Mesa (master): turnip: multiViewport and VK_EXT_shader_viewport_index_layer
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Tue Sep 15 16:32:22 UTC 2020
Module: Mesa
Branch: master
Commit: efff73422049d738b4adfaa34d5fde78a2805d5e
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=efff73422049d738b4adfaa34d5fde78a2805d5e
Author: Jonathan Marek <jonathan at marek.ca>
Date: Tue Jul 14 10:38:09 2020 -0400
turnip: multiViewport and VK_EXT_shader_viewport_index_layer
Passes at least:
dEQP-VK.dynamic_state.vp_state.viewport_array
dEQP-VK.draw.shader_viewport_index.*
dEQP-VK.draw.shader_layer.*
Signed-off-by: Jonathan Marek <jonathan at marek.ca>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5832>
---
src/freedreno/vulkan/tu_cmd_buffer.c | 16 +--
src/freedreno/vulkan/tu_device.c | 6 +-
src/freedreno/vulkan/tu_extensions.py | 1 +
src/freedreno/vulkan/tu_pipeline.c | 193 +++++++++++++++++++---------------
src/freedreno/vulkan/tu_private.h | 14 ++-
src/freedreno/vulkan/tu_shader.c | 1 +
6 files changed, 136 insertions(+), 95 deletions(-)
diff --git a/src/freedreno/vulkan/tu_cmd_buffer.c b/src/freedreno/vulkan/tu_cmd_buffer.c
index f77f849f405..7a67f5de154 100644
--- a/src/freedreno/vulkan/tu_cmd_buffer.c
+++ b/src/freedreno/vulkan/tu_cmd_buffer.c
@@ -2025,11 +2025,13 @@ tu_CmdSetViewport(VkCommandBuffer commandBuffer,
const VkViewport *pViewports)
{
TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
- struct tu_cs cs = tu_cmd_dynamic_state(cmd, VK_DYNAMIC_STATE_VIEWPORT, 18);
+ struct tu_cs cs;
- assert(firstViewport == 0 && viewportCount == 1);
+ memcpy(&cmd->state.viewport[firstViewport], pViewports, viewportCount * sizeof(*pViewports));
+ cmd->state.max_viewport = MAX2(cmd->state.max_viewport, firstViewport + viewportCount);
- tu6_emit_viewport(&cs, pViewports);
+ cs = tu_cmd_dynamic_state(cmd, VK_DYNAMIC_STATE_VIEWPORT, 8 + 10 * cmd->state.max_viewport);
+ tu6_emit_viewport(&cs, cmd->state.viewport, cmd->state.max_viewport);
}
void
@@ -2039,11 +2041,13 @@ tu_CmdSetScissor(VkCommandBuffer commandBuffer,
const VkRect2D *pScissors)
{
TU_FROM_HANDLE(tu_cmd_buffer, cmd, commandBuffer);
- struct tu_cs cs = tu_cmd_dynamic_state(cmd, VK_DYNAMIC_STATE_SCISSOR, 3);
+ struct tu_cs cs;
- assert(firstScissor == 0 && scissorCount == 1);
+ memcpy(&cmd->state.scissor[firstScissor], pScissors, scissorCount * sizeof(*pScissors));
+ cmd->state.max_scissor = MAX2(cmd->state.max_scissor, firstScissor + scissorCount);
- tu6_emit_scissor(&cs, pScissors);
+ cs = tu_cmd_dynamic_state(cmd, VK_DYNAMIC_STATE_SCISSOR, 1 + 2 * cmd->state.max_scissor);
+ tu6_emit_scissor(&cs, cmd->state.scissor, cmd->state.max_scissor);
}
void
diff --git a/src/freedreno/vulkan/tu_device.c b/src/freedreno/vulkan/tu_device.c
index ac5c95cbd48..11390fa2ca2 100644
--- a/src/freedreno/vulkan/tu_device.c
+++ b/src/freedreno/vulkan/tu_device.c
@@ -389,7 +389,7 @@ tu_GetPhysicalDeviceFeatures(VkPhysicalDevice physicalDevice,
.wideLines = false,
.largePoints = true,
.alphaToOne = true,
- .multiViewport = false,
+ .multiViewport = true,
.samplerAnisotropy = true,
.textureCompressionETC2 = true,
.textureCompressionASTC_LDR = true,
@@ -490,8 +490,8 @@ tu_GetPhysicalDeviceFeatures2(VkPhysicalDevice physicalDevice,
features->vulkanMemoryModel = false;
features->vulkanMemoryModelDeviceScope = false;
features->vulkanMemoryModelAvailabilityVisibilityChains = false;
- features->shaderOutputViewportIndex = false;
- features->shaderOutputLayer = false;
+ features->shaderOutputViewportIndex = true;
+ features->shaderOutputLayer = true;
features->subgroupBroadcastDynamicId = false;
break;
}
diff --git a/src/freedreno/vulkan/tu_extensions.py b/src/freedreno/vulkan/tu_extensions.py
index 16722a3c48b..c95167e4cbb 100644
--- a/src/freedreno/vulkan/tu_extensions.py
+++ b/src/freedreno/vulkan/tu_extensions.py
@@ -94,6 +94,7 @@ EXTENSIONS = [
Extension('VK_EXT_custom_border_color', 12, True),
Extension('VK_KHR_multiview', 1, True),
Extension('VK_EXT_host_query_reset', 1, True),
+ Extension('VK_EXT_shader_viewport_index_layer', 1, True),
]
MAX_API_VERSION = VkVersion(MAX_API_VERSION)
diff --git a/src/freedreno/vulkan/tu_pipeline.c b/src/freedreno/vulkan/tu_pipeline.c
index 4b2be417576..f2dbe9701f4 100644
--- a/src/freedreno/vulkan/tu_pipeline.c
+++ b/src/freedreno/vulkan/tu_pipeline.c
@@ -831,16 +831,22 @@ tu6_emit_vpc(struct tu_cs *cs,
ir3_find_output_regid(last_shader, VARYING_SLOT_PSIZ);
const uint32_t layer_regid =
ir3_find_output_regid(last_shader, VARYING_SLOT_LAYER);
+ const uint32_t view_regid =
+ ir3_find_output_regid(last_shader, VARYING_SLOT_VIEWPORT);
uint32_t primitive_regid = gs ?
ir3_find_sysval_regid(gs, SYSTEM_VALUE_PRIMITIVE_ID) : regid(63, 0);
uint32_t flags_regid = gs ?
ir3_find_output_regid(gs, VARYING_SLOT_GS_VERTEX_FLAGS_IR3) : 0;
- uint32_t pointsize_loc = 0xff, position_loc = 0xff, layer_loc = 0xff;
+ uint32_t pointsize_loc = 0xff, position_loc = 0xff, layer_loc = 0xff, view_loc = 0xff;
if (layer_regid != regid(63, 0)) {
layer_loc = linkage.max_loc;
ir3_link_add(&linkage, layer_regid, 0x1, linkage.max_loc);
}
+ if (view_regid != regid(63, 0)) {
+ view_loc = linkage.max_loc;
+ ir3_link_add(&linkage, view_regid, 0x1, linkage.max_loc);
+ }
if (position_regid != regid(63, 0)) {
position_loc = linkage.max_loc;
ir3_link_add(&linkage, position_regid, 0xf, linkage.max_loc);
@@ -895,6 +901,7 @@ tu6_emit_vpc(struct tu_cs *cs,
tu_cs_emit(cs, A6XX_PC_VS_OUT_CNTL_STRIDE_IN_VPC(linkage.max_loc) |
CONDREG(pointsize_regid, A6XX_PC_VS_OUT_CNTL_PSIZE) |
CONDREG(layer_regid, A6XX_PC_VS_OUT_CNTL_LAYER) |
+ CONDREG(view_regid, A6XX_PC_VS_OUT_CNTL_VIEW) |
CONDREG(primitive_regid, A6XX_PC_VS_OUT_CNTL_PRIMITIVE_ID));
tu_cs_emit_pkt4(cs, cfg->reg_sp_xs_primitive_cntl, 1);
@@ -902,10 +909,12 @@ tu6_emit_vpc(struct tu_cs *cs,
A6XX_SP_GS_PRIMITIVE_CNTL_FLAGS_REGID(flags_regid));
tu_cs_emit_pkt4(cs, cfg->reg_vpc_xs_layer_cntl, 1);
- tu_cs_emit(cs, A6XX_VPC_GS_LAYER_CNTL_LAYERLOC(layer_loc) | 0xff00);
+ tu_cs_emit(cs, A6XX_VPC_VS_LAYER_CNTL_LAYERLOC(layer_loc) |
+ A6XX_VPC_VS_LAYER_CNTL_VIEWLOC(view_loc));
tu_cs_emit_pkt4(cs, cfg->reg_gras_xs_layer_cntl, 1);
- tu_cs_emit(cs, CONDREG(layer_regid, A6XX_GRAS_GS_LAYER_CNTL_WRITES_LAYER));
+ tu_cs_emit(cs, CONDREG(layer_regid, A6XX_GRAS_GS_LAYER_CNTL_WRITES_LAYER) |
+ CONDREG(view_regid, A6XX_GRAS_GS_LAYER_CNTL_WRITES_VIEW));
tu_cs_emit_regs(cs, A6XX_PC_PRIMID_PASSTHRU(primid_passthru));
@@ -1547,63 +1556,73 @@ tu6_emit_vertex_input(struct tu_cs *cs,
}
void
-tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewport)
+tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewports, uint32_t num_viewport)
{
- float offsets[3];
- float scales[3];
- scales[0] = viewport->width / 2.0f;
- scales[1] = viewport->height / 2.0f;
- scales[2] = viewport->maxDepth - viewport->minDepth;
- offsets[0] = viewport->x + scales[0];
- offsets[1] = viewport->y + scales[1];
- offsets[2] = viewport->minDepth;
-
- VkOffset2D min;
- VkOffset2D max;
- min.x = (int32_t) viewport->x;
- max.x = (int32_t) ceilf(viewport->x + viewport->width);
- if (viewport->height >= 0.0f) {
- min.y = (int32_t) viewport->y;
- max.y = (int32_t) ceilf(viewport->y + viewport->height);
- } else {
- min.y = (int32_t)(viewport->y + viewport->height);
- max.y = (int32_t) ceilf(viewport->y);
- }
- /* the spec allows viewport->height to be 0.0f */
- if (min.y == max.y)
- max.y++;
- assert(min.x >= 0 && min.x < max.x);
- assert(min.y >= 0 && min.y < max.y);
-
- VkExtent2D guardband_adj;
- guardband_adj.width = fd_calc_guardband(offsets[0], scales[0], false);
- guardband_adj.height = fd_calc_guardband(offsets[1], scales[1], false);
+ VkExtent2D guardband = {511, 511};
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_CL_VPORT_XOFFSET(0), num_viewport * 6);
+ for (uint32_t i = 0; i < num_viewport; i++) {
+ const VkViewport *viewport = &viewports[i];
+ float offsets[3];
+ float scales[3];
+ scales[0] = viewport->width / 2.0f;
+ scales[1] = viewport->height / 2.0f;
+ scales[2] = viewport->maxDepth - viewport->minDepth;
+ offsets[0] = viewport->x + scales[0];
+ offsets[1] = viewport->y + scales[1];
+ offsets[2] = viewport->minDepth;
+ for (uint32_t j = 0; j < 3; j++) {
+ tu_cs_emit(cs, fui(offsets[j]));
+ tu_cs_emit(cs, fui(scales[j]));
+ }
- tu_cs_emit_regs(cs,
- A6XX_GRAS_CL_VPORT_XOFFSET(0, offsets[0]),
- A6XX_GRAS_CL_VPORT_XSCALE(0, scales[0]),
- A6XX_GRAS_CL_VPORT_YOFFSET(0, offsets[1]),
- A6XX_GRAS_CL_VPORT_YSCALE(0, scales[1]),
- A6XX_GRAS_CL_VPORT_ZOFFSET(0, offsets[2]),
- A6XX_GRAS_CL_VPORT_ZSCALE(0, scales[2]));
-
- tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL(0), 2);
- tu_cs_emit(cs, A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL_X(min.x) |
+ guardband.width =
+ MIN2(guardband.width, fd_calc_guardband(offsets[0], scales[0], false));
+ guardband.height =
+ MIN2(guardband.height, fd_calc_guardband(offsets[1], scales[1], false));
+ }
+
+ tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL(0), num_viewport * 2);
+ for (uint32_t i = 0; i < num_viewport; i++) {
+ const VkViewport *viewport = &viewports[i];
+ VkOffset2D min;
+ VkOffset2D max;
+ min.x = (int32_t) viewport->x;
+ max.x = (int32_t) ceilf(viewport->x + viewport->width);
+ if (viewport->height >= 0.0f) {
+ min.y = (int32_t) viewport->y;
+ max.y = (int32_t) ceilf(viewport->y + viewport->height);
+ } else {
+ min.y = (int32_t)(viewport->y + viewport->height);
+ max.y = (int32_t) ceilf(viewport->y);
+ }
+ /* the spec allows viewport->height to be 0.0f */
+ if (min.y == max.y)
+ max.y++;
+ /* allow viewport->width = 0.0f for un-initialized viewports: */
+ if (min.x == max.x)
+ max.x++;
+ assert(min.x >= 0 && min.x < max.x);
+ assert(min.y >= 0 && min.y < max.y);
+ tu_cs_emit(cs, A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL_X(min.x) |
A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL_Y(min.y));
- tu_cs_emit(cs, A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL_X(max.x - 1) |
+ tu_cs_emit(cs, A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL_X(max.x - 1) |
A6XX_GRAS_SC_VIEWPORT_SCISSOR_TL_Y(max.y - 1));
+ }
+ tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_CL_Z_CLAMP(0), num_viewport * 2);
+ for (uint32_t i = 0; i < num_viewport; i++) {
+ const VkViewport *viewport = &viewports[i];
+ tu_cs_emit(cs, fui(MIN2(viewport->minDepth, viewport->maxDepth)));
+ tu_cs_emit(cs, fui(MAX2(viewport->minDepth, viewport->maxDepth)));
+ }
tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_CL_GUARDBAND_CLIP_ADJ, 1);
- tu_cs_emit(cs,
- A6XX_GRAS_CL_GUARDBAND_CLIP_ADJ_HORZ(guardband_adj.width) |
- A6XX_GRAS_CL_GUARDBAND_CLIP_ADJ_VERT(guardband_adj.height));
+ tu_cs_emit(cs, A6XX_GRAS_CL_GUARDBAND_CLIP_ADJ_HORZ(guardband.width) |
+ A6XX_GRAS_CL_GUARDBAND_CLIP_ADJ_VERT(guardband.height));
- float z_clamp_min = MIN2(viewport->minDepth, viewport->maxDepth);
- float z_clamp_max = MAX2(viewport->minDepth, viewport->maxDepth);
-
- tu_cs_emit_regs(cs,
- A6XX_GRAS_CL_Z_CLAMP_MIN(0, z_clamp_min),
- A6XX_GRAS_CL_Z_CLAMP_MAX(0, z_clamp_max));
+ /* TODO: what to do about this and multi viewport ? */
+ float z_clamp_min = num_viewport ? MIN2(viewports[0].minDepth, viewports[0].maxDepth) : 0;
+ float z_clamp_max = num_viewport ? MAX2(viewports[0].minDepth, viewports[0].maxDepth) : 0;
tu_cs_emit_regs(cs,
A6XX_RB_Z_CLAMP_MIN(z_clamp_min),
@@ -1611,32 +1630,35 @@ tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewport)
}
void
-tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scissor)
+tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scissors, uint32_t scissor_count)
{
- VkOffset2D min = scissor->offset;
- VkOffset2D max = {
- scissor->offset.x + scissor->extent.width,
- scissor->offset.y + scissor->extent.height,
- };
+ tu_cs_emit_pkt4(cs, REG_A6XX_GRAS_SC_SCREEN_SCISSOR_TL(0), scissor_count * 2);
- /* special case for empty scissor with max == 0 to avoid overflow */
- if (max.x == 0)
- min.x = max.x = 1;
- if (max.y == 0)
- min.y = max.y = 1;
+ for (uint32_t i = 0; i < scissor_count; i++) {
+ const VkRect2D *scissor = &scissors[i];
- /* avoid overflow with large scissor
- * note the max will be limited to min - 1, so that empty scissor works
- */
- uint32_t scissor_max = BITFIELD_MASK(15);
- min.x = MIN2(scissor_max, min.x);
- min.y = MIN2(scissor_max, min.y);
- max.x = MIN2(scissor_max, max.x);
- max.y = MIN2(scissor_max, max.y);
+ uint32_t min_x = scissor->offset.x;
+ uint32_t min_y = scissor->offset.y;
+ uint32_t max_x = min_x + scissor->extent.width - 1;
+ uint32_t max_y = min_y + scissor->extent.height - 1;
- tu_cs_emit_regs(cs,
- A6XX_GRAS_SC_SCREEN_SCISSOR_TL(0, .x = min.x, .y = min.y),
- A6XX_GRAS_SC_SCREEN_SCISSOR_BR(0, .x = max.x - 1, .y = max.y - 1));
+ if (!scissor->extent.width || !scissor->extent.height) {
+ min_x = min_y = 1;
+ max_x = max_y = 0;
+ } else {
+ /* avoid overflow */
+ uint32_t scissor_max = BITFIELD_MASK(15);
+ min_x = MIN2(scissor_max, min_x);
+ min_y = MIN2(scissor_max, min_y);
+ max_x = MIN2(scissor_max, max_x);
+ max_y = MIN2(scissor_max, max_y);
+ }
+
+ tu_cs_emit(cs, A6XX_GRAS_SC_SCREEN_SCISSOR_TL_X(min_x) |
+ A6XX_GRAS_SC_SCREEN_SCISSOR_TL_Y(min_y));
+ tu_cs_emit(cs, A6XX_GRAS_SC_SCREEN_SCISSOR_BR_X(max_x) |
+ A6XX_GRAS_SC_SCREEN_SCISSOR_BR_Y(max_y));
+ }
}
void
@@ -2034,9 +2056,16 @@ tu_pipeline_builder_compile_shaders(struct tu_pipeline_builder *builder,
builder->shaders[stage] = shader;
}
- struct tu_shader *gs = builder->shaders[MESA_SHADER_GEOMETRY];
- key.layer_zero =
- !gs || !(gs->ir3_shader->nir->info.outputs_written & VARYING_SLOT_LAYER);
+ struct tu_shader *last_shader = builder->shaders[MESA_SHADER_GEOMETRY];
+ if (!last_shader)
+ last_shader = builder->shaders[MESA_SHADER_TESS_EVAL];
+ if (!last_shader)
+ last_shader = builder->shaders[MESA_SHADER_VERTEX];
+
+ uint64_t outputs_written = last_shader->ir3_shader->nir->info.outputs_written;
+
+ key.layer_zero = !(outputs_written & VARYING_BIT_LAYER);
+ key.view_zero = !(outputs_written & VARYING_BIT_VIEWPORT);
pipeline->tess.patch_type = key.tessellation;
@@ -2254,11 +2283,11 @@ tu_pipeline_builder_parse_viewport(struct tu_pipeline_builder *builder,
struct tu_cs cs;
- if (tu_pipeline_static_state(pipeline, &cs, VK_DYNAMIC_STATE_VIEWPORT, 18))
- tu6_emit_viewport(&cs, vp_info->pViewports);
+ if (tu_pipeline_static_state(pipeline, &cs, VK_DYNAMIC_STATE_VIEWPORT, 8 + 10 * vp_info->viewportCount))
+ tu6_emit_viewport(&cs, vp_info->pViewports, vp_info->viewportCount);
- if (tu_pipeline_static_state(pipeline, &cs, VK_DYNAMIC_STATE_SCISSOR, 3))
- tu6_emit_scissor(&cs, vp_info->pScissors);
+ if (tu_pipeline_static_state(pipeline, &cs, VK_DYNAMIC_STATE_SCISSOR, 1 + 2 * vp_info->scissorCount))
+ tu6_emit_scissor(&cs, vp_info->pScissors, vp_info->scissorCount);
}
static void
diff --git a/src/freedreno/vulkan/tu_private.h b/src/freedreno/vulkan/tu_private.h
index 2902050662a..811ade731fd 100644
--- a/src/freedreno/vulkan/tu_private.h
+++ b/src/freedreno/vulkan/tu_private.h
@@ -85,7 +85,7 @@ typedef uint32_t xcb_window_t;
#define MAX_VERTEX_ATTRIBS 32
#define MAX_RTS 8
#define MAX_VSC_PIPES 32
-#define MAX_VIEWPORTS 1
+#define MAX_VIEWPORTS 16
#define MAX_SCISSORS 16
#define MAX_DISCARD_RECTANGLES 4
#define MAX_PUSH_CONSTANTS_SIZE 128
@@ -859,11 +859,17 @@ struct tu_cmd_state
struct tu_pipeline *pipeline;
struct tu_pipeline *compute_pipeline;
- /* Vertex buffers */
+ /* Vertex buffers, viewports, and scissors
+ * the states for these can be updated partially, so we need to save these
+ * to be able to emit a complete draw state
+ */
struct {
uint64_t base;
uint32_t size;
} vb[MAX_VBS];
+ VkViewport viewport[MAX_VIEWPORTS];
+ VkRect2D scissor[MAX_SCISSORS];
+ uint32_t max_viewport, max_scissor;
/* for dynamic states that can't be emitted directly */
uint32_t dynamic_stencil_mask;
@@ -1115,10 +1121,10 @@ struct tu_pipeline
};
void
-tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewport);
+tu6_emit_viewport(struct tu_cs *cs, const VkViewport *viewport, uint32_t num_viewport);
void
-tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scissor);
+tu6_emit_scissor(struct tu_cs *cs, const VkRect2D *scs, uint32_t scissor_count);
void
tu6_emit_sample_locations(struct tu_cs *cs, const VkSampleLocationsInfoEXT *samp_loc);
diff --git a/src/freedreno/vulkan/tu_shader.c b/src/freedreno/vulkan/tu_shader.c
index 188bd723cd0..4348ea1985e 100644
--- a/src/freedreno/vulkan/tu_shader.c
+++ b/src/freedreno/vulkan/tu_shader.c
@@ -68,6 +68,7 @@ tu_spirv_to_nir(struct ir3_compiler *compiler,
.variable_pointers = true,
.stencil_export = true,
.multiview = true,
+ .shader_viewport_index_layer = true,
},
};
const nir_shader_compiler_options *nir_options =
More information about the mesa-commit
mailing list