#include "radv_private.h"
#include "radv_radeon_winsys.h"
+#include "radv_shader.h"
#include "radv_cs.h"
#include "sid.h"
#include "gfx9d.h"
#include "vk_format.h"
+#include "radv_debug.h"
#include "radv_meta.h"
#include "ac_debug.h"
},
};
-void
+static void
radv_dynamic_state_copy(struct radv_dynamic_state *dest,
const struct radv_dynamic_state *src,
uint32_t copy_mask)
{
+ /* 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;
+
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);
}
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);
}
VkCommandBuffer* pCommandBuffer)
{
struct radv_cmd_buffer *cmd_buffer;
- VkResult result;
unsigned ring;
cmd_buffer = vk_alloc(&pool->alloc, sizeof(*cmd_buffer), 8,
VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
cmd_buffer->cs = device->ws->cs_create(device->ws, ring);
if (!cmd_buffer->cs) {
- result = VK_ERROR_OUT_OF_HOST_MEMORY;
- goto fail;
+ vk_free(&cmd_buffer->pool->alloc, cmd_buffer);
+ return vk_error(VK_ERROR_OUT_OF_HOST_MEMORY);
}
*pCommandBuffer = radv_cmd_buffer_to_handle(cmd_buffer);
list_inithead(&cmd_buffer->upload.list);
return VK_SUCCESS;
-
-fail:
- vk_free(&cmd_buffer->pool->alloc, cmd_buffer);
-
- return result;
}
static void
vk_free(&cmd_buffer->pool->alloc, cmd_buffer);
}
-static void radv_reset_cmd_buffer(struct radv_cmd_buffer *cmd_buffer)
+static VkResult
+radv_reset_cmd_buffer(struct radv_cmd_buffer *cmd_buffer)
{
cmd_buffer->device->ws->cs_reset(cmd_buffer->cs);
free(up);
}
+ cmd_buffer->push_constant_stages = 0;
cmd_buffer->scratch_size_needed = 0;
cmd_buffer->compute_scratch_size_needed = 0;
cmd_buffer->esgs_ring_size_needed = 0;
&fence_ptr);
cmd_buffer->gfx9_fence_bo = cmd_buffer->upload.upload_bo;
}
+
+ return cmd_buffer->record_result;
}
static bool
return true;
}
+static void
+radv_emit_write_data_packet(struct radeon_winsys_cs *cs, uint64_t va,
+ unsigned count, const uint32_t *data)
+{
+ radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 2 + count, 0));
+ radeon_emit(cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
+ S_370_WR_CONFIRM(1) |
+ S_370_ENGINE_SEL(V_370_ME));
+ radeon_emit(cs, va);
+ radeon_emit(cs, va >> 32);
+ radeon_emit_array(cs, data, count);
+}
+
void radv_cmd_buffer_trace_emit(struct radv_cmd_buffer *cmd_buffer)
{
struct radv_device *device = cmd_buffer->device;
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;
++cmd_buffer->state.trace_id;
device->ws->cs_add_buffer(cs, device->trace_bo, 8);
- radeon_emit(cs, PKT3(PKT3_WRITE_DATA, 3, 0));
- radeon_emit(cs, S_370_DST_SEL(V_370_MEM_ASYNC) |
- S_370_WR_CONFIRM(1) |
- S_370_ENGINE_SEL(V_370_ME));
- radeon_emit(cs, va);
- radeon_emit(cs, va >> 32);
- radeon_emit(cs, cmd_buffer->state.trace_id);
+ 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)
+{
+ if (cmd_buffer->device->instance->debug_flags & RADV_DEBUG_SYNC_SHADERS) {
+ enum radv_cmd_flush_bits flags;
+
+ /* Force wait for graphics/compute engines to be idle. */
+ flags = RADV_CMD_FLAG_PS_PARTIAL_FLUSH |
+ RADV_CMD_FLAG_CS_PARTIAL_FLUSH;
+
+ si_cs_emit_cache_flush(cmd_buffer->cs, false,
+ cmd_buffer->device->physical_device->rad_info.chip_class,
+ NULL, 0,
+ radv_cmd_buffer_uses_mec(cmd_buffer),
+ flags);
+ }
+
+ radv_cmd_buffer_trace_emit(cmd_buffer);
+}
+
+static void
+radv_save_pipeline(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_pipeline *pipeline, enum ring_type ring)
+{
+ struct radv_device *device = cmd_buffer->device;
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
+ uint32_t data[2];
+ uint64_t va;
+
+ if (!device->trace_bo)
+ return;
+
+ va = radv_buffer_get_va(device->trace_bo);
+
+ switch (ring) {
+ case RING_GFX:
+ va += 8;
+ break;
+ case RING_COMPUTE:
+ va += 16;
+ break;
+ default:
+ assert(!"invalid ring type");
+ }
+
+ MAYBE_UNUSED unsigned cdw_max = radeon_check_space(device->ws,
+ cmd_buffer->cs, 6);
+
+ data[0] = (uintptr_t)pipeline;
+ data[1] = (uintptr_t)pipeline >> 32;
+
+ device->ws->cs_add_buffer(cs, device->trace_bo, 8);
+ radv_emit_write_data_packet(cs, va, 2, data);
+}
+
+static void
+radv_save_descriptors(struct radv_cmd_buffer *cmd_buffer)
+{
+ struct radv_device *device = cmd_buffer->device;
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
+ uint32_t data[MAX_SETS * 2] = {};
+ uint64_t va;
+
+ if (!device->trace_bo)
+ return;
+
+ va = radv_buffer_get_va(device->trace_bo) + 24;
+
+ MAYBE_UNUSED unsigned cdw_max = radeon_check_space(device->ws,
+ cmd_buffer->cs, 4 + MAX_SETS * 2);
+
+ for (int i = 0; i < MAX_SETS; i++) {
+ struct radv_descriptor_set *set = cmd_buffer->state.descriptors[i];
+ if (!set)
+ continue;
+
+ data[i * 2] = (uintptr_t)set;
+ data[i * 2 + 1] = (uintptr_t)set >> 32;
+ }
+
+ device->ws->cs_add_buffer(cs, device->trace_bo, 8);
+ radv_emit_write_data_packet(cs, va, MAX_SETS * 2, data);
+}
+
static void
radv_emit_graphics_blend_state(struct radv_cmd_buffer *cmd_buffer,
struct radv_pipeline *pipeline)
radeon_set_context_reg(cmd_buffer->cs, R_028010_DB_RENDER_OVERRIDE2, ds->db_render_override2);
}
-/* 12.4 fixed-point */
-static unsigned radv_pack_float_12p4(float x)
-{
- return x <= 0 ? 0 :
- x >= 4096 ? 0xffff : x * 16;
-}
-
-uint32_t
-radv_shader_stage_to_user_data_0(gl_shader_stage stage, bool has_gs, bool has_tess)
-{
- switch (stage) {
- case MESA_SHADER_FRAGMENT:
- return R_00B030_SPI_SHADER_USER_DATA_PS_0;
- case MESA_SHADER_VERTEX:
- if (has_tess)
- return R_00B530_SPI_SHADER_USER_DATA_LS_0;
- else
- return has_gs ? R_00B330_SPI_SHADER_USER_DATA_ES_0 : R_00B130_SPI_SHADER_USER_DATA_VS_0;
- case MESA_SHADER_GEOMETRY:
- return R_00B230_SPI_SHADER_USER_DATA_GS_0;
- case MESA_SHADER_COMPUTE:
- return R_00B900_COMPUTE_USER_DATA_0;
- case MESA_SHADER_TESS_CTRL:
- return R_00B430_SPI_SHADER_USER_DATA_HS_0;
- case MESA_SHADER_TESS_EVAL:
- if (has_gs)
- return R_00B330_SPI_SHADER_USER_DATA_ES_0;
- else
- return R_00B130_SPI_SHADER_USER_DATA_VS_0;
- default:
- unreachable("unknown shader");
- }
-}
-
struct ac_userdata_info *
radv_lookup_user_sgpr(struct radv_pipeline *pipeline,
gl_shader_stage stage,
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);
+ radeon_set_context_reg(cmd_buffer->cs, R_028804_DB_EQAA, ms->db_eqaa);
+ radeon_set_context_reg(cmd_buffer->cs, R_028A4C_PA_SC_MODE_CNTL_1, ms->pa_sc_mode_cntl_1);
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_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);
}
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;
+ uint64_t va = radv_buffer_get_va(shader->bo) + shader->bo_offset;
unsigned export_count;
ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
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;
+ uint64_t va = radv_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);
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;
+ uint64_t va = radv_buffer_get_va(shader->bo) + shader->bo_offset;
uint32_t rsrc2 = shader->rsrc2;
ws->cs_add_buffer(cmd_buffer->cs, shader->bo, 8);
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;
+ uint64_t va = radv_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);
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;
+ va = radv_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);
assert (pipeline->shaders[MESA_SHADER_FRAGMENT]);
ps = pipeline->shaders[MESA_SHADER_FRAGMENT];
- va = ws->buffer_get_va(ps->bo) + ps->bo_offset;
+ va = radv_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);
}
}
-static void polaris_set_vgt_vertex_reuse(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
+static void
+radv_emit_vgt_vertex_reuse(struct radv_cmd_buffer *cmd_buffer,
+ struct radv_pipeline *pipeline)
{
- uint32_t vtx_reuse_depth = 30;
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
+
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(cs, R_028C58_VGT_VERTEX_REUSE_BLOCK_CNTL,
+ pipeline->graphics.vtx_reuse_depth);
}
static void
-radv_emit_graphics_pipeline(struct radv_cmd_buffer *cmd_buffer,
- struct radv_pipeline *pipeline)
+radv_emit_graphics_pipeline(struct radv_cmd_buffer *cmd_buffer)
{
+ struct radv_pipeline *pipeline = cmd_buffer->state.pipeline;
+
if (!pipeline || cmd_buffer->state.emitted_pipeline == pipeline)
return;
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);
+ radv_emit_vgt_vertex_reuse(cmd_buffer, pipeline);
cmd_buffer->scratch_size_needed =
MAX2(cmd_buffer->scratch_size_needed,
}
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);
+
cmd_buffer->state.emitted_pipeline = pipeline;
}
radv_emit_scissor(struct radv_cmd_buffer *cmd_buffer)
{
uint32_t count = cmd_buffer->state.dynamic.scissor.count;
+
+ if (cmd_buffer->device->physical_device->rad_info.chip_class >= GFX9) {
+ 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.pipeline->graphics.ms.pa_sc_mode_cntl_0 | S_028A48_VPORT_SCISSOR_ENABLE(count ? 1 : 0));
}
+static void
+radv_emit_line_width(struct radv_cmd_buffer *cmd_buffer)
+{
+ unsigned width = cmd_buffer->state.dynamic.line_width * 8;
+
+ radeon_set_context_reg(cmd_buffer->cs, R_028A08_PA_SU_LINE_CNTL,
+ S_028A08_WIDTH(CLAMP(width, 0, 0xFFF)));
+}
+
+static void
+radv_emit_blend_constants(struct radv_cmd_buffer *cmd_buffer)
+{
+ struct radv_dynamic_state *d = &cmd_buffer->state.dynamic;
+
+ radeon_set_context_reg_seq(cmd_buffer->cs, R_028414_CB_BLEND_RED, 4);
+ radeon_emit_array(cmd_buffer->cs, (uint32_t *)d->blend_constants, 4);
+}
+
+static void
+radv_emit_stencil(struct radv_cmd_buffer *cmd_buffer)
+{
+ struct radv_dynamic_state *d = &cmd_buffer->state.dynamic;
+
+ radeon_set_context_reg_seq(cmd_buffer->cs,
+ R_028430_DB_STENCILREFMASK, 2);
+ radeon_emit(cmd_buffer->cs,
+ S_028430_STENCILTESTVAL(d->stencil_reference.front) |
+ S_028430_STENCILMASK(d->stencil_compare_mask.front) |
+ S_028430_STENCILWRITEMASK(d->stencil_write_mask.front) |
+ S_028430_STENCILOPVAL(1));
+ radeon_emit(cmd_buffer->cs,
+ S_028434_STENCILTESTVAL_BF(d->stencil_reference.back) |
+ S_028434_STENCILMASK_BF(d->stencil_compare_mask.back) |
+ S_028434_STENCILWRITEMASK_BF(d->stencil_write_mask.back) |
+ S_028434_STENCILOPVAL_BF(1));
+}
+
+static void
+radv_emit_depth_bounds(struct radv_cmd_buffer *cmd_buffer)
+{
+ struct radv_dynamic_state *d = &cmd_buffer->state.dynamic;
+
+ radeon_set_context_reg(cmd_buffer->cs, R_028020_DB_DEPTH_BOUNDS_MIN,
+ fui(d->depth_bounds.min));
+ radeon_set_context_reg(cmd_buffer->cs, R_028024_DB_DEPTH_BOUNDS_MAX,
+ fui(d->depth_bounds.max));
+}
+
+static void
+radv_emit_depth_biais(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 */
+ }
+}
+
static void
radv_emit_fb_color_state(struct radv_cmd_buffer *cmd_buffer,
int index,
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;
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);
+ uint64_t va = radv_buffer_get_va(image->bo);
va += image->offset + image->clear_value_offset;
if (!image->surface.htile_size)
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)
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)
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)
static void
radv_cmd_buffer_flush_dynamic_state(struct radv_cmd_buffer *cmd_buffer)
{
- struct radv_dynamic_state *d = &cmd_buffer->state.dynamic;
-
if (G_028810_DX_RASTERIZATION_KILL(cmd_buffer->state.pipeline->graphics.raster.pa_cl_clip_cntl))
return;
if (cmd_buffer->state.dirty & (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) {
- unsigned width = cmd_buffer->state.dynamic.line_width * 8;
- radeon_set_context_reg(cmd_buffer->cs, R_028A08_PA_SU_LINE_CNTL,
- S_028A08_WIDTH(CLAMP(width, 0, 0xFFF)));
- }
+ if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_DYNAMIC_LINE_WIDTH)
+ radv_emit_line_width(cmd_buffer);
- if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS) {
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028414_CB_BLEND_RED, 4);
- radeon_emit_array(cmd_buffer->cs, (uint32_t*)d->blend_constants, 4);
- }
+ if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_DYNAMIC_BLEND_CONSTANTS)
+ radv_emit_blend_constants(cmd_buffer);
if (cmd_buffer->state.dirty & (RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE |
RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK |
- RADV_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK)) {
- radeon_set_context_reg_seq(cmd_buffer->cs, R_028430_DB_STENCILREFMASK, 2);
- radeon_emit(cmd_buffer->cs, S_028430_STENCILTESTVAL(d->stencil_reference.front) |
- S_028430_STENCILMASK(d->stencil_compare_mask.front) |
- S_028430_STENCILWRITEMASK(d->stencil_write_mask.front) |
- S_028430_STENCILOPVAL(1));
- radeon_emit(cmd_buffer->cs, S_028434_STENCILTESTVAL_BF(d->stencil_reference.back) |
- S_028434_STENCILMASK_BF(d->stencil_compare_mask.back) |
- S_028434_STENCILWRITEMASK_BF(d->stencil_write_mask.back) |
- S_028434_STENCILOPVAL_BF(1));
- }
+ 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)) {
- radeon_set_context_reg(cmd_buffer->cs, R_028020_DB_DEPTH_BOUNDS_MIN, fui(d->depth_bounds.min));
- radeon_set_context_reg(cmd_buffer->cs, R_028024_DB_DEPTH_BOUNDS_MAX, fui(d->depth_bounds.max));
- }
+ 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)) {
- struct radv_raster_state *raster = &cmd_buffer->state.pipeline->graphics.raster;
- 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 */
- }
- }
+ RADV_CMD_DIRTY_DYNAMIC_DEPTH_BIAS))
+ radv_emit_depth_biais(cmd_buffer);
cmd_buffer->state.dirty = 0;
}
radv_flush_push_descriptors(struct radv_cmd_buffer *cmd_buffer)
{
struct radv_descriptor_set *set = &cmd_buffer->push_descriptors.set;
- uint32_t *ptr = NULL;
unsigned bo_offset;
- if (!radv_cmd_buffer_upload_alloc(cmd_buffer, set->size, 32,
- &bo_offset,
- (void**) &ptr))
+ if (!radv_cmd_buffer_upload_data(cmd_buffer, set->size, 32,
+ set->mapped_ptr,
+ &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;
-
- memcpy(ptr, set->mapped_ptr, set->size);
}
static void
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) {
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;
+ for_each_bit(i, cmd_buffer->state.descriptors_dirty) {
struct radv_descriptor_set *set = cmd_buffer->state.descriptors[i];
if (!set)
continue;
}
cmd_buffer->state.descriptors_dirty = 0;
cmd_buffer->state.push_descriptors_dirty = false;
+
+ radv_save_descriptors(cmd_buffer);
+
assert(cmd_buffer->cs->cdw <= cdw_max);
}
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,
}
}
-static void
+static bool
radv_cmd_buffer_update_vertex_descriptors(struct radv_cmd_buffer *cmd_buffer)
{
struct radv_device *device = cmd_buffer->device;
if ((cmd_buffer->state.pipeline != cmd_buffer->state.emitted_pipeline || cmd_buffer->state.vb_dirty) &&
- cmd_buffer->state.pipeline->num_vertex_attribs &&
+ cmd_buffer->state.pipeline->vertex_elements.count &&
cmd_buffer->state.pipeline->shaders[MESA_SHADER_VERTEX]->info.info.vs.has_vertex_buffers) {
+ struct radv_vertex_elements_info *velems = &cmd_buffer->state.pipeline->vertex_elements;
unsigned vb_offset;
void *vb_ptr;
uint32_t i = 0;
- uint32_t num_attribs = cmd_buffer->state.pipeline->num_vertex_attribs;
+ uint32_t count = velems->count;
uint64_t va;
/* allocate some descriptor state for vertex buffers */
- radv_cmd_buffer_upload_alloc(cmd_buffer, num_attribs * 16, 256,
- &vb_offset, &vb_ptr);
+ if (!radv_cmd_buffer_upload_alloc(cmd_buffer, count * 16, 256,
+ &vb_offset, &vb_ptr))
+ return false;
- for (i = 0; i < num_attribs; i++) {
+ for (i = 0; i < count; i++) {
uint32_t *desc = &((uint32_t *)vb_ptr)[i * 4];
uint32_t offset;
- int vb = cmd_buffer->state.pipeline->va_binding[i];
+ int vb = velems->binding[i];
struct radv_buffer *buffer = cmd_buffer->state.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 + cmd_buffer->state.pipeline->va_offset[i];
+ offset = cmd_buffer->state.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);
if (cmd_buffer->device->physical_device->rad_info.chip_class <= CIK && stride)
- desc[2] = (buffer->size - offset - cmd_buffer->state.pipeline->va_format_size[i]) / stride + 1;
+ desc[2] = (buffer->size - offset - velems->format_size[i]) / stride + 1;
else
desc[2] = buffer->size - offset;
- desc[3] = cmd_buffer->state.pipeline->va_rsrc_word3[i];
+ 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_dirty = false;
+
+ return true;
}
static void
bool indirect_draw,
uint32_t draw_vertex_count)
{
- struct radv_pipeline *pipeline = cmd_buffer->state.pipeline;
uint32_t ia_multi_vgt_param;
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws,
cmd_buffer->cs, 4096);
- radv_cmd_buffer_update_vertex_descriptors(cmd_buffer);
+ 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, pipeline);
+ radv_emit_graphics_pipeline(cmd_buffer);
if (cmd_buffer->state.dirty & RADV_CMD_DIRTY_RENDER_TARGETS)
radv_emit_framebuffer_state(cmd_buffer);
list_del(&cmd_buffer->pool_link);
list_addtail(&cmd_buffer->pool_link, &pool->cmd_buffers);
- radv_reset_cmd_buffer(cmd_buffer);
+ result = radv_reset_cmd_buffer(cmd_buffer);
cmd_buffer->_loader_data.loaderMagic = ICD_LOADER_MAGIC;
cmd_buffer->level = pAllocateInfo->level;
pCommandBuffers[i] = radv_cmd_buffer_to_handle(cmd_buffer);
- result = VK_SUCCESS;
} else {
result = radv_create_cmd_buffer(device, pool, pAllocateInfo->level,
&pCommandBuffers[i]);
VkCommandBufferResetFlags flags)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
- radv_reset_cmd_buffer(cmd_buffer);
- return VK_SUCCESS;
+ return radv_reset_cmd_buffer(cmd_buffer);
}
static void emit_gfx_buffer_state(struct radv_cmd_buffer *cmd_buffer)
{
struct radv_device *device = cmd_buffer->device;
if (device->gfx_init) {
- uint64_t va = device->ws->buffer_get_va(device->gfx_init);
+ uint64_t va = radv_buffer_get_va(device->gfx_init);
device->ws->cs_add_buffer(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);
const VkCommandBufferBeginInfo *pBeginInfo)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
- VkResult result = VK_SUCCESS;
+ VkResult result;
- radv_reset_cmd_buffer(cmd_buffer);
+ 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;
}
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_FROM_HANDLE(radv_buffer, index_buffer, buffer);
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;
if (!set->mapped_ptr) {
cmd_buffer->push_descriptors.capacity = 0;
- cmd_buffer->record_result = VK_ERROR_OUT_OF_DEVICE_MEMORY;
+ cmd_buffer->record_result = VK_ERROR_OUT_OF_HOST_MEMORY;
return false;
}
struct radv_descriptor_set *push_set = &cmd_buffer->meta_push_descriptors;
unsigned bo_offset;
+ assert(set == 0);
assert(layout->set[set].layout->flags & VK_DESCRIPTOR_SET_LAYOUT_CREATE_PUSH_DESCRIPTOR_BIT_KHR);
push_set->size = layout->set[set].layout->size;
(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,
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;
+ va = radv_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);
S_00B81C_NUM_THREAD_FULL(compute_shader->info.cs.block_size[2]));
assert(cmd_buffer->cs->cdw <= cdw_max);
+ radv_save_pipeline(cmd_buffer, pipeline, RING_COMPUTE);
}
static void radv_mark_descriptor_sets_dirty(struct radv_cmd_buffer *cmd_buffer)
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);
+
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);
+
cmd_buffer->state.pipeline = pipeline;
if (!pipeline)
break;
const VkViewport* pViewports)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
-
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));
const VkRect2D* pScissors)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
-
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));
assert(secondary->ring_offsets_idx == primary->ring_offsets_idx);
}
primary->device->ws->cs_execute_secondary(primary->cs, secondary->cs);
+
+ primary->state.emitted_pipeline = secondary->state.emitted_pipeline;
+ primary->state.emitted_compute_pipeline = secondary->state.emitted_compute_pipeline;
+ primary->state.last_primitive_reset_en = secondary->state.last_primitive_reset_en;
+ primary->state.last_primitive_reset_index = secondary->state.last_primitive_reset_index;
}
- /* if we execute secondary we need to re-emit out pipelines */
+ /* if we execute secondary we need to mark some stuff to reset dirty */
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);
}
}
VkCommandPoolResetFlags flags)
{
RADV_FROM_HANDLE(radv_cmd_pool, pool, commandPool);
+ VkResult result;
list_for_each_entry(struct radv_cmd_buffer, cmd_buffer,
&pool->cmd_buffers, pool_link) {
- radv_reset_cmd_buffer(cmd_buffer);
+ result = radv_reset_cmd_buffer(cmd_buffer);
+ if (result != VK_SUCCESS)
+ return result;
}
return VK_SUCCESS;
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);
assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_trace_emit(cmd_buffer);
+ radv_cmd_buffer_after_draw(cmd_buffer);
}
}
assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_trace_emit(cmd_buffer);
+ radv_cmd_buffer_after_draw(cmd_buffer);
}
static void
uint32_t base_reg = cmd_buffer->state.pipeline->graphics.vtx_base_sgpr;
assert(base_reg);
- radeon_emit(cs, PKT3(indexed ? PKT3_DRAW_INDEX_INDIRECT_MULTI :
- PKT3_DRAW_INDIRECT_MULTI,
- 8, false));
- radeon_emit(cs, 0);
- radeon_emit(cs, (base_reg - SI_SH_REG_OFFSET) >> 2);
- radeon_emit(cs, ((base_reg + 4) - SI_SH_REG_OFFSET) >> 2);
- radeon_emit(cs, (((base_reg + 8) - SI_SH_REG_OFFSET) >> 2) |
- S_2C3_DRAW_INDEX_ENABLE(draw_id_enable) |
- S_2C3_COUNT_INDIRECT_ENABLE(!!count_va));
- radeon_emit(cs, draw_count); /* count */
- radeon_emit(cs, count_va); /* count_addr */
- radeon_emit(cs, count_va >> 32);
- radeon_emit(cs, stride); /* stride */
- radeon_emit(cs, di_src_sel);
+ if (draw_count == 1 && !count_va && !draw_id_enable) {
+ radeon_emit(cs, PKT3(indexed ? PKT3_DRAW_INDEX_INDIRECT :
+ PKT3_DRAW_INDIRECT, 3, false));
+ radeon_emit(cs, 0);
+ radeon_emit(cs, (base_reg - SI_SH_REG_OFFSET) >> 2);
+ radeon_emit(cs, ((base_reg + 4) - SI_SH_REG_OFFSET) >> 2);
+ radeon_emit(cs, di_src_sel);
+ } else {
+ radeon_emit(cs, PKT3(indexed ? PKT3_DRAW_INDEX_INDIRECT_MULTI :
+ PKT3_DRAW_INDIRECT_MULTI,
+ 8, false));
+ radeon_emit(cs, 0);
+ radeon_emit(cs, (base_reg - SI_SH_REG_OFFSET) >> 2);
+ radeon_emit(cs, ((base_reg + 4) - SI_SH_REG_OFFSET) >> 2);
+ radeon_emit(cs, (((base_reg + 8) - SI_SH_REG_OFFSET) >> 2) |
+ S_2C3_DRAW_INDEX_ENABLE(draw_id_enable) |
+ S_2C3_COUNT_INDIRECT_ENABLE(!!count_va));
+ radeon_emit(cs, draw_count); /* count */
+ radeon_emit(cs, count_va); /* count_addr */
+ radeon_emit(cs, count_va >> 32);
+ radeon_emit(cs, stride); /* stride */
+ radeon_emit(cs, di_src_sel);
+ }
}
static void
RADV_FROM_HANDLE(radv_buffer, count_buffer, _count_buffer);
struct radeon_winsys_cs *cs = cmd_buffer->cs;
- uint64_t indirect_va = cmd_buffer->device->ws->buffer_get_va(buffer->bo);
+ uint64_t indirect_va = radv_buffer_get_va(buffer->bo);
indirect_va += offset + buffer->offset;
uint64_t count_va = 0;
if (count_buffer) {
- count_va = cmd_buffer->device->ws->buffer_get_va(count_buffer->bo);
+ count_va = radv_buffer_get_va(count_buffer->bo);
count_va += count_offset + count_buffer->offset;
+
+ cmd_buffer->device->ws->cs_add_buffer(cs, count_buffer->bo, 8);
}
if (!draw_count)
radv_cs_emit_indirect_draw_packet(cmd_buffer, indexed, draw_count, count_va, stride);
}
}
- radv_cmd_buffer_trace_emit(cmd_buffer);
+ radv_cmd_buffer_after_draw(cmd_buffer);
}
static void
MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 31 * MAX_VIEWS);
- radeon_emit(cmd_buffer->cs, PKT3(PKT3_INDEX_TYPE, 0, 0));
- radeon_emit(cmd_buffer->cs, cmd_buffer->state.index_type);
+ 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);
+ }
radeon_emit(cmd_buffer->cs, PKT3(PKT3_INDEX_BASE, 1, 0));
radeon_emit(cmd_buffer->cs, index_va);
maxDrawCount, stride);
}
+struct radv_dispatch_info {
+ /**
+ * Determine the layout of the grid (in block units) to be used.
+ */
+ uint32_t blocks[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_flush_compute_state(struct radv_cmd_buffer *cmd_buffer)
+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];
+ struct radeon_winsys *ws = cmd_buffer->device->ws;
+ struct radeon_winsys_cs *cs = cmd_buffer->cs;
+ struct ac_userdata_info *loc;
+ uint8_t grid_used;
+
+ grid_used = compute_shader->info.info.cs.grid_components_used;
+
+ 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;
+
+ ws->cs_add_buffer(cs, info->indirect->bo, 8);
+
+ if (loc->sgpr_idx != -1) {
+ for (unsigned i = 0; i < grid_used; ++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, 1);
+ } 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, 1);
+ }
+ } else {
+ unsigned blocks[3] = { info->blocks[0], info->blocks[1], info->blocks[2] };
+ unsigned dispatch_initiator = S_00B800_COMPUTE_SHADER_EN(1) |
+ S_00B800_FORCE_START_AT_000(1);
+
+ if (cmd_buffer->device->physical_device->rad_info.chip_class >= CIK) {
+ /* If the KMD allows it (there is a KMD hw register for
+ * it), allow launching waves out-of-order.
+ */
+ dispatch_initiator |= S_00B800_ORDER_MODE(1);
+ }
+
+ 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]);
+
+ 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 == grid_used);
+
+ radeon_set_sh_reg_seq(cs, R_00B900_COMPUTE_USER_DATA_0 +
+ loc->sgpr_idx * 4, grid_used);
+ radeon_emit(cs, blocks[0]);
+ if (grid_used > 1)
+ radeon_emit(cs, blocks[1]);
+ if (grid_used > 2)
+ radeon_emit(cs, blocks[2]);
+ }
+
+ 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_dispatch(struct radv_cmd_buffer *cmd_buffer,
+ const struct radv_dispatch_info *info)
{
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_emit_dispatch_packets(cmd_buffer, info);
+
+ radv_cmd_buffer_after_draw(cmd_buffer);
}
void radv_CmdDispatch(
uint32_t z)
{
RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);
+ struct radv_dispatch_info info = {};
- radv_flush_compute_state(cmd_buffer);
-
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 10);
+ info.blocks[0] = x;
+ info.blocks[1] = y;
+ info.blocks[2] = z;
- 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);
-
- assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_trace_emit(cmd_buffer);
+ radv_dispatch(cmd_buffer, &info);
}
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);
- }
- }
+ struct radv_dispatch_info info = {};
- 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);
-
- 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_trace_emit(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);
-
- MAYBE_UNUSED unsigned cdw_max = radeon_check_space(cmd_buffer->device->ws, cmd_buffer->cs, 15);
+ struct radv_dispatch_info info = {};
- 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]));
+ info.blocks[0] = x;
+ info.blocks[1] = y;
+ info.blocks[2] = z;
+ info.unaligned = 1;
- 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));
-
- assert(cmd_buffer->cs->cdw <= cdw_max);
- radv_cmd_buffer_trace_emit(cmd_buffer);
+ radv_dispatch(cmd_buffer, &info);
}
void radv_CmdEndRenderPass(
VkImageLayout dst_layout,
unsigned src_queue_mask,
unsigned dst_queue_mask,
- const VkImageSubresourceRange *range,
- VkImageAspectFlags pending_clears)
+ const VkImageSubresourceRange *range)
{
if (src_layout == VK_IMAGE_LAYOUT_UNDEFINED) {
if (image->fmask.size)
VkImageLayout dst_layout,
unsigned src_queue_mask,
unsigned dst_queue_mask,
- const VkImageSubresourceRange *range,
- VkImageAspectFlags pending_clears)
+ const VkImageSubresourceRange *range)
{
if (src_layout == VK_IMAGE_LAYOUT_UNDEFINED) {
radv_initialize_dcc(cmd_buffer, image, 0x20202020u);
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);
+ dst_queue_mask, range);
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);
+ 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);
cmd_buffer->state.predicating,
cmd_buffer->device->physical_device->rad_info.chip_class,
false,
- EVENT_TYPE_BOTTOM_OF_PIPE_TS, 0,
+ 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);