cs, 0);
+ radeon_emit(cs, dispatch_initiator);
+ }
+ } else {
+ unsigned blocks[3] = { info->blocks[0], info->blocks[1], info->blocks[2] };
+ unsigned offsets[3] = { info->offsets[0], info->offsets[1], info->offsets[2] };
+
+ if (info->unaligned) {
+ unsigned *cs_block_size = compute_shader->info.cs.block_size;
+ unsigned remainder[3];
+
+ /* If aligned, these should be an entire block size,
+ * not 0.
+ */
+ remainder[0] = blocks[0] + cs_block_size[0] -
+ align_u32_npot(blocks[0], cs_block_size[0]);
+ remainder[1] = blocks[1] + cs_block_size[1] -
+ align_u32_npot(blocks[1], cs_block_size[1]);
+ remainder[2] = blocks[2] + cs_block_size[2] -
+ align_u32_npot(blocks[2], cs_block_size[2]);
+
+ blocks[0] = round_up_u32(blocks[0], cs_block_size[0]);
+ blocks[1] = round_up_u32(blocks[1], cs_block_size[1]);
+ blocks[2] = round_up_u32(blocks[2], cs_block_size[2]);
+
+ for(unsigned i = 0; i < 3; ++i) {
+ assert(offsets[i] % cs_block_size[i] == 0);
+ offsets[i] /= cs_block_size[i];
+ }
+
+ radeon_set_sh_reg_seq(cs, R_00B81C_COMPUTE_NUM_THREAD_X, 3);
+ radeon_emit(cs,
+ S_00B81C_NUM_THREAD_FULL(cs_block_size[0]) |
+ S_00B81C_NUM_THREAD_PARTIAL(remainder[0]));
+ radeon_emit(cs,
+ S_00B81C_NUM_THREAD_FULL(cs_block_size[1]) |
+ S_00B81C_NUM_THREAD_PARTIAL(remainder[1]));
+ radeon_emit(cs,
+ S_00B81C_NUM_THREAD_FULL(cs_block_size[2]) |
+ S_00B81C_NUM_THREAD_PARTIAL(remainder[2]));
+
+ dispatch_initiator |= S_00B800_PARTIAL_TG_EN(1);
+ }
+
+ if (loc->sgpr_idx != -1) {
+ assert(!loc->indirect);
+ assert(loc->num_sgprs == 3);
+
+ radeon_set_sh_reg_seq(cs, R_00B900_COMPUTE_USER_DATA_0 +
+ loc->sgpr_idx * 4, 3);
+ radeon_emit(cs, blocks[0]);
+ radeon_emit(cs, blocks[1]);
+ radeon_emit(cs, blocks[2]);
+ }
+
+ if (offsets[0] || offsets[1] || offsets[2]) {
+ radeon_set_sh_reg_seq(cs, R_00B810_COMPUTE_START_X, 3);
+ radeon_emit(cs, offsets[0]);
+ radeon_emit(cs, offsets[1]);
+ radeon_emit(cs, offsets[2]);
+
+ /* The blocks in the packet are not counts but end values. */
+ for (unsigned i = 0; i < 3; ++i)
+ blocks[i] += offsets[i];
+ } else {
+ dispatch_initiator |= S_00B800_FORCE_START_AT_000(1);
+ }
+
+ radeon_emit(cs, PKT3(PKT3_DISPATCH_DIRECT, 3, 0) |
+ PKT3_SHADER_TYPE_S(1));
+ radeon_emit(cs, blocks[0]);
+ radeon_emit(cs, blocks[1]);
+ radeon_emit(cs, blocks[2]);
+ radeon_emit(cs, dispatch_initiator);
+ }
+
+ assert(cmd_buffer->cs->cdw <= cdw_max);
}
static void
-radv_flush_compute_state(struct radv_cmd_buffer *cmd_buffer)
+radv_upload_compute_shader_descriptors(struct radv_cmd_buffer *cmd_buffer)
{
- radv_emit_compute_pipeline(cmd_buffer);
radv_flush_descriptors(cmd_buffer, VK_SHADER_STAGE_COMPUTE_BIT);
- radv_flush_constants(cmd_buffer, cmd_buffer->state.compute_pipeline,
- VK_SHADER_STAGE_COMPUTE_BIT);
- si_emit_cache_flush(cmd_buffer);
+ radv_flush_constants(cmd_buffer, VK_SHADER_STAGE_COMPUTE_BIT);
}
-void radv_CmdDispatch(
+static void
+radv_dispatch(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_dispatch_info *info)
+{
+ struct radv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
+ bool has_prefetch =
+ cmd_buffer->device->physical_device->rad_info.chip_class >= CIK;
+ bool pipeline_is_dirty = pipeline &&
+ pipeline != cmd_buffer->state.emitted_compute_pipeline;
+
+ if (cmd_buffer->state.flush_bits & (RADV_CMD_FLAG_FLUSH_AND_INV_CB |
+ RADV_CMD_FLAG_FLUSH_AND_INV_DB |
+ RADV_CMD_FLAG_PS_PARTIAL_FLUSH |
+ RADV_CMD_FLAG_CS_PARTIAL_FLUSH)) {
+ /* If we have to wait for idle, set all states first, so that
+ * all SET packets are processed in parallel with previous draw
+ * calls. Then upload descriptors, set shader pointers, and
+ * dispatch, and prefetch at the end. This ensures that the
+ * time the CUs are idle is very short. (there are only SET_SH
+ * packets between the wait and the draw)
+ */
+ radv_emit_compute_pipeline(cmd_buffer);
+ si_emit_cache_flush(cmd_buffer);
+ /* <-- CUs are idle here --> */
+
+ radv_upload_compute_shader_descriptors(cmd_buffer);
+
+ radv_emit_dispatch_packets(cmd_buffer, info);
+ /* <-- CUs are busy here --> */
+
+ /* Start prefetches after the dispatch has been started. Both
+ * will run in parallel, but starting the dispatch first is
+ * more important.
+ */
+ if (has_prefetch && pipeline_is_dirty) {
+ radv_emit_shader_prefetch(cmd_buffer,
+ pipeline->shaders[MESA_SHADER_COMPUTE]);
+ }
+ } else {
+ /* If we don't wait for idle, start prefetches first, then set
+ * states, and dispatch at the end.
+ */
+ si_emit_cache_flush(cmd_buffer);
+
+ if (has_prefetch && pipeline_is_dirty) {
+ radv_emit_shader_prefetch(cmd_buffer,
+ pipeline->shaders[MESA_SHADER_COMPUTE]);
+ }
+
+ radv_upload_compute_shader_descriptors(cmd_buffer);
+
+ radv_emit_compute_pipeline(cmd_buffer);
+ radv_emit_dispatch_packets(cmd_buffer, info);
+ }
+
+ radv_cmd_buffer_after_draw(cmd_buffer, RADV_CMD_FLAG_CS_PARTIAL_FLUSH);
+}
+
+void radv_CmdDispatchBase(
VkCommandBuffer commandBuffer,
+ uint32_t base_x,
+ uint32_t base_y,
+ uint32_t base_z,
uint32_t x,
uint32_t y,
uint32_t z)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ struct radv_dispatch_info info = {};
- radv_flush_compute_state(cmd_buffer);
+ info.blocks[0] = x;
+ info.blocks[1] = y;
+ info.blocks[2] = z;
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 10);
-
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(cmd_buffer->state.compute_pipeline,
- MESA_SHADER_COMPUTE, AC_UD_CS_GRID_SIZE);
- if (loc->sgpr_idx != -1) {
- assert(!loc->indirect);
- uint8_t grid_used = cmd_buffer->state.compute_pipeline->shaders[MESA_SHADER_COMPUTE]->info.info.cs.grid_components_used;
- assert(loc->num_sgprs == grid_used);
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B900_COMPUTE_USER_DATA_0 + loc->sgpr_idx * 4, grid_used);
- radeon_emit(cmd_buffer->cs, x);
- if (grid_used > 1)
- radeon_emit(cmd_buffer->cs, y);
- if (grid_used > 2)
- radeon_emit(cmd_buffer->cs, z);
- }
-
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_DISPATCH_DIRECT, 3, 0) |
- PKT3_SHADER_TYPE_S(1));
- radeon_emit(cmd_buffer->cs, x);
- radeon_emit(cmd_buffer->cs, y);
- radeon_emit(cmd_buffer->cs, z);
- radeon_emit(cmd_buffer->cs, 1);
+ info.offsets[0] = base_x;
+ info.offsets[1] = base_y;
+ info.offsets[2] = base_z;
+ radv_dispatch(cmd_buffer, &info);
+}
- assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_after_draw(cmd_buffer);
+void radv_CmdDispatch(
+ VkCommandBuffer commandBuffer,
+ uint32_t x,
+ uint32_t y,
+ uint32_t z)
+{
+ radv_CmdDispatchBase(commandBuffer, 0, 0, 0, x, y, z);
}
void radv_CmdDispatchIndirect(
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
RADV_FROM_HANDLE(radv_buffer, buffer, _buffer);
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(buffer->bo);
- va += buffer->offset + offset;
-
- cmd_buffer->device->ws->cs_add_buffer(cmd_buffer->cs, buffer->bo, 8);
-
- radv_flush_compute_state(cmd_buffer);
-
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 25);
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(cmd_buffer->state.compute_pipeline,
- MESA_SHADER_COMPUTE, AC_UD_CS_GRID_SIZE);
- if (loc->sgpr_idx != -1) {
- uint8_t grid_used = cmd_buffer->state.compute_pipeline->shaders[MESA_SHADER_COMPUTE]->info.info.cs.grid_components_used;
- for (unsigned i = 0; i < grid_used; ++i) {
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_COPY_DATA, 4, 0));
- radeon_emit(cmd_buffer->cs, COPY_DATA_SRC_SEL(COPY_DATA_MEM) |
- COPY_DATA_DST_SEL(COPY_DATA_REG));
- radeon_emit(cmd_buffer->cs, (va + 4 * i));
- radeon_emit(cmd_buffer->cs, (va + 4 * i) >> 32);
- radeon_emit(cmd_buffer->cs, ((R_00B900_COMPUTE_USER_DATA_0 + loc->sgpr_idx * 4) >> 2) + i);
- radeon_emit(cmd_buffer->cs, 0);
- }
- }
-
- if (radv_cmd_buffer_uses_mec(cmd_buffer)) {
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_DISPATCH_INDIRECT, 2, 0) |
- PKT3_SHADER_TYPE_S(1));
- radeon_emit(cmd_buffer->cs, va);
- radeon_emit(cmd_buffer->cs, va >> 32);
- radeon_emit(cmd_buffer->cs, 1);
- } else {
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_SET_BASE, 2, 0) |
- PKT3_SHADER_TYPE_S(1));
- radeon_emit(cmd_buffer->cs, 1);
- radeon_emit(cmd_buffer->cs, va);
- radeon_emit(cmd_buffer->cs, va >> 32);
+ struct radv_dispatch_info info = {};
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_DISPATCH_INDIRECT, 1, 0) |
- PKT3_SHADER_TYPE_S(1));
- radeon_emit(cmd_buffer->cs, 0);
- radeon_emit(cmd_buffer->cs, 1);
- }
+ info.indirect = buffer;
+ info.indirect_offset = offset;
- assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_after_draw(cmd_buffer);
+ radv_dispatch(cmd_buffer, &info);
}
void radv_unaligned_dispatch(
uint32_t y,
uint32_t z)
{
- struct radv_pipeline *pipeline = cmd_buffer->state.compute_pipeline;
- struct radv_shader_variant *compute_shader = pipeline->shaders[MESA_SHADER_COMPUTE];
- uint32_t blocks[3], remainder[3];
-
- blocks[0] = round_up_u32(x, compute_shader->info.cs.block_size[0]);
- blocks[1] = round_up_u32(y, compute_shader->info.cs.block_size[1]);
- blocks[2] = round_up_u32(z, compute_shader->info.cs.block_size[2]);
-
- /* If aligned, these should be an entire block size, not 0 */
- remainder[0] = x + compute_shader->info.cs.block_size[0] - align_u32_npot(x, compute_shader->info.cs.block_size[0]);
- remainder[1] = y + compute_shader->info.cs.block_size[1] - align_u32_npot(y, compute_shader->info.cs.block_size[1]);
- remainder[2] = z + compute_shader->info.cs.block_size[2] - align_u32_npot(z, compute_shader->info.cs.block_size[2]);
-
- radv_flush_compute_state(cmd_buffer);
+ struct radv_dispatch_info info = {};
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 15);
-
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B81C_COMPUTE_NUM_THREAD_X, 3);
- radeon_emit(cmd_buffer->cs,
- S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[0]) |
- S_00B81C_NUM_THREAD_PARTIAL(remainder[0]));
- radeon_emit(cmd_buffer->cs,
- S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[1]) |
- S_00B81C_NUM_THREAD_PARTIAL(remainder[1]));
- radeon_emit(cmd_buffer->cs,
- S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[2]) |
- S_00B81C_NUM_THREAD_PARTIAL(remainder[2]));
-
- struct ac_userdata_info *loc = radv_lookup_user_sgpr(cmd_buffer->state.compute_pipeline,
- MESA_SHADER_COMPUTE, AC_UD_CS_GRID_SIZE);
- if (loc->sgpr_idx != -1) {
- uint8_t grid_used = cmd_buffer->state.compute_pipeline->shaders[MESA_SHADER_COMPUTE]->info.info.cs.grid_components_used;
- radeon_set_sh_reg_seq(cmd_buffer->cs, R_00B900_COMPUTE_USER_DATA_0 + loc->sgpr_idx * 4, grid_used);
- radeon_emit(cmd_buffer->cs, blocks[0]);
- if (grid_used > 1)
- radeon_emit(cmd_buffer->cs, blocks[1]);
- if (grid_used > 2)
- radeon_emit(cmd_buffer->cs, blocks[2]);
- }
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_DISPATCH_DIRECT, 3, 0) |
- PKT3_SHADER_TYPE_S(1));
- radeon_emit(cmd_buffer->cs, blocks[0]);
- radeon_emit(cmd_buffer->cs, blocks[1]);
- radeon_emit(cmd_buffer->cs, blocks[2]);
- radeon_emit(cmd_buffer->cs, S_00B800_COMPUTE_SHADER_EN(1) |
- S_00B800_PARTIAL_TG_EN(1));
+ info.blocks[0] = x;
+ info.blocks[1] = y;
+ info.blocks[2] = z;
+ info.unaligned = 1;
- assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_after_draw(cmd_buffer);
+ radv_dispatch(cmd_buffer, &info);
}
void radv_CmdEndRenderPass(
/*
* For HTILE we have the following interesting clear words:
- * 0x0000030f: Uncompressed.
+ * 0xfffff30f: Uncompressed, full depth range, for depth+stencil HTILE
+ * 0xfffc000f: Uncompressed, full depth range, for depth only HTILE.
* 0xfffffff0: Clear depth to 1.0
* 0x00000000: Clear depth to 0.0
*/
uint64_t size = image->surface.htile_slice_size * layer_count;
uint64_t offset = image->offset + image->htile_offset +
image->surface.htile_slice_size * range->baseArrayLayer;
+ struct radv_cmd_state *state = &cmd_buffer->state;
- cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_DB |
- RADV_CMD_FLAG_FLUSH_AND_INV_DB_META;
+ state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_DB |
+ RADV_CMD_FLAG_FLUSH_AND_INV_DB_META;
- radv_fill_buffer(cmd_buffer, image->bo, offset, size, clear_word);
+ state->flush_bits |= radv_fill_buffer(cmd_buffer, image->bo, offset,
+ size, clear_word);
- cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_DB_META |
- RADV_CMD_FLAG_CS_PARTIAL_FLUSH |
- RADV_CMD_FLAG_INV_VMEM_L1 |
- RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2;
+ state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_DB_META;
}
static void radv_handle_depth_image_transition(struct radv_cmd_buffer *cmd_buffer,
const VkImageSubresourceRange *range,
VkImageAspectFlags pending_clears)
{
+ if (!radv_image_has_htile(image))
+ return;
+
if (dst_layout == VK_IMAGE_LAYOUT_DEPTH_STENCIL_ATTACHMENT_OPTIMAL &&
(pending_clears & vk_format_aspects(image->vk_format)) == vk_format_aspects(image->vk_format) &&
cmd_buffer->state.render_area.offset.x == 0 && cmd_buffer->state.render_area.offset.y == 0 &&
radv_initialize_htile(cmd_buffer, image, range, 0);
} else if (!radv_layout_is_htile_compressed(image, src_layout, src_queue_mask) &&
radv_layout_is_htile_compressed(image, dst_layout, dst_queue_mask)) {
- radv_initialize_htile(cmd_buffer, image, range, 0xffffffff);
+ uint32_t clear_value = vk_format_is_stencil(image->vk_format) ? 0xfffff30f : 0xfffc000f;
+ radv_initialize_htile(cmd_buffer, image, range, clear_value);
} else if (radv_layout_is_htile_compressed(image, src_layout, src_queue_mask) &&
!radv_layout_is_htile_compressed(image, dst_layout, dst_queue_mask)) {
VkImageSubresourceRange local_range = *range;
}
}
-void radv_initialise_cmask(struct radv_cmd_buffer *cmd_buffer,
- struct radv_image *image, uint32_t value)
+static void radv_initialise_cmask(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image, uint32_t value)
{
- cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
- RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
+ struct radv_cmd_state *state = &cmd_buffer->state;
- radv_fill_buffer(cmd_buffer, image->bo, image->offset + image->cmask.offset,
- image->cmask.size, value);
+ state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
+ RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
- cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB_META |
- RADV_CMD_FLAG_CS_PARTIAL_FLUSH |
- RADV_CMD_FLAG_INV_VMEM_L1 |
- RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2;
-}
+ state->flush_bits |= radv_clear_cmask(cmd_buffer, image, value);
-static void radv_handle_cmask_image_transition(struct radv_cmd_buffer *cmd_buffer,
- struct radv_image *image,
- VkImageLayout src_layout,
- VkImageLayout dst_layout,
- unsigned src_queue_mask,
- unsigned dst_queue_mask,
- const VkImageSubresourceRange *range,
- VkImageAspectFlags pending_clears)
-{
- if (src_layout == VK_IMAGE_LAYOUT_UNDEFINED) {
- if (image->fmask.size)
- radv_initialise_cmask(cmd_buffer, image, 0xccccccccu);
- else
- radv_initialise_cmask(cmd_buffer, image, 0xffffffffu);
- } else if (radv_layout_can_fast_clear(image, src_layout, src_queue_mask) &&
- !radv_layout_can_fast_clear(image, dst_layout, dst_queue_mask)) {
- radv_fast_clear_flush_image_inplace(cmd_buffer, image, range);
- }
+ state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
}
void radv_initialize_dcc(struct radv_cmd_buffer *cmd_buffer,
struct radv_image *image, uint32_t value)
{
+ struct radv_cmd_state *state = &cmd_buffer->state;
+
+ state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
+ RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
+
+ state->flush_bits |= radv_clear_dcc(cmd_buffer, image, value);
- cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
- RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
+ state->flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
+ RADV_CMD_FLAG_FLUSH_AND_INV_CB_META;
+}
+
+/**
+ * Initialize DCC/FMASK/CMASK metadata for a color image.
+ */
+static void radv_init_color_image_metadata(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image,
+ VkImageLayout src_layout,
+ VkImageLayout dst_layout,
+ unsigned src_queue_mask,
+ unsigned dst_queue_mask)
+{
+ if (radv_image_has_cmask(image)) {
+ uint32_t value = 0xffffffffu; /* Fully expanded mode. */
+
+ /* TODO: clarify this. */
+ if (radv_image_has_fmask(image)) {
+ value = 0xccccccccu;
+ }
+
+ radv_initialise_cmask(cmd_buffer, image, value);
+ }
+
+ if (radv_image_has_dcc(image)) {
+ uint32_t value = 0xffffffffu; /* Fully expanded mode. */
- radv_fill_buffer(cmd_buffer, image->bo, image->offset + image->dcc_offset,
- image->surface.dcc_size, value);
+ if (radv_layout_dcc_compressed(image, dst_layout,
+ dst_queue_mask)) {
+ value = 0x20202020u;
+ }
- cmd_buffer->state.flush_bits |= RADV_CMD_FLAG_FLUSH_AND_INV_CB |
- RADV_CMD_FLAG_FLUSH_AND_INV_CB_META |
- RADV_CMD_FLAG_CS_PARTIAL_FLUSH |
- RADV_CMD_FLAG_INV_VMEM_L1 |
- RADV_CMD_FLAG_WRITEBACK_GLOBAL_L2;
+ radv_initialize_dcc(cmd_buffer, image, value);
+ }
}
-static void radv_handle_dcc_image_transition(struct radv_cmd_buffer *cmd_buffer,
- struct radv_image *image,
- VkImageLayout src_layout,
- VkImageLayout dst_layout,
- unsigned src_queue_mask,
- unsigned dst_queue_mask,
- const VkImageSubresourceRange *range,
- VkImageAspectFlags pending_clears)
+/**
+ * Handle color image transitions for DCC/FMASK/CMASK.
+ */
+static void radv_handle_color_image_transition(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_image *image,
+ VkImageLayout src_layout,
+ VkImageLayout dst_layout,
+ unsigned src_queue_mask,
+ unsigned dst_queue_mask,
+ const VkImageSubresourceRange *range)
{
if (src_layout == VK_IMAGE_LAYOUT_UNDEFINED) {
- radv_initialize_dcc(cmd_buffer, image, 0x20202020u);
- } else if (radv_layout_can_fast_clear(image, src_layout, src_queue_mask) &&
- !radv_layout_can_fast_clear(image, dst_layout, dst_queue_mask)) {
- radv_fast_clear_flush_image_inplace(cmd_buffer, image, range);
+ radv_init_color_image_metadata(cmd_buffer, image,
+ src_layout, dst_layout,
+ src_queue_mask, dst_queue_mask);
+ return;
+ }
+
+ if (radv_image_has_dcc(image)) {
+ if (src_layout == VK_IMAGE_LAYOUT_PREINITIALIZED) {
+ radv_initialize_dcc(cmd_buffer, image, 0xffffffffu);
+ } else if (radv_layout_dcc_compressed(image, src_layout, src_queue_mask) &&
+ !radv_layout_dcc_compressed(image, dst_layout, dst_queue_mask)) {
+ radv_decompress_dcc(cmd_buffer, image, range);
+ } else if (radv_layout_can_fast_clear(image, src_layout, src_queue_mask) &&
+ !radv_layout_can_fast_clear(image, dst_layout, dst_queue_mask)) {
+ radv_fast_clear_flush_image_inplace(cmd_buffer, image, range);
+ }
+ } else if (radv_image_has_cmask(image) || radv_image_has_fmask(image)) {
+ if (radv_layout_can_fast_clear(image, src_layout, src_queue_mask) &&
+ !radv_layout_can_fast_clear(image, dst_layout, dst_queue_mask)) {
+ radv_fast_clear_flush_image_inplace(cmd_buffer, image, range);
+ }
}
}
return;
}
- unsigned src_queue_mask = radv_image_queue_family_mask(image, src_family, cmd_buffer->queue_family_index);
- unsigned dst_queue_mask = radv_image_queue_family_mask(image, dst_family, cmd_buffer->queue_family_index);
-
- if (image->surface.htile_size)
- radv_handle_depth_image_transition(cmd_buffer, image, src_layout,
- dst_layout, src_queue_mask,
- dst_queue_mask, range,
- pending_clears);
+ unsigned src_queue_mask =
+ radv_image_queue_family_mask(image, src_family,
+ cmd_buffer->queue_family_index);
+ unsigned dst_queue_mask =
+ radv_image_queue_family_mask(image, dst_family,
+ cmd_buffer->queue_family_index);
- if (image->cmask.size)
- radv_handle_cmask_image_transition(cmd_buffer, image, src_layout,
- dst_layout, src_queue_mask,
- dst_queue_mask, range,
- pending_clears);
-
- if (image->surface.dcc_size)
- radv_handle_dcc_image_transition(cmd_buffer, image, src_layout,
- dst_layout, src_queue_mask,
- dst_queue_mask, range,
- pending_clears);
+ if (vk_format_is_depth(image->vk_format)) {
+ radv_handle_depth_image_transition(cmd_buffer, image,
+ src_layout, dst_layout,
+ src_queue_mask, dst_queue_mask,
+ range, pending_clears);
+ } else {
+ radv_handle_color_image_transition(cmd_buffer, image,
+ src_layout, dst_layout,
+ src_queue_mask, dst_queue_mask,
+ range);
+ }
}
void radv_CmdPipelineBarrier(
unsigned value)
{
struct radeon_winsys_cs *cs = cmd_buffer->cs;
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(event->bo);
+ uint64_t va = radv_buffer_get_va(event->bo);
- cmd_buffer->device->ws->cs_add_buffer(cs, event->bo, 8);
+ radv_cs_add_buffer(cmd_buffer->device->ws, cs, event->bo, 8);
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cs, 18);
si_cs_emit_write_event_eop(cs,
cmd_buffer->state.predicating,
cmd_buffer->device->physical_device->rad_info.chip_class,
- false,
- EVENT_TYPE_BOTTOM_OF_PIPE_TS, 0,
+ radv_cmd_buffer_uses_mec(cmd_buffer),
+ V_028A90_BOTTOM_OF_PIPE_TS, 0,
1, va, 2, value);
assert(cmd_buffer->cs->cdw <= cdw_max);
for (unsigned i = 0; i < eventCount; ++i) {
RADV_FROM_HANDLE(radv_event, event, pEvents[i]);
- uint64_t va = cmd_buffer->device->ws->buffer_get_va(event->bo);
+ uint64_t va = radv_buffer_get_va(event->bo);
- cmd_buffer->device->ws->cs_add_buffer(cs, event->bo, 8);
+ radv_cs_add_buffer(cmd_buffer->device->ws, cs, event->bo, 8);
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cs, 7);
RADV_CMD_FLAG_INV_VMEM_L1 |
RADV_CMD_FLAG_INV_SMEM_L1;
}
+
+
+void radv_CmdSetDeviceMask(VkCommandBuffer commandBuffer,
+ uint32_t deviceMask)
+{
+ /* No-op */
+}