<div dir="ltr"><div class="gmail_quote"><div class="gmail_attr">Gfx10 remembers sample positions in compressed Z/S memory, so the hw doesn't need the decompress pass for shader loads.</div><div class="gmail_attr"><br></div><div class="gmail_attr">Marek<br></div><div dir="ltr" class="gmail_attr"><br></div><div dir="ltr" class="gmail_attr">On Wed, May 22, 2019 at 4:20 PM Marek Olšák <<a href="mailto:maraeo@gmail.com">maraeo@gmail.com</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex"><div dir="ltr"><div class="gmail_quote"><div class="gmail_attr">The depth decompress pass needs to know the sample locations.<br></div><div class="gmail_attr"><br></div><div class="gmail_attr">If shader loads read from compressed depth, the texture hardware will always use the standard locations for decompression.</div><div class="gmail_attr"><br></div><div class="gmail_attr">Marek<br></div><div dir="ltr" class="gmail_attr"><br></div><div dir="ltr" class="gmail_attr">On Tue, May 21, 2019 at 8:17 PM Bas Nieuwenhuizen <<a href="mailto:bas@basnieuwenhuizen.nl" target="_blank">bas@basnieuwenhuizen.nl</a>> wrote:<br></div><blockquote class="gmail_quote" style="margin:0px 0px 0px 0.8ex;border-left:1px solid rgb(204,204,204);padding-left:1ex">So this does not seem to use the sample locations during layout transitions?<br>
<br>
AFAIK those are needed for e.g. HTILE decompression as it is based on<br>
equations somehow.<br>
<br>
On Thu, May 16, 2019 at 11:51 AM Samuel Pitoiset<br>
<<a href="mailto:samuel.pitoiset@gmail.com" target="_blank">samuel.pitoiset@gmail.com</a>> wrote:<br>
><br>
> Basically, this extension allows applications to use custom<br>
> sample locations. It doesn't support variable sample locations<br>
> during subpass. Note that we don't have to upload the user<br>
> sample locations because the spec doesn't allow this.<br>
><br>
> Only enabled on VI+ because it's untested on older chips.<br>
><br>
> v2: - change sampleLocationCoordinateRange[1] to 0.9375<br>
> - compute and emit PA_SC_CENTROID_PRIORITY_{0,1}<br>
> - rebased on top of master<br>
> - some cleanups<br>
><br>
> Signed-off-by: Samuel Pitoiset <<a href="mailto:samuel.pitoiset@gmail.com" target="_blank">samuel.pitoiset@gmail.com</a>><br>
> ---<br>
> src/amd/vulkan/radv_cmd_buffer.c | 223 ++++++++++++++++++++++++++++++<br>
> src/amd/vulkan/radv_device.c | 27 ++++<br>
> src/amd/vulkan/radv_extensions.py | 1 +<br>
> src/amd/vulkan/radv_pipeline.c | 30 ++++<br>
> src/amd/vulkan/radv_private.h | 26 +++-<br>
> 5 files changed, 300 insertions(+), 7 deletions(-)<br>
><br>
> diff --git a/src/amd/vulkan/radv_cmd_buffer.c b/src/amd/vulkan/radv_cmd_buffer.c<br>
> index 4f592bc7f68..fb79c1c6713 100644<br>
> --- a/src/amd/vulkan/radv_cmd_buffer.c<br>
> +++ b/src/amd/vulkan/radv_cmd_buffer.c<br>
> @@ -105,6 +105,7 @@ radv_bind_dynamic_state(struct radv_cmd_buffer *cmd_buffer,<br>
> dest->viewport.count = src->viewport.count;<br>
> dest->scissor.count = src->scissor.count;<br>
> dest->discard_rectangle.count = src->discard_rectangle.count;<br>
> + dest->sample_location.count = src->sample_location.count;<br>
><br>
> if (copy_mask & RADV_DYNAMIC_VIEWPORT) {<br>
> if (memcmp(&dest->viewport.viewports, &src->viewport.viewports,<br>
> @@ -192,6 +193,22 @@ radv_bind_dynamic_state(struct radv_cmd_buffer *cmd_buffer,<br>
> }<br>
> }<br>
><br>
> + if (copy_mask & RADV_DYNAMIC_SAMPLE_LOCATIONS) {<br>
> + if (dest->sample_location.per_pixel != src->sample_location.per_pixel ||<br>
> + dest->sample_location.grid_size.width != src->sample_location.grid_size.width ||<br>
> + dest->sample_location.grid_size.height != src->sample_location.grid_size.height ||<br>
> + memcmp(&dest->sample_location.locations,<br>
> + &src->sample_location.locations,<br>
> + src->sample_location.count * sizeof(VkSampleLocationEXT))) {<br>
> + dest->sample_location.per_pixel = src->sample_location.per_pixel;<br>
> + dest->sample_location.grid_size = src->sample_location.grid_size;<br>
> + typed_memcpy(dest->sample_location.locations,<br>
> + src->sample_location.locations,<br>
> + src->sample_location.count);<br>
> + dest_mask |= RADV_DYNAMIC_SAMPLE_LOCATIONS;<br>
> + }<br>
> + }<br>
> +<br>
> cmd_buffer->state.dirty |= dest_mask;<br>
> }<br>
><br>
> @@ -632,6 +649,190 @@ radv_emit_descriptor_pointers(struct radv_cmd_buffer *cmd_buffer,<br>
> }<br>
> }<br>
><br>
> +/**<br>
> + * Convert the user sample locations to hardware sample locations (the values<br>
> + * that will be emitted by PA_SC_AA_SAMPLE_LOCS_PIXEL_*).<br>
> + */<br>
> +static void<br>
> +radv_convert_user_sample_locs(struct radv_sample_locations_state *state,<br>
> + uint32_t x, uint32_t y, VkOffset2D *sample_locs)<br>
> +{<br>
> + uint32_t x_offset = x % state->grid_size.width;<br>
> + uint32_t y_offset = y % state->grid_size.height;<br>
> + uint32_t num_samples = (uint32_t)state->per_pixel;<br>
> + VkSampleLocationEXT *user_locs;<br>
> + uint32_t pixel_offset;<br>
> +<br>
> + pixel_offset = (x_offset + y_offset * state->grid_size.width) * num_samples;<br>
> +<br>
> + assert(pixel_offset <= MAX_SAMPLE_LOCATIONS);<br>
> + user_locs = &state->locations[pixel_offset];<br>
> +<br>
> + for (uint32_t i = 0; i < num_samples; i++) {<br>
> + float shifted_pos_x = user_locs[i].x - 0.5;<br>
> + float shifted_pos_y = user_locs[i].y - 0.5;<br>
> +<br>
> + int32_t scaled_pos_x = floor(shifted_pos_x * 16);<br>
> + int32_t scaled_pos_y = floor(shifted_pos_y * 16);<br>
> +<br>
> + sample_locs[i].x = CLAMP(scaled_pos_x, -8, 7);<br>
> + sample_locs[i].y = CLAMP(scaled_pos_y, -8, 7);<br>
> + }<br>
> +}<br>
> +<br>
> +/**<br>
> + * Compute the PA_SC_AA_SAMPLE_LOCS_PIXEL_* mask based on hardware sample<br>
> + * locations.<br>
> + */<br>
> +static void<br>
> +radv_compute_sample_locs_pixel(uint32_t num_samples, VkOffset2D *sample_locs,<br>
> + uint32_t *sample_locs_pixel)<br>
> +{<br>
> + for (uint32_t i = 0; i < num_samples; i++) {<br>
> + uint32_t sample_reg_idx = i / 4;<br>
> + uint32_t sample_loc_idx = i % 4;<br>
> + int32_t pos_x = sample_locs[i].x;<br>
> + int32_t pos_y = sample_locs[i].y;<br>
> +<br>
> + uint32_t shift_x = 8 * sample_loc_idx;<br>
> + uint32_t shift_y = shift_x + 4;<br>
> +<br>
> + sample_locs_pixel[sample_reg_idx] |= (pos_x & 0xf) << shift_x;<br>
> + sample_locs_pixel[sample_reg_idx] |= (pos_y & 0xf) << shift_y;<br>
> + }<br>
> +}<br>
> +<br>
> +/**<br>
> + * Compute the PA_SC_CENTROID_PRIORITY_* mask based on the top left hardware<br>
> + * sample locations.<br>
> + */<br>
> +static uint64_t<br>
> +radv_compute_centroid_priority(struct radv_cmd_buffer *cmd_buffer,<br>
> + VkOffset2D *sample_locs,<br>
> + uint32_t num_samples)<br>
> +{<br>
> + uint32_t centroid_priorities[num_samples];<br>
> + uint32_t sample_mask = num_samples - 1;<br>
> + uint32_t distances[num_samples];<br>
> + uint64_t centroid_priority = 0;<br>
> +<br>
> + /* Compute the distances from center for each sample. */<br>
> + for (int i = 0; i < num_samples; i++) {<br>
> + distances[i] = (sample_locs[i].x * sample_locs[i].x) +<br>
> + (sample_locs[i].y * sample_locs[i].y);<br>
> + }<br>
> +<br>
> + /* Compute the centroid priorities by looking at the distances array. */<br>
> + for (int i = 0; i < num_samples; i++) {<br>
> + uint32_t min_idx = 0;<br>
> +<br>
> + for (int j = 1; j < num_samples; j++) {<br>
> + if (distances[j] < distances[min_idx])<br>
> + min_idx = j;<br>
> + }<br>
> +<br>
> + centroid_priorities[i] = min_idx;<br>
> + distances[min_idx] = 0xffffffff;<br>
> + }<br>
> +<br>
> + /* Compute the final centroid priority. */<br>
> + for (int i = 0; i < 8; i++) {<br>
> + centroid_priority |=<br>
> + centroid_priorities[i & sample_mask] << (i * 4);<br>
> + }<br>
> +<br>
> + return centroid_priority << 32 | centroid_priority;<br>
> +}<br>
> +<br>
> +/**<br>
> + * Emit the sample locations that are specified with VK_EXT_sample_locations.<br>
> + */<br>
> +static void<br>
> +radv_emit_sample_locations(struct radv_cmd_buffer *cmd_buffer)<br>
> +{<br>
> + struct radv_pipeline *pipeline = cmd_buffer->state.pipeline;<br>
> + struct radv_multisample_state *ms = &pipeline-><a href="http://graphics.ms" rel="noreferrer" target="_blank">graphics.ms</a>;<br>
> + struct radv_sample_locations_state *sample_location =<br>
> + &cmd_buffer->state.dynamic.sample_location;<br>
> + uint32_t num_samples = (uint32_t)sample_location->per_pixel;<br>
> + struct radeon_cmdbuf *cs = cmd_buffer->cs;<br>
> + uint32_t sample_locs_pixel[4][2] = {};<br>
> + VkOffset2D sample_locs[4][8]; /* 8 is the max. sample count supported */<br>
> + uint32_t max_sample_dist = 0;<br>
> + uint64_t centroid_priority;<br>
> +<br>
> + if (!cmd_buffer->state.dynamic.sample_location.count)<br>
> + return;<br>
> +<br>
> + /* Convert the user sample locations to hardware sample locations. */<br>
> + radv_convert_user_sample_locs(sample_location, 0, 0, sample_locs[0]);<br>
> + radv_convert_user_sample_locs(sample_location, 1, 0, sample_locs[1]);<br>
> + radv_convert_user_sample_locs(sample_location, 0, 1, sample_locs[2]);<br>
> + radv_convert_user_sample_locs(sample_location, 1, 1, sample_locs[3]);<br>
> +<br>
> + /* Compute the PA_SC_AA_SAMPLE_LOCS_PIXEL_* mask. */<br>
> + for (uint32_t i = 0; i < 4; i++) {<br>
> + radv_compute_sample_locs_pixel(num_samples, sample_locs[i],<br>
> + sample_locs_pixel[i]);<br>
> + }<br>
> +<br>
> + /* Compute the PA_SC_CENTROID_PRIORITY_* mask. */<br>
> + centroid_priority =<br>
> + radv_compute_centroid_priority(cmd_buffer, sample_locs[0],<br>
> + num_samples);<br>
> +<br>
> + /* Compute the maximum sample distance from the specified locations. */<br>
> + for (uint32_t i = 0; i < num_samples; i++) {<br>
> + VkOffset2D offset = sample_locs[0][i];<br>
> + max_sample_dist = MAX2(max_sample_dist,<br>
> + MAX2(abs(offset.x), abs(offset.y)));<br>
> + }<br>
> +<br>
> + /* Emit the specified user sample locations. */<br>
> + switch (num_samples) {<br>
> + case 2:<br>
> + case 4:<br>
> + radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_pixel[0][0]);<br>
> + radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_pixel[1][0]);<br>
> + radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_pixel[2][0]);<br>
> + radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_pixel[3][0]);<br>
> + break;<br>
> + case 8:<br>
> + radeon_set_context_reg(cs, R_028BF8_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0, sample_locs_pixel[0][0]);<br>
> + radeon_set_context_reg(cs, R_028C08_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_0, sample_locs_pixel[1][0]);<br>
> + radeon_set_context_reg(cs, R_028C18_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_0, sample_locs_pixel[2][0]);<br>
> + radeon_set_context_reg(cs, R_028C28_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_0, sample_locs_pixel[3][0]);<br>
> + radeon_set_context_reg(cs, R_028BFC_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_1, sample_locs_pixel[0][1]);<br>
> + radeon_set_context_reg(cs, R_028C0C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y0_1, sample_locs_pixel[1][1]);<br>
> + radeon_set_context_reg(cs, R_028C1C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y1_1, sample_locs_pixel[2][1]);<br>
> + radeon_set_context_reg(cs, R_028C2C_PA_SC_AA_SAMPLE_LOCS_PIXEL_X1Y1_1, sample_locs_pixel[3][1]);<br>
> + break;<br>
> + default:<br>
> + unreachable("invalid number of samples");<br>
> + }<br>
> +<br>
> + /* Emit the maximum sample distance and the centroid priority. */<br>
> + uint32_t pa_sc_aa_config = ms->pa_sc_aa_config;<br>
> +<br>
> + pa_sc_aa_config &= C_028BE0_MAX_SAMPLE_DIST;<br>
> + pa_sc_aa_config |= S_028BE0_MAX_SAMPLE_DIST(max_sample_dist);<br>
> +<br>
> + radeon_set_context_reg_seq(cs, R_028BE0_PA_SC_AA_CONFIG, 1);<br>
> + radeon_emit(cs, pa_sc_aa_config);<br>
> +<br>
> + radeon_set_context_reg_seq(cs, R_028BD4_PA_SC_CENTROID_PRIORITY_0, 2);<br>
> + radeon_emit(cs, centroid_priority);<br>
> + radeon_emit(cs, centroid_priority >> 32);<br>
> +<br>
> + /* GFX9: Flush DFSM when the AA mode changes. */<br>
> + if (cmd_buffer->device->dfsm_allowed) {<br>
> + radeon_emit(cs, PKT3(PKT3_EVENT_WRITE, 0, 0));<br>
> + radeon_emit(cs, EVENT_TYPE(V_028A90_FLUSH_DFSM) | EVENT_INDEX(0));<br>
> + }<br>
> +<br>
> + cmd_buffer->state.context_roll_without_scissor_emitted = true;<br>
> +}<br>
> +<br>
> static void<br>
> radv_emit_inline_push_consts(struct radv_cmd_buffer *cmd_buffer,<br>
> struct radv_pipeline *pipeline,<br>
> @@ -1775,6 +1976,9 @@ radv_cmd_buffer_flush_dynamic_state(struct radv_cmd_buffer *cmd_buffer)<br>
> if (states & RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE)<br>
> radv_emit_discard_rectangle(cmd_buffer);<br>
><br>
> + if (states & RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS)<br>
> + radv_emit_sample_locations(cmd_buffer);<br>
> +<br>
> cmd_buffer->state.dirty &= ~states;<br>
> }<br>
><br>
> @@ -3266,6 +3470,25 @@ void radv_CmdSetDiscardRectangleEXT(<br>
> state->dirty |= RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE;<br>
> }<br>
><br>
> +void radv_CmdSetSampleLocationsEXT(<br>
> + VkCommandBuffer commandBuffer,<br>
> + const VkSampleLocationsInfoEXT* pSampleLocationsInfo)<br>
> +{<br>
> + RADV_FROM_HANDLE(radv_cmd_buffer, cmd_buffer, commandBuffer);<br>
> + struct radv_cmd_state *state = &cmd_buffer->state;<br>
> +<br>
> + assert(pSampleLocationsInfo->sampleLocationsCount <= MAX_SAMPLE_LOCATIONS);<br>
> +<br>
> + state->dynamic.sample_location.per_pixel = pSampleLocationsInfo->sampleLocationsPerPixel;<br>
> + state->dynamic.sample_location.grid_size = pSampleLocationsInfo->sampleLocationGridSize;<br>
> + state->dynamic.sample_location.count = pSampleLocationsInfo->sampleLocationsCount;<br>
> + typed_memcpy(&state->dynamic.sample_location.locations[0],<br>
> + pSampleLocationsInfo->pSampleLocations,<br>
> + pSampleLocationsInfo->sampleLocationsCount);<br>
> +<br>
> + state->dirty |= RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS;<br>
> +}<br>
> +<br>
> void radv_CmdExecuteCommands(<br>
> VkCommandBuffer commandBuffer,<br>
> uint32_t commandBufferCount,<br>
> diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c<br>
> index c0e317a97e5..e523bba1223 100644<br>
> --- a/src/amd/vulkan/radv_device.c<br>
> +++ b/src/amd/vulkan/radv_device.c<br>
> @@ -1360,6 +1360,19 @@ void radv_GetPhysicalDeviceProperties2(<br>
> props->maxDescriptorSetUpdateAfterBindInlineUniformBlocks = MAX_INLINE_UNIFORM_BLOCK_COUNT;<br>
> break;<br>
> }<br>
> + case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_SAMPLE_LOCATIONS_PROPERTIES_EXT: {<br>
> + VkPhysicalDeviceSampleLocationsPropertiesEXT *properties =<br>
> + (VkPhysicalDeviceSampleLocationsPropertiesEXT *)ext;<br>
> + properties->sampleLocationSampleCounts = VK_SAMPLE_COUNT_2_BIT |<br>
> + VK_SAMPLE_COUNT_4_BIT |<br>
> + VK_SAMPLE_COUNT_8_BIT;<br>
> + properties->maxSampleLocationGridSize = (VkExtent2D){ 2 , 2 };<br>
> + properties->sampleLocationCoordinateRange[0] = 0.0f;<br>
> + properties->sampleLocationCoordinateRange[1] = 0.9375f;<br>
> + properties->sampleLocationSubPixelBits = 4;<br>
> + properties->variableSampleLocations = VK_FALSE;<br>
> + break;<br>
> + }<br>
> default:<br>
> break;<br>
> }<br>
> @@ -5361,3 +5374,17 @@ VkResult radv_GetCalibratedTimestampsEXT(<br>
><br>
> return VK_SUCCESS;<br>
> }<br>
> +<br>
> +void radv_GetPhysicalDeviceMultisamplePropertiesEXT(<br>
> + VkPhysicalDevice physicalDevice,<br>
> + VkSampleCountFlagBits samples,<br>
> + VkMultisamplePropertiesEXT* pMultisampleProperties)<br>
> +{<br>
> + if (samples & (VK_SAMPLE_COUNT_2_BIT |<br>
> + VK_SAMPLE_COUNT_4_BIT |<br>
> + VK_SAMPLE_COUNT_8_BIT)) {<br>
> + pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 2, 2 };<br>
> + } else {<br>
> + pMultisampleProperties->maxSampleLocationGridSize = (VkExtent2D){ 0, 0 };<br>
> + }<br>
> +}<br>
> diff --git a/src/amd/vulkan/radv_extensions.py b/src/amd/vulkan/radv_extensions.py<br>
> index 0b5af56a435..2c2c4dc74d2 100644<br>
> --- a/src/amd/vulkan/radv_extensions.py<br>
> +++ b/src/amd/vulkan/radv_extensions.py<br>
> @@ -119,6 +119,7 @@ EXTENSIONS = [<br>
> Extension('VK_EXT_memory_priority', 1, True),<br>
> Extension('VK_EXT_pci_bus_info', 2, True),<br>
> Extension('VK_EXT_pipeline_creation_feedback', 1, True),<br>
> + Extension('VK_EXT_sample_locations', 1, 'device->rad_info.chip_class >= GFX8'),<br>
> Extension('VK_EXT_sampler_filter_minmax', 1, 'device->rad_info.chip_class >= GFX7'),<br>
> Extension('VK_EXT_scalar_block_layout', 1, 'device->rad_info.chip_class >= GFX7'),<br>
> Extension('VK_EXT_shader_viewport_index_layer', 1, True),<br>
> diff --git a/src/amd/vulkan/radv_pipeline.c b/src/amd/vulkan/radv_pipeline.c<br>
> index 56fd65bec29..7f8e00b5c3c 100644<br>
> --- a/src/amd/vulkan/radv_pipeline.c<br>
> +++ b/src/amd/vulkan/radv_pipeline.c<br>
> @@ -1267,6 +1267,8 @@ static unsigned radv_dynamic_state_mask(VkDynamicState state)<br>
> return RADV_DYNAMIC_STENCIL_REFERENCE;<br>
> case VK_DYNAMIC_STATE_DISCARD_RECTANGLE_EXT:<br>
> return RADV_DYNAMIC_DISCARD_RECTANGLE;<br>
> + case VK_DYNAMIC_STATE_SAMPLE_LOCATIONS_EXT:<br>
> + return RADV_DYNAMIC_SAMPLE_LOCATIONS;<br>
> default:<br>
> unreachable("Unhandled dynamic state");<br>
> }<br>
> @@ -1297,6 +1299,11 @@ static uint32_t radv_pipeline_needed_dynamic_state(const VkGraphicsPipelineCreat<br>
> if (!vk_find_struct_const(pCreateInfo->pNext, PIPELINE_DISCARD_RECTANGLE_STATE_CREATE_INFO_EXT))<br>
> states &= ~RADV_DYNAMIC_DISCARD_RECTANGLE;<br>
><br>
> + if (!pCreateInfo->pMultisampleState ||<br>
> + !vk_find_struct_const(pCreateInfo->pMultisampleState->pNext,<br>
> + PIPELINE_SAMPLE_LOCATIONS_STATE_CREATE_INFO_EXT))<br>
> + states &= ~RADV_DYNAMIC_SAMPLE_LOCATIONS;<br>
> +<br>
> /* TODO: blend constants & line width. */<br>
><br>
> return states;<br>
> @@ -1426,6 +1433,29 @@ radv_pipeline_init_dynamic_state(struct radv_pipeline *pipeline,<br>
> }<br>
> }<br>
><br>
> + if (needed_states & RADV_DYNAMIC_SAMPLE_LOCATIONS) {<br>
> + const VkPipelineSampleLocationsStateCreateInfoEXT *sample_location_info =<br>
> + vk_find_struct_const(pCreateInfo->pMultisampleState->pNext,<br>
> + PIPELINE_SAMPLE_LOCATIONS_STATE_CREATE_INFO_EXT);<br>
> + /* If sampleLocationsEnable is VK_FALSE, the default sample<br>
> + * locations are used and the values specified in<br>
> + * sampleLocationsInfo are ignored.<br>
> + */<br>
> + if (sample_location_info->sampleLocationsEnable) {<br>
> + const VkSampleLocationsInfoEXT *pSampleLocationsInfo =<br>
> + &sample_location_info->sampleLocationsInfo;<br>
> +<br>
> + assert(pSampleLocationsInfo->sampleLocationsCount <= MAX_SAMPLE_LOCATIONS);<br>
> +<br>
> + dynamic->sample_location.per_pixel = pSampleLocationsInfo->sampleLocationsPerPixel;<br>
> + dynamic->sample_location.grid_size = pSampleLocationsInfo->sampleLocationGridSize;<br>
> + dynamic->sample_location.count = pSampleLocationsInfo->sampleLocationsCount;<br>
> + typed_memcpy(&dynamic->sample_location.locations[0],<br>
> + pSampleLocationsInfo->pSampleLocations,<br>
> + pSampleLocationsInfo->sampleLocationsCount);<br>
> + }<br>
> + }<br>
> +<br>
> pipeline->dynamic_state.mask = states;<br>
> }<br>
><br>
> diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h<br>
> index 7834a505562..5c84bf74352 100644<br>
> --- a/src/amd/vulkan/radv_private.h<br>
> +++ b/src/amd/vulkan/radv_private.h<br>
> @@ -91,6 +91,7 @@ typedef uint32_t xcb_window_t;<br>
> #define MAX_VIEWPORTS 16<br>
> #define MAX_SCISSORS 16<br>
> #define MAX_DISCARD_RECTANGLES 4<br>
> +#define MAX_SAMPLE_LOCATIONS 32<br>
> #define MAX_PUSH_CONSTANTS_SIZE 128<br>
> #define MAX_PUSH_DESCRIPTORS 32<br>
> #define MAX_DYNAMIC_UNIFORM_BUFFERS 16<br>
> @@ -852,7 +853,8 @@ enum radv_dynamic_state_bits {<br>
> RADV_DYNAMIC_STENCIL_WRITE_MASK = 1 << 7,<br>
> RADV_DYNAMIC_STENCIL_REFERENCE = 1 << 8,<br>
> RADV_DYNAMIC_DISCARD_RECTANGLE = 1 << 9,<br>
> - RADV_DYNAMIC_ALL = (1 << 10) - 1,<br>
> + RADV_DYNAMIC_SAMPLE_LOCATIONS = 1 << 10,<br>
> + RADV_DYNAMIC_ALL = (1 << 11) - 1,<br>
> };<br>
><br>
> enum radv_cmd_dirty_bits {<br>
> @@ -868,12 +870,13 @@ enum radv_cmd_dirty_bits {<br>
> RADV_CMD_DIRTY_DYNAMIC_STENCIL_WRITE_MASK = 1 << 7,<br>
> RADV_CMD_DIRTY_DYNAMIC_STENCIL_REFERENCE = 1 << 8,<br>
> RADV_CMD_DIRTY_DYNAMIC_DISCARD_RECTANGLE = 1 << 9,<br>
> - RADV_CMD_DIRTY_DYNAMIC_ALL = (1 << 10) - 1,<br>
> - RADV_CMD_DIRTY_PIPELINE = 1 << 10,<br>
> - RADV_CMD_DIRTY_INDEX_BUFFER = 1 << 11,<br>
> - RADV_CMD_DIRTY_FRAMEBUFFER = 1 << 12,<br>
> - RADV_CMD_DIRTY_VERTEX_BUFFER = 1 << 13,<br>
> - RADV_CMD_DIRTY_STREAMOUT_BUFFER = 1 << 14,<br>
> + RADV_CMD_DIRTY_DYNAMIC_SAMPLE_LOCATIONS = 1 << 10,<br>
> + RADV_CMD_DIRTY_DYNAMIC_ALL = (1 << 11) - 1,<br>
> + RADV_CMD_DIRTY_PIPELINE = 1 << 11,<br>
> + RADV_CMD_DIRTY_INDEX_BUFFER = 1 << 12,<br>
> + RADV_CMD_DIRTY_FRAMEBUFFER = 1 << 13,<br>
> + RADV_CMD_DIRTY_VERTEX_BUFFER = 1 << 14,<br>
> + RADV_CMD_DIRTY_STREAMOUT_BUFFER = 1 << 15,<br>
> };<br>
><br>
> enum radv_cmd_flush_bits {<br>
> @@ -950,6 +953,13 @@ struct radv_discard_rectangle_state {<br>
> VkRect2D rectangles[MAX_DISCARD_RECTANGLES];<br>
> };<br>
><br>
> +struct radv_sample_locations_state {<br>
> + VkSampleCountFlagBits per_pixel;<br>
> + VkExtent2D grid_size;<br>
> + uint32_t count;<br>
> + VkSampleLocationEXT locations[MAX_SAMPLE_LOCATIONS];<br>
> +};<br>
> +<br>
> struct radv_dynamic_state {<br>
> /**<br>
> * Bitmask of (1 << VK_DYNAMIC_STATE_*).<br>
> @@ -992,6 +1002,8 @@ struct radv_dynamic_state {<br>
> } stencil_reference;<br>
><br>
> struct radv_discard_rectangle_state discard_rectangle;<br>
> +<br>
> + struct radv_sample_locations_state sample_location;<br>
> };<br>
><br>
> extern const struct radv_dynamic_state default_dynamic_state;<br>
> --<br>
> 2.21.0<br>
><br>
> _______________________________________________<br>
> mesa-dev mailing list<br>
> <a href="mailto:mesa-dev@lists.freedesktop.org" target="_blank">mesa-dev@lists.freedesktop.org</a><br>
> <a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a><br>
_______________________________________________<br>
mesa-dev mailing list<br>
<a href="mailto:mesa-dev@lists.freedesktop.org" target="_blank">mesa-dev@lists.freedesktop.org</a><br>
<a href="https://lists.freedesktop.org/mailman/listinfo/mesa-dev" rel="noreferrer" target="_blank">https://lists.freedesktop.org/mailman/listinfo/mesa-dev</a></blockquote></div></div>
</blockquote></div></div>