}
}
+static enum a4xx_index_size
+tu6_index_size(VkIndexType type)
+{
+ switch (type) {
+ case VK_INDEX_TYPE_UINT16:
+ return INDEX4_SIZE_16_BIT;
+ case VK_INDEX_TYPE_UINT32:
+ return INDEX4_SIZE_32_BIT;
+ default:
+ unreachable("invalid VkIndexType");
+ return INDEX4_SIZE_8_BIT;
+ }
+}
+
static void
tu6_emit_marker(struct tu_cmd_buffer *cmd, struct tu_cs *cs)
{
tu_cs_emit_write_reg(cs, cmd->marker_reg, ++cmd->marker_seqno);
}
-static void
+void
tu6_emit_event_write(struct tu_cmd_buffer *cmd,
struct tu_cs *cs,
enum vgt_event_type event,
const struct tu_image_level *slice =
&iview->image->levels[iview->base_mip];
const enum a6xx_tile_mode tile_mode = TILE6_LINEAR;
- const enum a6xx_color_fmt format = RB6_R8G8B8A8_UNORM;
- const enum a3xx_color_swap swap = WZYX;
uint32_t stride = 0;
uint32_t offset = 0;
if (vk_format_is_srgb(iview->vk_format))
srgb_cntl |= (1 << i);
+ const struct tu_native_format *format =
+ tu6_get_native_format(iview->vk_format);
+ assert(format && format->rb >= 0);
+
offset = slice->offset + slice->size * iview->base_layer;
stride = slice->pitch * vk_format_get_blocksize(iview->vk_format);
tu_cs_emit_pkt4(cs, REG_A6XX_RB_MRT_BUF_INFO(i), 6);
- tu_cs_emit(cs, A6XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format) |
+ tu_cs_emit(cs, A6XX_RB_MRT_BUF_INFO_COLOR_FORMAT(format->rb) |
A6XX_RB_MRT_BUF_INFO_COLOR_TILE_MODE(tile_mode) |
- A6XX_RB_MRT_BUF_INFO_COLOR_SWAP(swap));
+ A6XX_RB_MRT_BUF_INFO_COLOR_SWAP(format->swap));
tu_cs_emit(cs, A6XX_RB_MRT_PITCH(stride));
tu_cs_emit(cs, A6XX_RB_MRT_ARRAY_PITCH(slice->size));
tu_cs_emit_qw(cs, iview->image->bo->iova + iview->image->bo_offset +
cs, tiling->gmem_offsets[gmem_index++]); /* RB_MRT[i].BASE_GMEM */
tu_cs_emit_pkt4(cs, REG_A6XX_SP_FS_MRT_REG(i), 1);
- tu_cs_emit(cs, A6XX_SP_FS_MRT_REG_COLOR_FORMAT(RB6_R8G8B8A8_UNORM));
+ tu_cs_emit(cs, A6XX_SP_FS_MRT_REG_COLOR_FORMAT(format->rb));
#if 0
/* when we support UBWC, these would be the system memory
slice->pitch * vk_format_get_blocksize(iview->vk_format);
const enum a6xx_tile_mode tile_mode = TILE6_LINEAR;
const enum a3xx_msaa_samples samples = tu6_msaa_samples(1);
- const enum a6xx_color_fmt format = RB6_R8G8B8A8_UNORM;
- const enum a3xx_color_swap swap = WZYX;
tu_cs_emit_pkt4(cs, REG_A6XX_RB_BLIT_INFO, 1);
tu_cs_emit(cs, blit_info);
/* tile mode? */
+ const struct tu_native_format *format =
+ tu6_get_native_format(iview->vk_format);
+ assert(format && format->rb >= 0);
tu_cs_emit_pkt4(cs, REG_A6XX_RB_BLIT_DST_INFO, 5);
tu_cs_emit(cs, A6XX_RB_BLIT_DST_INFO_TILE_MODE(tile_mode) |
A6XX_RB_BLIT_DST_INFO_SAMPLES(samples) |
- A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(format) |
- A6XX_RB_BLIT_DST_INFO_COLOR_SWAP(swap));
+ A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(format->rb) |
+ A6XX_RB_BLIT_DST_INFO_COLOR_SWAP(format->swap));
tu_cs_emit_qw(cs,
iview->image->bo->iova + iview->image->bo_offset + offset);
tu_cs_emit(cs, A6XX_RB_BLIT_DST_PITCH(stride));
{
const enum a6xx_tile_mode tile_mode = TILE6_LINEAR;
const enum a3xx_msaa_samples samples = tu6_msaa_samples(1);
- const enum a6xx_color_fmt format = RB6_R8G8B8A8_UNORM;
+
+ const struct tu_native_format *format =
+ tu6_get_native_format(iview->vk_format);
+ assert(format && format->rb >= 0);
/* must be WZYX; other values are ignored */
const enum a3xx_color_swap swap = WZYX;
tu_cs_emit_pkt4(cs, REG_A6XX_RB_BLIT_DST_INFO, 1);
tu_cs_emit(cs, A6XX_RB_BLIT_DST_INFO_TILE_MODE(tile_mode) |
A6XX_RB_BLIT_DST_INFO_SAMPLES(samples) |
- A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(format) |
+ A6XX_RB_BLIT_DST_INFO_COLOR_FORMAT(format->rb) |
A6XX_RB_BLIT_DST_INFO_COLOR_SWAP(swap));
tu_cs_emit_pkt4(cs, REG_A6XX_RB_BLIT_INFO, 1);
const struct tu_tile *tile)
{
tu_cs_emit_pkt7(cs, CP_SET_MARKER, 1);
- tu_cs_emit(cs, A2XX_CP_SET_MARKER_0_MODE(0x7));
+ tu_cs_emit(cs, A6XX_CP_SET_MARKER_0_MODE(0x7));
tu6_emit_marker(cmd, cs);
tu_cs_emit_pkt7(cs, CP_SET_MARKER, 1);
- tu_cs_emit(cs, A2XX_CP_SET_MARKER_0_MODE(RM6_GMEM) | 0x10);
+ tu_cs_emit(cs, A6XX_CP_SET_MARKER_0_MODE(RM6_GMEM) | 0x10);
tu6_emit_marker(cmd, cs);
const uint32_t x1 = tile->begin.x;
tu6_emit_marker(cmd, cs);
tu_cs_emit_pkt7(cs, CP_SET_MARKER, 1);
- tu_cs_emit(cs, A2XX_CP_SET_MARKER_0_MODE(RM6_RESOLVE) | 0x10);
+ tu_cs_emit(cs, A6XX_CP_SET_MARKER_0_MODE(RM6_RESOLVE) | 0x10);
tu6_emit_marker(cmd, cs);
tu6_emit_blit_scissor(cmd, cs);
}
}
+static void
+tu6_emit_restart_index(struct tu_cs *cs, uint32_t restart_index)
+{
+ tu_cs_emit_pkt4(cs, REG_A6XX_PC_RESTART_INDEX, 1);
+ tu_cs_emit(cs, restart_index);
+}
+
static void
tu6_init_hw(struct tu_cmd_buffer *cmd, struct tu_cs *cs)
{
tu_cs_emit_write_reg(cs, REG_A6XX_RB_SRGB_CNTL, 0);
tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_8101, 0);
- tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_8109, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_SAMPLE_CNTL, 0);
tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_8110, 0);
tu_cs_emit_write_reg(cs, REG_A6XX_RB_RENDER_CONTROL0, 0x401);
tu_cs_emit_write_reg(cs, REG_A6XX_RB_RENDER_CONTROL1, 0);
tu_cs_emit_write_reg(cs, REG_A6XX_RB_FS_OUTPUT_CNTL0, 0);
- tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8810, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_SAMPLE_CNTL, 0);
tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8818, 0);
tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8819, 0);
tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_881A, 0);
struct tu_cs *cs,
const struct tu_tile *tile)
{
- VkResult result = tu_cs_reserve_space(cmd->device, cs, 64);
+ const uint32_t render_tile_space = 64 + tu_cs_get_call_size(&cmd->draw_cs);
+ VkResult result = tu_cs_reserve_space(cmd->device, cs, render_tile_space);
if (result != VK_SUCCESS) {
cmd->record_result = result;
return;
tu6_emit_tile_select(cmd, cs, tile);
tu_cs_emit_ib(cs, &cmd->state.tile_load_ib);
- /* draw IB? */
-
+ tu_cs_emit_call(cs, &cmd->draw_cs);
cmd->wait_for_idle = true;
tu_cs_emit_ib(cs, &cmd->state.tile_store_ib);
}
/* emit to tile-load sub_cs */
- tu_cs_reserve_space(cmd->device, &sub_cs, tile_load_space);
tu6_emit_tile_load(cmd, &sub_cs);
- tu_cs_sanity_check(&sub_cs);
cmd->state.tile_load_ib = tu_cs_end_sub_stream(&cmd->tile_cs, &sub_cs);
}
/* emit to tile-store sub_cs */
- tu_cs_reserve_space(cmd->device, &sub_cs, tile_store_space);
tu6_emit_tile_store(cmd, &sub_cs);
- tu_cs_sanity_check(&sub_cs);
cmd->state.tile_store_ib = tu_cs_end_sub_stream(&cmd->tile_cs, &sub_cs);
}
tu_bo_list_init(&cmd_buffer->bo_list);
tu_cs_init(&cmd_buffer->cs, TU_CS_MODE_GROW, 4096);
+ tu_cs_init(&cmd_buffer->draw_cs, TU_CS_MODE_GROW, 4096);
tu_cs_init(&cmd_buffer->tile_cs, TU_CS_MODE_SUB_STREAM, 1024);
*pCommandBuffer = tu_cmd_buffer_to_handle(cmd_buffer);
free(cmd_buffer->descriptors[i].push_set.set.mapped_ptr);
tu_cs_finish(cmd_buffer->device, &cmd_buffer->cs);
+ tu_cs_finish(cmd_buffer->device, &cmd_buffer->draw_cs);
tu_cs_finish(cmd_buffer->device, &cmd_buffer->tile_cs);
tu_bo_list_destroy(&cmd_buffer->bo_list);
tu_bo_list_reset(&cmd_buffer->bo_list);
tu_cs_reset(cmd_buffer->device, &cmd_buffer->cs);
+ tu_cs_reset(cmd_buffer->device, &cmd_buffer->draw_cs);
tu_cs_reset(cmd_buffer->device, &cmd_buffer->tile_cs);
for (unsigned i = 0; i < VK_PIPELINE_BIND_POINT_RANGE_SIZE; i++) {
const VkBuffer *pBuffers,
const VkDeviceSize *pOffsets)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+
+ assert(firstBinding + bindingCount <= MAX_VBS);
+
+ for (uint32_t i = 0; i < bindingCount; i++) {
+ cmd->state.vb.buffers[firstBinding + i] =
+ tu_buffer_from_handle(pBuffers[i]);
+ cmd->state.vb.offsets[firstBinding + i] = pOffsets[i];
+ }
+
+ /* VB states depend on VkPipelineVertexInputStateCreateInfo */
+ cmd->state.dirty |= TU_CMD_DIRTY_VERTEX_BUFFERS;
}
void
VkDeviceSize offset,
VkIndexType indexType)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+ TU_FROM_HANDLE(tu_buffer, buf, buffer);
+
+ /* initialize/update the restart index */
+ if (!cmd->state.index_buffer || cmd->state.index_type != indexType) {
+ struct tu_cs *draw_cs = &cmd->draw_cs;
+ VkResult result = tu_cs_reserve_space(cmd->device, draw_cs, 2);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ tu6_emit_restart_index(
+ draw_cs, indexType == VK_INDEX_TYPE_UINT32 ? 0xffffffff : 0xffff);
+
+ tu_cs_sanity_check(draw_cs);
+ }
+
+ /* track the BO */
+ if (cmd->state.index_buffer != buf)
+ tu_bo_list_add(&cmd->bo_list, buf->bo, MSM_SUBMIT_BO_READ);
+
+ cmd->state.index_buffer = buf;
+ cmd->state.index_offset = offset;
+ cmd->state.index_type = indexType;
}
void
MSM_SUBMIT_BO_WRITE);
}
+ for (uint32_t i = 0; i < cmd_buffer->draw_cs.bo_count; i++) {
+ tu_bo_list_add(&cmd_buffer->bo_list, cmd_buffer->draw_cs.bos[i],
+ MSM_SUBMIT_BO_READ | MSM_SUBMIT_BO_DUMP);
+ }
+
for (uint32_t i = 0; i < cmd_buffer->tile_cs.bo_count; i++) {
tu_bo_list_add(&cmd_buffer->bo_list, cmd_buffer->tile_cs.bos[i],
- MSM_SUBMIT_BO_READ);
+ MSM_SUBMIT_BO_READ | MSM_SUBMIT_BO_DUMP);
}
tu_cs_end(&cmd_buffer->cs);
VkPipelineBindPoint pipelineBindPoint,
VkPipeline _pipeline)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+ TU_FROM_HANDLE(tu_pipeline, pipeline, _pipeline);
+
+ switch (pipelineBindPoint) {
+ case VK_PIPELINE_BIND_POINT_GRAPHICS:
+ cmd->state.pipeline = pipeline;
+ cmd->state.dirty |= TU_CMD_DIRTY_PIPELINE;
+ break;
+ case VK_PIPELINE_BIND_POINT_COMPUTE:
+ tu_finishme("binding compute pipeline");
+ break;
+ default:
+ unreachable("unrecognized pipeline bind point");
+ break;
+ }
}
void
uint32_t viewportCount,
const VkViewport *pViewports)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+ struct tu_cs *draw_cs = &cmd->draw_cs;
+
+ VkResult result = tu_cs_reserve_space(cmd->device, draw_cs, 12);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ assert(firstViewport == 0 && viewportCount == 1);
+ tu6_emit_viewport(draw_cs, pViewports);
+
+ tu_cs_sanity_check(draw_cs);
}
void
uint32_t scissorCount,
const VkRect2D *pScissors)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+ struct tu_cs *draw_cs = &cmd->draw_cs;
+
+ VkResult result = tu_cs_reserve_space(cmd->device, draw_cs, 3);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ assert(firstScissor == 0 && scissorCount == 1);
+ tu6_emit_scissor(draw_cs, pScissors);
+
+ tu_cs_sanity_check(draw_cs);
}
void
tu_CmdSetLineWidth(VkCommandBuffer commandBuffer, float lineWidth)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+
+ cmd->state.dynamic.line_width = lineWidth;
+
+ /* line width depends on VkPipelineRasterizationStateCreateInfo */
+ cmd->state.dirty |= TU_CMD_DIRTY_DYNAMIC_LINE_WIDTH;
}
void
float depthBiasClamp,
float depthBiasSlopeFactor)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+ struct tu_cs *draw_cs = &cmd->draw_cs;
+
+ VkResult result = tu_cs_reserve_space(cmd->device, draw_cs, 4);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ tu6_emit_depth_bias(draw_cs, depthBiasConstantFactor, depthBiasClamp,
+ depthBiasSlopeFactor);
+
+ tu_cs_sanity_check(draw_cs);
}
void
tu_CmdSetBlendConstants(VkCommandBuffer commandBuffer,
const float blendConstants[4])
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+ struct tu_cs *draw_cs = &cmd->draw_cs;
+
+ VkResult result = tu_cs_reserve_space(cmd->device, draw_cs, 5);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ tu6_emit_blend_constants(draw_cs, blendConstants);
+
+ tu_cs_sanity_check(draw_cs);
}
void
VkStencilFaceFlags faceMask,
uint32_t compareMask)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+
+ if (faceMask & VK_STENCIL_FACE_FRONT_BIT)
+ cmd->state.dynamic.stencil_compare_mask.front = compareMask;
+ if (faceMask & VK_STENCIL_FACE_BACK_BIT)
+ cmd->state.dynamic.stencil_compare_mask.back = compareMask;
+
+ /* the front/back compare masks must be updated together */
+ cmd->state.dirty |= TU_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK;
}
void
VkStencilFaceFlags faceMask,
uint32_t writeMask)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+
+ if (faceMask & VK_STENCIL_FACE_FRONT_BIT)
+ cmd->state.dynamic.stencil_write_mask.front = writeMask;
+ if (faceMask & VK_STENCIL_FACE_BACK_BIT)
+ cmd->state.dynamic.stencil_write_mask.back = writeMask;
+
+ /* the front/back write masks must be updated together */
+ cmd->state.dirty |= TU_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK;
}
void
VkStencilFaceFlags faceMask,
uint32_t reference)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+
+ if (faceMask & VK_STENCIL_FACE_FRONT_BIT)
+ cmd->state.dynamic.stencil_reference.front = reference;
+ if (faceMask & VK_STENCIL_FACE_BACK_BIT)
+ cmd->state.dynamic.stencil_reference.back = reference;
+
+ /* the front/back references must be updated together */
+ cmd->state.dirty |= TU_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE;
}
void
void
tu_TrimCommandPool(VkDevice device,
VkCommandPool commandPool,
- VkCommandPoolTrimFlagsKHR flags)
+ VkCommandPoolTrimFlags flags)
{
TU_FROM_HANDLE(tu_cmd_pool, pool, commandPool);
tu_cmd_update_tiling_config(cmd_buffer, &pRenderPassBegin->renderArea);
tu_cmd_prepare_tile_load_ib(cmd_buffer);
tu_cmd_prepare_tile_store_ib(cmd_buffer);
+
+ /* draw_cs should contain entries only for this render pass */
+ assert(!cmd_buffer->draw_cs.entry_count);
+ tu_cs_begin(&cmd_buffer->draw_cs);
}
void
uint64_t count_buffer_offset;
};
+enum tu_draw_state_group_id
+{
+ TU_DRAW_STATE_PROGRAM,
+ TU_DRAW_STATE_PROGRAM_BINNING,
+ TU_DRAW_STATE_VI,
+ TU_DRAW_STATE_VI_BINNING,
+ TU_DRAW_STATE_VP,
+ TU_DRAW_STATE_RAST,
+ TU_DRAW_STATE_DS,
+ TU_DRAW_STATE_BLEND,
+
+ TU_DRAW_STATE_COUNT,
+};
+
+struct tu_draw_state_group
+{
+ enum tu_draw_state_group_id id;
+ uint32_t enable_mask;
+ const struct tu_cs_entry *ib;
+};
+
static void
-tu_draw(struct tu_cmd_buffer *cmd_buffer, const struct tu_draw_info *info)
+tu6_bind_draw_states(struct tu_cmd_buffer *cmd,
+ struct tu_cs *cs,
+ const struct tu_draw_info *draw)
{
+ const struct tu_pipeline *pipeline = cmd->state.pipeline;
+ const struct tu_dynamic_state *dynamic = &cmd->state.dynamic;
+ struct tu_draw_state_group draw_state_groups[TU_DRAW_STATE_COUNT];
+ uint32_t draw_state_group_count = 0;
+
+ VkResult result = tu_cs_reserve_space(cmd->device, cs, 256);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ /* TODO lrz */
+
+ uint32_t pc_primitive_cntl = 0;
+ if (pipeline->ia.primitive_restart && draw->indexed)
+ pc_primitive_cntl |= A6XX_PC_PRIMITIVE_CNTL_0_PRIMITIVE_RESTART;
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9806, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9990, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VFD_UNKNOWN_A008, 0);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_PC_PRIMITIVE_CNTL_0, 1);
+ tu_cs_emit(cs, pc_primitive_cntl);
+
+ if (cmd->state.dirty &
+ (TU_CMD_DIRTY_PIPELINE | TU_CMD_DIRTY_DYNAMIC_LINE_WIDTH) &&
+ (pipeline->dynamic_state.mask & TU_DYNAMIC_LINE_WIDTH)) {
+ tu6_emit_gras_su_cntl(cs, pipeline->rast.gras_su_cntl,
+ dynamic->line_width);
+ }
+
+ if ((cmd->state.dirty & TU_CMD_DIRTY_DYNAMIC_STENCIL_COMPARE_MASK) &&
+ (pipeline->dynamic_state.mask & TU_DYNAMIC_STENCIL_COMPARE_MASK)) {
+ tu6_emit_stencil_compare_mask(cs, dynamic->stencil_compare_mask.front,
+ dynamic->stencil_compare_mask.back);
+ }
+
+ if ((cmd->state.dirty & TU_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK) &&
+ (pipeline->dynamic_state.mask & TU_DYNAMIC_STENCIL_WRITE_MASK)) {
+ tu6_emit_stencil_write_mask(cs, dynamic->stencil_write_mask.front,
+ dynamic->stencil_write_mask.back);
+ }
+
+ if ((cmd->state.dirty & TU_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE) &&
+ (pipeline->dynamic_state.mask & TU_DYNAMIC_STENCIL_REFERENCE)) {
+ tu6_emit_stencil_reference(cs, dynamic->stencil_reference.front,
+ dynamic->stencil_reference.back);
+ }
+
+ if (cmd->state.dirty &
+ (TU_CMD_DIRTY_PIPELINE | TU_CMD_DIRTY_VERTEX_BUFFERS)) {
+ for (uint32_t i = 0; i < pipeline->vi.count; i++) {
+ const uint32_t binding = pipeline->vi.bindings[i];
+ const uint32_t stride = pipeline->vi.strides[i];
+ const struct tu_buffer *buf = cmd->state.vb.buffers[binding];
+ const VkDeviceSize offset = buf->bo_offset +
+ cmd->state.vb.offsets[binding] +
+ pipeline->vi.offsets[i];
+ const VkDeviceSize size =
+ offset < buf->bo->size ? buf->bo->size - offset : 0;
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VFD_FETCH(i), 4);
+ tu_cs_emit_qw(cs, buf->bo->iova + offset);
+ tu_cs_emit(cs, size);
+ tu_cs_emit(cs, stride);
+ }
+ }
+
+ /* TODO shader consts */
+
+ if (cmd->state.dirty & TU_CMD_DIRTY_PIPELINE) {
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_PROGRAM,
+ .enable_mask = 0x6,
+ .ib = &pipeline->program.state_ib,
+ };
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_PROGRAM_BINNING,
+ .enable_mask = 0x1,
+ .ib = &pipeline->program.binning_state_ib,
+ };
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_VI,
+ .enable_mask = 0x6,
+ .ib = &pipeline->vi.state_ib,
+ };
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_VI_BINNING,
+ .enable_mask = 0x1,
+ .ib = &pipeline->vi.binning_state_ib,
+ };
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_VP,
+ .enable_mask = 0x7,
+ .ib = &pipeline->vp.state_ib,
+ };
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_RAST,
+ .enable_mask = 0x7,
+ .ib = &pipeline->rast.state_ib,
+ };
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_DS,
+ .enable_mask = 0x7,
+ .ib = &pipeline->ds.state_ib,
+ };
+ draw_state_groups[draw_state_group_count++] =
+ (struct tu_draw_state_group) {
+ .id = TU_DRAW_STATE_BLEND,
+ .enable_mask = 0x7,
+ .ib = &pipeline->blend.state_ib,
+ };
+ }
+
+ tu_cs_emit_pkt7(cs, CP_SET_DRAW_STATE, 3 * draw_state_group_count);
+ for (uint32_t i = 0; i < draw_state_group_count; i++) {
+ const struct tu_draw_state_group *group = &draw_state_groups[i];
+
+ uint32_t cp_set_draw_state =
+ CP_SET_DRAW_STATE__0_COUNT(group->ib->size / 4) |
+ CP_SET_DRAW_STATE__0_ENABLE_MASK(group->enable_mask) |
+ CP_SET_DRAW_STATE__0_GROUP_ID(group->id);
+ uint64_t iova;
+ if (group->ib->size) {
+ iova = group->ib->bo->iova + group->ib->offset;
+ } else {
+ cp_set_draw_state |= CP_SET_DRAW_STATE__0_DISABLE;
+ iova = 0;
+ }
+
+ tu_cs_emit(cs, cp_set_draw_state);
+ tu_cs_emit_qw(cs, iova);
+ }
+
+ tu_cs_sanity_check(cs);
+
+ /* track BOs */
+ if (cmd->state.dirty & TU_CMD_DIRTY_PIPELINE) {
+ tu_bo_list_add(&cmd->bo_list, &pipeline->program.binary_bo,
+ MSM_SUBMIT_BO_READ | MSM_SUBMIT_BO_DUMP);
+ for (uint32_t i = 0; i < pipeline->cs.bo_count; i++) {
+ tu_bo_list_add(&cmd->bo_list, pipeline->cs.bos[i],
+ MSM_SUBMIT_BO_READ | MSM_SUBMIT_BO_DUMP);
+ }
+ }
+ if (cmd->state.dirty & TU_CMD_DIRTY_VERTEX_BUFFERS) {
+ for (uint32_t i = 0; i < MAX_VBS; i++) {
+ const struct tu_buffer *buf = cmd->state.vb.buffers[i];
+ if (buf)
+ tu_bo_list_add(&cmd->bo_list, buf->bo, MSM_SUBMIT_BO_READ);
+ }
+ }
+
+ cmd->state.dirty = 0;
+}
+
+static void
+tu6_emit_draw_direct(struct tu_cmd_buffer *cmd,
+ struct tu_cs *cs,
+ const struct tu_draw_info *draw)
+{
+
+ const enum pc_di_primtype primtype = cmd->state.pipeline->ia.primtype;
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VFD_INDEX_OFFSET, 2);
+ tu_cs_emit(cs, draw->vertex_offset);
+ tu_cs_emit(cs, draw->first_instance);
+
+ /* TODO hw binning */
+ if (draw->indexed) {
+ const enum a4xx_index_size index_size =
+ tu6_index_size(cmd->state.index_type);
+ const uint32_t index_bytes =
+ (cmd->state.index_type == VK_INDEX_TYPE_UINT32) ? 4 : 2;
+ const struct tu_buffer *buf = cmd->state.index_buffer;
+ const VkDeviceSize offset = buf->bo_offset + cmd->state.index_offset +
+ index_bytes * draw->first_index;
+ const uint32_t size = index_bytes * draw->count;
+
+ const uint32_t cp_draw_indx =
+ CP_DRAW_INDX_OFFSET_0_PRIM_TYPE(primtype) |
+ CP_DRAW_INDX_OFFSET_0_SOURCE_SELECT(DI_SRC_SEL_DMA) |
+ CP_DRAW_INDX_OFFSET_0_INDEX_SIZE(index_size) |
+ CP_DRAW_INDX_OFFSET_0_VIS_CULL(IGNORE_VISIBILITY) | 0x2000;
+
+ tu_cs_emit_pkt7(cs, CP_DRAW_INDX_OFFSET, 7);
+ tu_cs_emit(cs, cp_draw_indx);
+ tu_cs_emit(cs, draw->instance_count);
+ tu_cs_emit(cs, draw->count);
+ tu_cs_emit(cs, 0x0); /* XXX */
+ tu_cs_emit_qw(cs, buf->bo->iova + offset);
+ tu_cs_emit(cs, size);
+ } else {
+ const uint32_t cp_draw_indx =
+ CP_DRAW_INDX_OFFSET_0_PRIM_TYPE(primtype) |
+ CP_DRAW_INDX_OFFSET_0_SOURCE_SELECT(DI_SRC_SEL_AUTO_INDEX) |
+ CP_DRAW_INDX_OFFSET_0_VIS_CULL(IGNORE_VISIBILITY) | 0x2000;
+
+ tu_cs_emit_pkt7(cs, CP_DRAW_INDX_OFFSET, 3);
+ tu_cs_emit(cs, cp_draw_indx);
+ tu_cs_emit(cs, draw->instance_count);
+ tu_cs_emit(cs, draw->count);
+ }
+}
+
+static void
+tu_draw(struct tu_cmd_buffer *cmd, const struct tu_draw_info *draw)
+{
+ struct tu_cs *cs = &cmd->draw_cs;
+
+ tu6_bind_draw_states(cmd, cs, draw);
+
+ VkResult result = tu_cs_reserve_space(cmd->device, cs, 32);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ if (draw->indirect) {
+ tu_finishme("indirect draw");
+ return;
+ }
+
+ /* TODO tu6_emit_marker should pick different regs depending on cs */
+ tu6_emit_marker(cmd, cs);
+ tu6_emit_draw_direct(cmd, cs, draw);
+ tu6_emit_marker(cmd, cs);
+
+ cmd->wait_for_idle = true;
+
+ tu_cs_sanity_check(cs);
}
void
{
TU_FROM_HANDLE(tu_cmd_buffer, cmd_buffer, commandBuffer);
+ tu_cs_end(&cmd_buffer->draw_cs);
+
tu_cmd_render_tiles(cmd_buffer);
+ /* discard draw_cs entries now that the tiles are rendered */
+ tu_cs_discard_entries(&cmd_buffer->draw_cs);
+
vk_free(&cmd_buffer->pool->alloc, cmd_buffer->state.attachments);
cmd_buffer->state.attachments = NULL;
projects / mesa.git / blobdiff