#include "tu_private.h"
+#include "registers/adreno_pm4.xml.h"
+#include "registers/adreno_common.xml.h"
+#include "registers/a6xx.xml.h"
+
#include "vk_format.h"
-#include "adreno_pm4.xml.h"
+
#include "tu_cs.h"
void
return VK_SUCCESS;
}
+static VkResult
+tu_tiling_config_update_gmem_layout(struct tu_tiling_config *tiling,
+ const struct tu_device *dev)
+{
+ const uint32_t gmem_size = dev->physical_device->gmem_size;
+ uint32_t offset = 0;
+
+ for (uint32_t i = 0; i < tiling->buffer_count; i++) {
+ /* 16KB-aligned */
+ offset = align(offset, 0x4000);
+
+ tiling->gmem_offsets[i] = offset;
+ offset += tiling->tile0.extent.width * tiling->tile0.extent.height *
+ tiling->buffer_cpp[i];
+ }
+
+ return offset <= gmem_size ? VK_SUCCESS : VK_ERROR_OUT_OF_DEVICE_MEMORY;
+}
+
+static void
+tu_tiling_config_update_tile_layout(struct tu_tiling_config *tiling,
+ const struct tu_device *dev)
+{
+ const uint32_t tile_align_w = dev->physical_device->tile_align_w;
+ const uint32_t tile_align_h = dev->physical_device->tile_align_h;
+ const uint32_t max_tile_width = 1024; /* A6xx */
+
+ tiling->tile0.offset = (VkOffset2D) {
+ .x = tiling->render_area.offset.x & ~(tile_align_w - 1),
+ .y = tiling->render_area.offset.y & ~(tile_align_h - 1),
+ };
+
+ const uint32_t ra_width =
+ tiling->render_area.extent.width +
+ (tiling->render_area.offset.x - tiling->tile0.offset.x);
+ const uint32_t ra_height =
+ tiling->render_area.extent.height +
+ (tiling->render_area.offset.y - tiling->tile0.offset.y);
+
+ /* start from 1 tile */
+ tiling->tile_count = (VkExtent2D) {
+ .width = 1,
+ .height = 1,
+ };
+ tiling->tile0.extent = (VkExtent2D) {
+ .width = align(ra_width, tile_align_w),
+ .height = align(ra_height, tile_align_h),
+ };
+
+ /* do not exceed max tile width */
+ while (tiling->tile0.extent.width > max_tile_width) {
+ tiling->tile_count.width++;
+ tiling->tile0.extent.width =
+ align(ra_width / tiling->tile_count.width, tile_align_w);
+ }
+
+ /* do not exceed gmem size */
+ while (tu_tiling_config_update_gmem_layout(tiling, dev) != VK_SUCCESS) {
+ if (tiling->tile0.extent.width > tiling->tile0.extent.height) {
+ tiling->tile_count.width++;
+ tiling->tile0.extent.width =
+ align(ra_width / tiling->tile_count.width, tile_align_w);
+ } else {
+ tiling->tile_count.height++;
+ tiling->tile0.extent.height =
+ align(ra_height / tiling->tile_count.height, tile_align_h);
+ }
+ }
+}
+
+static void
+tu_tiling_config_update_pipe_layout(struct tu_tiling_config *tiling,
+ const struct tu_device *dev)
+{
+ const uint32_t max_pipe_count = 32; /* A6xx */
+
+ /* start from 1 tile per pipe */
+ tiling->pipe0 = (VkExtent2D) {
+ .width = 1,
+ .height = 1,
+ };
+ tiling->pipe_count = tiling->tile_count;
+
+ /* do not exceed max pipe count vertically */
+ while (tiling->pipe_count.height > max_pipe_count) {
+ tiling->pipe0.height += 2;
+ tiling->pipe_count.height =
+ (tiling->tile_count.height + tiling->pipe0.height - 1) /
+ tiling->pipe0.height;
+ }
+
+ /* do not exceed max pipe count */
+ while (tiling->pipe_count.width * tiling->pipe_count.height >
+ max_pipe_count) {
+ tiling->pipe0.width += 1;
+ tiling->pipe_count.width =
+ (tiling->tile_count.width + tiling->pipe0.width - 1) /
+ tiling->pipe0.width;
+ }
+}
+
+static void
+tu_tiling_config_update_pipes(struct tu_tiling_config *tiling,
+ const struct tu_device *dev)
+{
+ const uint32_t max_pipe_count = 32; /* A6xx */
+ const uint32_t used_pipe_count =
+ tiling->pipe_count.width * tiling->pipe_count.height;
+ const VkExtent2D last_pipe = {
+ .width = tiling->tile_count.width % tiling->pipe0.width,
+ .height = tiling->tile_count.height % tiling->pipe0.height,
+ };
+
+ assert(used_pipe_count <= max_pipe_count);
+ assert(max_pipe_count <= ARRAY_SIZE(tiling->pipe_config));
+
+ for (uint32_t y = 0; y < tiling->pipe_count.height; y++) {
+ for (uint32_t x = 0; x < tiling->pipe_count.width; x++) {
+ const uint32_t pipe_x = tiling->pipe0.width * x;
+ const uint32_t pipe_y = tiling->pipe0.height * y;
+ const uint32_t pipe_w = (x == tiling->pipe_count.width - 1)
+ ? last_pipe.width
+ : tiling->pipe0.width;
+ const uint32_t pipe_h = (y == tiling->pipe_count.height - 1)
+ ? last_pipe.height
+ : tiling->pipe0.height;
+ const uint32_t n = tiling->pipe_count.width * y + x;
+
+ tiling->pipe_config[n] = A6XX_VSC_PIPE_CONFIG_REG_X(pipe_x) |
+ A6XX_VSC_PIPE_CONFIG_REG_Y(pipe_y) |
+ A6XX_VSC_PIPE_CONFIG_REG_W(pipe_w) |
+ A6XX_VSC_PIPE_CONFIG_REG_H(pipe_h);
+ tiling->pipe_sizes[n] = CP_SET_BIN_DATA5_0_VSC_SIZE(pipe_w * pipe_h);
+ }
+ }
+
+ memset(tiling->pipe_config + used_pipe_count, 0,
+ sizeof(uint32_t) * (max_pipe_count - used_pipe_count));
+}
+
+static void
+tu_tiling_config_update(struct tu_tiling_config *tiling,
+ const struct tu_device *dev,
+ const uint32_t *buffer_cpp,
+ uint32_t buffer_count,
+ const VkRect2D *render_area)
+{
+ /* see if there is any real change */
+ const bool ra_changed =
+ render_area &&
+ memcmp(&tiling->render_area, render_area, sizeof(*render_area));
+ const bool buf_changed = tiling->buffer_count != buffer_count ||
+ memcmp(tiling->buffer_cpp, buffer_cpp,
+ sizeof(*buffer_cpp) * buffer_count);
+ if (!ra_changed && !buf_changed)
+ return;
+
+ if (ra_changed)
+ tiling->render_area = *render_area;
+
+ if (buf_changed) {
+ memcpy(tiling->buffer_cpp, buffer_cpp,
+ sizeof(*buffer_cpp) * buffer_count);
+ tiling->buffer_count = buffer_count;
+ }
+
+ tu_tiling_config_update_tile_layout(tiling, dev);
+ tu_tiling_config_update_pipe_layout(tiling, dev);
+ tu_tiling_config_update_pipes(tiling, dev);
+}
+
+static void
+tu_tiling_config_get_tile(const struct tu_tiling_config *tiling,
+ const struct tu_device *dev,
+ uint32_t tx,
+ uint32_t ty,
+ struct tu_tile *tile)
+{
+ /* find the pipe and the slot for tile (tx, ty) */
+ const uint32_t px = tx / tiling->pipe0.width;
+ const uint32_t py = ty / tiling->pipe0.height;
+ const uint32_t sx = tx - tiling->pipe0.width * px;
+ const uint32_t sy = ty - tiling->pipe0.height * py;
+
+ assert(tx < tiling->tile_count.width && ty < tiling->tile_count.height);
+ assert(px < tiling->pipe_count.width && py < tiling->pipe_count.height);
+ assert(sx < tiling->pipe0.width && sy < tiling->pipe0.height);
+
+ /* convert to 1D indices */
+ tile->pipe = tiling->pipe_count.width * py + px;
+ tile->slot = tiling->pipe0.width * sy + sx;
+
+ /* get the blit area for the tile */
+ tile->begin = (VkOffset2D) {
+ .x = tiling->tile0.offset.x + tiling->tile0.extent.width * tx,
+ .y = tiling->tile0.offset.y + tiling->tile0.extent.height * ty,
+ };
+ tile->end.x =
+ (tx == tiling->tile_count.width - 1)
+ ? tiling->render_area.offset.x + tiling->render_area.extent.width
+ : tile->begin.x + tiling->tile0.extent.width;
+ tile->end.y =
+ (ty == tiling->tile_count.height - 1)
+ ? tiling->render_area.offset.y + tiling->render_area.extent.height
+ : tile->begin.y + tiling->tile0.extent.height;
+}
+
+static void
+tu6_emit_marker(struct tu_cmd_buffer *cmd)
+{
+ tu_cs_emit_write_reg(cmd->cur_cs, cmd->marker_reg, ++cmd->marker_seqno);
+}
+
+static void
+tu6_emit_event_write(struct tu_cmd_buffer *cmd,
+ enum vgt_event_type event,
+ bool need_seqno)
+{
+ struct tu_cs *cs = cmd->cur_cs;
+
+ tu_cs_emit_pkt7(cs, CP_EVENT_WRITE, need_seqno ? 4 : 1);
+ tu_cs_emit(cs, CP_EVENT_WRITE_0_EVENT(event));
+ if (need_seqno) {
+ tu_cs_emit_qw(cs, cmd->scratch_bo.iova);
+ tu_cs_emit(cs, ++cmd->scratch_seqno);
+ }
+}
+
+static void
+tu6_emit_cache_flush(struct tu_cmd_buffer *cmd)
+{
+ tu6_emit_event_write(cmd, 0x31, false);
+}
+
+static void
+tu6_init_hw(struct tu_cmd_buffer *cmd)
+{
+ struct tu_cs *cs = cmd->cur_cs;
+
+ VkResult result = tu_cs_reserve_space(cmd->device, cs, 256);
+ if (result != VK_SUCCESS) {
+ cmd->record_result = result;
+ return;
+ }
+
+ tu6_emit_cache_flush(cmd);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_HLSQ_UPDATE_CNTL, 0xfffff);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_CCU_CNTL, 0x7c400004);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8E04, 0x00100000);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_AE04, 0x8);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_AE00, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_AE0F, 0x3f);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_B605, 0x44);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_B600, 0x100000);
+ tu_cs_emit_write_reg(cs, REG_A6XX_HLSQ_UNKNOWN_BE00, 0x80);
+ tu_cs_emit_write_reg(cs, REG_A6XX_HLSQ_UNKNOWN_BE01, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9600, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_8600, 0x880);
+ tu_cs_emit_write_reg(cs, REG_A6XX_HLSQ_UNKNOWN_BE04, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_AE03, 0x00000410);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_IBO_COUNT, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_B182, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_HLSQ_UNKNOWN_BB11, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_UCHE_UNKNOWN_0E12, 0x3200000);
+ tu_cs_emit_write_reg(cs, REG_A6XX_UCHE_CLIENT_PF, 4);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8E01, 0x0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_AB00, 0x5);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VFD_UNKNOWN_A009, 0x00000001);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8811, 0x00000010);
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_MODE_CNTL, 0x1f);
+
+ 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_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_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);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_881B, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_881C, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_881D, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_881E, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_88F0, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9101, 0xffff00);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9107, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9236, 1);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9300, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_SO_OVERRIDE,
+ A6XX_VPC_SO_OVERRIDE_SO_DISABLE);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9801, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9806, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9980, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9B06, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9B06, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_A81B, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_UNKNOWN_B183, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_8099, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_809B, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_80A0, 2);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_80AF, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9210, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9211, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9602, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9981, 0x3);
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_UNKNOWN_9E72, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_VPC_UNKNOWN_9108, 0x3);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_TP_UNKNOWN_B304, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_SP_TP_UNKNOWN_B309, 0x000000a2);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8804, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_80A4, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_80A5, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_GRAS_UNKNOWN_80A6, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8805, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8806, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8878, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_RB_UNKNOWN_8879, 0);
+ tu_cs_emit_write_reg(cs, REG_A6XX_HLSQ_CONTROL_5_REG, 0xfc);
+
+ tu6_emit_marker(cmd);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_VFD_MODE_CNTL, 0x00000000);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_VFD_UNKNOWN_A008, 0);
+
+ tu_cs_emit_write_reg(cs, REG_A6XX_PC_MODE_CNTL, 0x0000001f);
+
+ /* we don't use this yet.. probably best to disable.. */
+ tu_cs_emit_pkt7(cs, CP_SET_DRAW_STATE, 3);
+ tu_cs_emit(cs, CP_SET_DRAW_STATE__0_COUNT(0) |
+ CP_SET_DRAW_STATE__0_DISABLE_ALL_GROUPS |
+ CP_SET_DRAW_STATE__0_GROUP_ID(0));
+ tu_cs_emit(cs, CP_SET_DRAW_STATE__1_ADDR_LO(0));
+ tu_cs_emit(cs, CP_SET_DRAW_STATE__2_ADDR_HI(0));
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_BUFFER_BASE_LO(0), 3);
+ tu_cs_emit(cs, 0x00000000); /* VPC_SO_BUFFER_BASE_LO_0 */
+ tu_cs_emit(cs, 0x00000000); /* VPC_SO_BUFFER_BASE_HI_0 */
+ tu_cs_emit(cs, 0x00000000); /* VPC_SO_BUFFER_SIZE_0 */
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_FLUSH_BASE_LO(0), 2);
+ tu_cs_emit(cs, 0x00000000); /* VPC_SO_FLUSH_BASE_LO_0 */
+ tu_cs_emit(cs, 0x00000000); /* VPC_SO_FLUSH_BASE_HI_0 */
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_BUF_CNTL, 1);
+ tu_cs_emit(cs, 0x00000000); /* VPC_SO_BUF_CNTL */
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_BUFFER_OFFSET(0), 1);
+ tu_cs_emit(cs, 0x00000000); /* UNKNOWN_E2AB */
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_BUFFER_BASE_LO(1), 3);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_BUFFER_OFFSET(1), 6);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_BUFFER_OFFSET(2), 6);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_VPC_SO_BUFFER_OFFSET(3), 3);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_SP_HS_CTRL_REG0, 1);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_SP_GS_CTRL_REG0, 1);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_LRZ_CNTL, 1);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_RB_LRZ_CNTL, 1);
+ tu_cs_emit(cs, 0x00000000);
+
+ tu_cs_reserve_space_assert(cs);
+}
+
+static void
+tu_cmd_render_tiles(struct tu_cmd_buffer *cmd)
+{
+ const struct tu_tiling_config *tiling = &cmd->state.tiling_config;
+
+ for (uint32_t y = 0; y < tiling->tile_count.height; y++) {
+ for (uint32_t x = 0; x < tiling->tile_count.width; x++) {
+ struct tu_tile tile;
+ tu_tiling_config_get_tile(tiling, cmd->device, x, y, &tile);
+ /* TODO */
+ }
+ }
+}
+
+static void
+tu_cmd_update_tiling_config(struct tu_cmd_buffer *cmd,
+ const VkRect2D *render_area)
+{
+ const struct tu_device *dev = cmd->device;
+ const struct tu_render_pass *pass = cmd->state.pass;
+ const struct tu_subpass *subpass = cmd->state.subpass;
+ struct tu_tiling_config *tiling = &cmd->state.tiling_config;
+
+ uint32_t buffer_cpp[MAX_RTS + 2];
+ uint32_t buffer_count = 0;
+
+ for (uint32_t i = 0; i < subpass->color_count; ++i) {
+ const uint32_t a = subpass->color_attachments[i].attachment;
+ if (a == VK_ATTACHMENT_UNUSED)
+ continue;
+
+ const struct tu_render_pass_attachment *att = &pass->attachments[a];
+ buffer_cpp[buffer_count++] =
+ vk_format_get_blocksize(att->format) * att->samples;
+ }
+
+ if (subpass->depth_stencil_attachment.attachment != VK_ATTACHMENT_UNUSED) {
+ const uint32_t a = subpass->depth_stencil_attachment.attachment;
+ const struct tu_render_pass_attachment *att = &pass->attachments[a];
+
+ /* TODO */
+ assert(att->format != VK_FORMAT_D32_SFLOAT_S8_UINT);
+
+ buffer_cpp[buffer_count++] =
+ vk_format_get_blocksize(att->format) * att->samples;
+ }
+
+ tu_tiling_config_update(tiling, dev, buffer_cpp, buffer_count,
+ render_area);
+}
+
const struct tu_dynamic_state default_dynamic_state = {
.viewport =
{
list_inithead(&cmd_buffer->upload.list);
+ cmd_buffer->marker_reg = REG_A6XX_CP_SCRATCH_REG(
+ cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY ? 7 : 6);
+
+ VkResult result = tu_bo_init_new(device, &cmd_buffer->scratch_bo, 0x1000);
+ if (result != VK_SUCCESS)
+ return result;
+
return VK_SUCCESS;
}
static void
tu_cmd_buffer_destroy(struct tu_cmd_buffer *cmd_buffer)
{
+ tu_bo_finish(cmd_buffer->device, &cmd_buffer->scratch_bo);
+
list_del(&cmd_buffer->pool_link);
for (unsigned i = 0; i < VK_PIPELINE_BIND_POINT_RANGE_SIZE; i++)
return cmd_buffer->record_result;
}
+static VkResult
+tu_cmd_state_setup_attachments(struct tu_cmd_buffer *cmd_buffer,
+ struct tu_render_pass *pass,
+ const VkRenderPassBeginInfo *info)
+{
+ struct tu_cmd_state *state = &cmd_buffer->state;
+
+ if (pass->attachment_count == 0) {
+ state->attachments = NULL;
+ return VK_SUCCESS;
+ }
+
+ state->attachments =
+ vk_alloc(&cmd_buffer->pool->alloc,
+ pass->attachment_count * sizeof(state->attachments[0]), 8,
+ VK_SYSTEM_ALLOCATION_SCOPE_OBJECT);
+ if (state->attachments == NULL) {
+ cmd_buffer->record_result = VK_ERROR_OUT_OF_HOST_MEMORY;
+ return cmd_buffer->record_result;
+ }
+
+ for (uint32_t i = 0; i < pass->attachment_count; ++i) {
+ const struct tu_render_pass_attachment *att = &pass->attachments[i];
+ VkImageAspectFlags att_aspects = vk_format_aspects(att->format);
+ VkImageAspectFlags clear_aspects = 0;
+
+ if (att_aspects == VK_IMAGE_ASPECT_COLOR_BIT) {
+ /* color attachment */
+ if (att->load_op == VK_ATTACHMENT_LOAD_OP_CLEAR) {
+ clear_aspects |= VK_IMAGE_ASPECT_COLOR_BIT;
+ }
+ } else {
+ /* depthstencil attachment */
+ if ((att_aspects & VK_IMAGE_ASPECT_DEPTH_BIT) &&
+ att->load_op == VK_ATTACHMENT_LOAD_OP_CLEAR) {
+ clear_aspects |= VK_IMAGE_ASPECT_DEPTH_BIT;
+ if ((att_aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
+ att->stencil_load_op == VK_ATTACHMENT_LOAD_OP_DONT_CARE)
+ clear_aspects |= VK_IMAGE_ASPECT_STENCIL_BIT;
+ }
+ if ((att_aspects & VK_IMAGE_ASPECT_STENCIL_BIT) &&
+ att->stencil_load_op == VK_ATTACHMENT_LOAD_OP_CLEAR) {
+ clear_aspects |= VK_IMAGE_ASPECT_STENCIL_BIT;
+ }
+ }
+
+ state->attachments[i].pending_clear_aspects = clear_aspects;
+ state->attachments[i].cleared_views = 0;
+ if (clear_aspects && info) {
+ assert(info->clearValueCount > i);
+ state->attachments[i].clear_value = info->pClearValues[i];
+ }
+
+ state->attachments[i].current_layout = att->initial_layout;
+ }
+
+ return VK_SUCCESS;
+}
+
VkResult
tu_AllocateCommandBuffers(VkDevice _device,
const VkCommandBufferAllocateInfo *pAllocateInfo,
memset(&cmd_buffer->state, 0, sizeof(cmd_buffer->state));
cmd_buffer->usage_flags = pBeginInfo->flags;
+ result = tu_cs_begin(cmd_buffer->device, &cmd_buffer->cs, 4096);
+ if (result != VK_SUCCESS)
+ return result;
+
+ cmd_buffer->marker_seqno = 0;
+ cmd_buffer->scratch_seqno = 0;
+
+ cmd_buffer->cur_cs = &cmd_buffer->cs;
+
/* setup initial configuration into command buffer */
if (cmd_buffer->level == VK_COMMAND_BUFFER_LEVEL_PRIMARY) {
switch (cmd_buffer->queue_family_index) {
case TU_QUEUE_GENERAL:
- /* init */
+ tu6_init_hw(cmd_buffer);
break;
default:
break;
}
}
- result = tu_cs_begin(cmd_buffer->device, &cmd_buffer->cs, 4096);
- if (result != VK_SUCCESS)
- return result;
-
cmd_buffer->status = TU_CMD_BUFFER_STATUS_RECORDING;
- /* Put some stuff in so we do not have empty command buffers. */
- tu_cs_emit_pkt7(&cmd_buffer->cs, CP_NOP, 4);
- tu_cs_emit(&cmd_buffer->cs, 0);
- tu_cs_emit(&cmd_buffer->cs, 0);
- tu_cs_emit(&cmd_buffer->cs, 0);
- tu_cs_emit(&cmd_buffer->cs, 0);
-
return VK_SUCCESS;
}
{
TU_FROM_HANDLE(tu_cmd_buffer, cmd_buffer, commandBuffer);
+ if (cmd_buffer->scratch_seqno) {
+ tu_bo_list_add(&cmd_buffer->bo_list, &cmd_buffer->scratch_bo,
+ MSM_SUBMIT_BO_WRITE);
+ }
+
VkResult result = tu_cs_end(&cmd_buffer->cs);
if (result != VK_SUCCESS)
cmd_buffer->record_result = result;
+ assert(!cmd_buffer->state.attachments);
+
cmd_buffer->status = TU_CMD_BUFFER_STATUS_EXECUTABLE;
return cmd_buffer->record_result;
const VkRenderPassBeginInfo *pRenderPassBegin,
VkSubpassContents contents)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd_buffer, commandBuffer);
+ TU_FROM_HANDLE(tu_render_pass, pass, pRenderPassBegin->renderPass);
+ TU_FROM_HANDLE(tu_framebuffer, framebuffer, pRenderPassBegin->framebuffer);
+ VkResult result;
+
+ cmd_buffer->state.pass = pass;
+ cmd_buffer->state.subpass = pass->subpasses;
+ cmd_buffer->state.framebuffer = framebuffer;
+
+ result =
+ tu_cmd_state_setup_attachments(cmd_buffer, pass, pRenderPassBegin);
+ if (result != VK_SUCCESS)
+ return;
+
+ tu_cmd_update_tiling_config(cmd_buffer, &pRenderPassBegin->renderArea);
}
void
void
tu_CmdNextSubpass(VkCommandBuffer commandBuffer, VkSubpassContents contents)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
+
+ tu_cmd_render_tiles(cmd);
+
+ cmd->state.subpass++;
+ tu_cmd_update_tiling_config(cmd, NULL);
}
void
void
tu_CmdEndRenderPass(VkCommandBuffer commandBuffer)
{
+ TU_FROM_HANDLE(tu_cmd_buffer, cmd_buffer, commandBuffer);
+
+ tu_cmd_render_tiles(cmd_buffer);
+
+ vk_free(&cmd_buffer->pool->alloc, cmd_buffer->state.attachments);
+ cmd_buffer->state.attachments = NULL;
+
+ cmd_buffer->state.pass = NULL;
+ cmd_buffer->state.subpass = NULL;
+ cmd_buffer->state.framebuffer = NULL;
}
void