Mesa (master): radv: Implement VK_EXT_custom_border_color
GitLab Mirror
gitlab-mirror at kemper.freedesktop.org
Fri May 15 01:46:10 UTC 2020
Module: Mesa
Branch: master
Commit: 57e796a12a883f5845c8e0ed05ec6bdb6d055b53
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=57e796a12a883f5845c8e0ed05ec6bdb6d055b53
Author: Joshua Ashton <joshua at froggi.es>
Date: Tue Apr 7 06:11:24 2020 +0100
radv: Implement VK_EXT_custom_border_color
Reviewed-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas at basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/4877>
---
docs/relnotes/new_features.txt | 1 +
src/amd/vulkan/radv_device.c | 133 +++++++++++++++++++++++++++++++++++++-
src/amd/vulkan/radv_extensions.py | 1 +
src/amd/vulkan/radv_private.h | 17 +++++
src/amd/vulkan/si_cmd_buffer.c | 10 +++
5 files changed, 160 insertions(+), 2 deletions(-)
diff --git a/docs/relnotes/new_features.txt b/docs/relnotes/new_features.txt
index 857b888ac41..3562f5c7122 100644
--- a/docs/relnotes/new_features.txt
+++ b/docs/relnotes/new_features.txt
@@ -1,2 +1,3 @@
GL_ARB_compute_variable_group_size on Iris.
VK_EXT_private_data on ANV and RADV.
+VK_EXT_custom_border_color on RADV.
diff --git a/src/amd/vulkan/radv_device.c b/src/amd/vulkan/radv_device.c
index bdfb2dad355..b4b429a4ed7 100644
--- a/src/amd/vulkan/radv_device.c
+++ b/src/amd/vulkan/radv_device.c
@@ -1294,6 +1294,12 @@ void radv_GetPhysicalDeviceFeatures2(
features->robustBufferAccess2 = true;
features->robustImageAccess2 = true;
features->nullDescriptor = true;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: {
+ VkPhysicalDeviceCustomBorderColorFeaturesEXT *features =
+ (VkPhysicalDeviceCustomBorderColorFeaturesEXT *)ext;
+ features->customBorderColors = true;
+ features->customBorderColorWithoutFormat = true;
break;
}
case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_PRIVATE_DATA_FEATURES_EXT: {
@@ -1938,6 +1944,11 @@ void radv_GetPhysicalDeviceProperties2(
(VkPhysicalDeviceRobustness2PropertiesEXT *)ext;
properties->robustStorageBufferAccessSizeAlignment = 4;
properties->robustUniformBufferAccessSizeAlignment = 4;
+ }
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_PROPERTIES_EXT: {
+ VkPhysicalDeviceCustomBorderColorPropertiesEXT *props =
+ (VkPhysicalDeviceCustomBorderColorPropertiesEXT *)ext;
+ props->maxCustomBorderColorSamplers = RADV_BORDER_COLOR_COUNT;
break;
}
default:
@@ -2936,6 +2947,37 @@ check_physical_device_features(VkPhysicalDevice physicalDevice,
return VK_SUCCESS;
}
+static VkResult radv_device_init_border_color(struct radv_device *device)
+{
+ device->border_color_data.bo =
+ device->ws->buffer_create(device->ws,
+ RADV_BORDER_COLOR_BUFFER_SIZE,
+ 4096,
+ RADEON_DOMAIN_VRAM,
+ RADEON_FLAG_CPU_ACCESS |
+ RADEON_FLAG_READ_ONLY |
+ RADEON_FLAG_NO_INTERPROCESS_SHARING,
+ RADV_BO_PRIORITY_SHADER);
+
+ if (device->border_color_data.bo == NULL)
+ return vk_error(device->physical_device->instance, VK_ERROR_OUT_OF_DEVICE_MEMORY);
+
+ device->border_color_data.colors_gpu_ptr =
+ device->ws->buffer_map(device->border_color_data.bo);
+ pthread_mutex_init(&device->border_color_data.mutex, NULL);
+
+ return VK_SUCCESS;
+}
+
+static void radv_device_finish_border_color(struct radv_device *device)
+{
+ if (device->border_color_data.bo) {
+ device->ws->buffer_destroy(device->border_color_data.bo);
+
+ pthread_mutex_destroy(&device->border_color_data.mutex);
+ }
+}
+
VkResult radv_CreateDevice(
VkPhysicalDevice physicalDevice,
const VkDeviceCreateInfo* pCreateInfo,
@@ -2949,6 +2991,7 @@ VkResult radv_CreateDevice(
bool keep_shader_info = false;
bool robust_buffer_access = false;
bool overallocation_disallowed = false;
+ bool custom_border_colors = false;
/* Check enabled features */
if (pCreateInfo->pEnabledFeatures) {
@@ -2980,6 +3023,11 @@ VkResult radv_CreateDevice(
overallocation_disallowed = true;
break;
}
+ case VK_STRUCTURE_TYPE_PHYSICAL_DEVICE_CUSTOM_BORDER_COLOR_FEATURES_EXT: {
+ const VkPhysicalDeviceCustomBorderColorFeaturesEXT *border_color_features = (const void *)ext;
+ custom_border_colors = border_color_features->customBorderColors;
+ break;
+ }
default:
break;
}
@@ -3150,6 +3198,13 @@ VkResult radv_CreateDevice(
radv_device_init_msaa(device);
+ /* If the border color extension is enabled, let's create the buffer we need. */
+ if (custom_border_colors) {
+ result = radv_device_init_border_color(device);
+ if (result != VK_SUCCESS)
+ goto fail;
+ }
+
for (int family = 0; family < RADV_MAX_QUEUE_FAMILIES; ++family) {
device->empty_cs[family] = device->ws->cs_create(device->ws, family);
switch (family) {
@@ -3221,6 +3276,8 @@ fail:
if (device->gfx_init)
device->ws->buffer_destroy(device->gfx_init);
+ radv_device_finish_border_color(device);
+
for (unsigned i = 0; i < RADV_MAX_QUEUE_FAMILIES; i++) {
for (unsigned q = 0; q < device->queue_count[i]; q++)
radv_queue_finish(&device->queues[i][q]);
@@ -3247,6 +3304,8 @@ void radv_DestroyDevice(
if (device->gfx_init)
device->ws->buffer_destroy(device->gfx_init);
+ radv_device_finish_border_color(device);
+
for (unsigned i = 0; i < RADV_MAX_QUEUE_FAMILIES; i++) {
for (unsigned q = 0; q < device->queue_count[i]; q++)
radv_queue_finish(&device->queues[i][q]);
@@ -7094,6 +7153,9 @@ radv_tex_bordercolor(VkBorderColor bcolor)
case VK_BORDER_COLOR_FLOAT_OPAQUE_WHITE:
case VK_BORDER_COLOR_INT_OPAQUE_WHITE:
return V_008F3C_SQ_TEX_BORDER_COLOR_OPAQUE_WHITE;
+ case VK_BORDER_COLOR_FLOAT_CUSTOM_EXT:
+ case VK_BORDER_COLOR_INT_CUSTOM_EXT:
+ return V_008F3C_SQ_TEX_BORDER_COLOR_REGISTER;
default:
break;
}
@@ -7149,6 +7211,40 @@ static inline int S_FIXED(float value, unsigned frac_bits)
return value * (1 << frac_bits);
}
+static uint32_t radv_register_border_color(struct radv_device *device,
+ VkClearColorValue value)
+{
+ uint32_t slot;
+
+ pthread_mutex_lock(&device->border_color_data.mutex);
+
+ for (slot = 0; slot < RADV_BORDER_COLOR_COUNT; slot++) {
+ if (!device->border_color_data.used[slot]) {
+ /* Copy to the GPU wrt endian-ness. */
+ util_memcpy_cpu_to_le32(&device->border_color_data.colors_gpu_ptr[slot],
+ &value,
+ sizeof(VkClearColorValue));
+
+ device->border_color_data.used[slot] = true;
+ break;
+ }
+ }
+
+ pthread_mutex_unlock(&device->border_color_data.mutex);
+
+ return slot;
+}
+
+static void radv_unregister_border_color(struct radv_device *device,
+ uint32_t slot)
+{
+ pthread_mutex_lock(&device->border_color_data.mutex);
+
+ device->border_color_data.used[slot] = false;
+
+ pthread_mutex_unlock(&device->border_color_data.mutex);
+}
+
static void
radv_init_sampler(struct radv_device *device,
struct radv_sampler *sampler,
@@ -7161,6 +7257,11 @@ radv_init_sampler(struct radv_device *device,
unsigned filter_mode = V_008F30_SQ_IMG_FILTER_MODE_BLEND;
unsigned depth_compare_func = V_008F30_SQ_TEX_DEPTH_COMPARE_NEVER;
bool trunc_coord = pCreateInfo->minFilter == VK_FILTER_NEAREST && pCreateInfo->magFilter == VK_FILTER_NEAREST;
+ bool uses_border_color = pCreateInfo->addressModeU == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
+ pCreateInfo->addressModeV == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER ||
+ pCreateInfo->addressModeW == VK_SAMPLER_ADDRESS_MODE_CLAMP_TO_BORDER;
+ VkBorderColor border_color = uses_border_color ? pCreateInfo->borderColor : VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
+ uint32_t border_color_ptr;
const struct VkSamplerReductionModeCreateInfo *sampler_reduction =
vk_find_struct_const(pCreateInfo->pNext,
@@ -7171,6 +7272,30 @@ radv_init_sampler(struct radv_device *device,
if (pCreateInfo->compareEnable)
depth_compare_func = radv_tex_compare(pCreateInfo->compareOp);
+ sampler->border_color_slot = RADV_BORDER_COLOR_COUNT;
+
+ if (border_color == VK_BORDER_COLOR_FLOAT_CUSTOM_EXT || border_color == VK_BORDER_COLOR_INT_CUSTOM_EXT) {
+ const VkSamplerCustomBorderColorCreateInfoEXT *custom_border_color =
+ vk_find_struct_const(pCreateInfo->pNext,
+ SAMPLER_CUSTOM_BORDER_COLOR_CREATE_INFO_EXT);
+
+ assert(custom_border_color);
+
+ sampler->border_color_slot =
+ radv_register_border_color(device, custom_border_color->customBorderColor);
+
+ /* Did we fail to find a slot? */
+ if (sampler->border_color_slot == RADV_BORDER_COLOR_COUNT) {
+ fprintf(stderr, "WARNING: no free border color slots, defaulting to TRANS_BLACK.\n");
+ border_color = VK_BORDER_COLOR_FLOAT_TRANSPARENT_BLACK;
+ }
+ }
+
+ /* If we don't have a custom color, set the ptr to 0 */
+ border_color_ptr = sampler->border_color_slot != RADV_BORDER_COLOR_COUNT
+ ? sampler->border_color_slot
+ : 0;
+
sampler->state[0] = (S_008F30_CLAMP_X(radv_tex_wrap(pCreateInfo->addressModeU)) |
S_008F30_CLAMP_Y(radv_tex_wrap(pCreateInfo->addressModeV)) |
S_008F30_CLAMP_Z(radv_tex_wrap(pCreateInfo->addressModeW)) |
@@ -7191,8 +7316,8 @@ radv_init_sampler(struct radv_device *device,
S_008F38_XY_MIN_FILTER(radv_tex_filter(pCreateInfo->minFilter, max_aniso)) |
S_008F38_MIP_FILTER(radv_tex_mipfilter(pCreateInfo->mipmapMode)) |
S_008F38_MIP_POINT_PRECLAMP(0));
- sampler->state[3] = (S_008F3C_BORDER_COLOR_PTR(0) |
- S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(pCreateInfo->borderColor)));
+ sampler->state[3] = (S_008F3C_BORDER_COLOR_PTR(border_color_ptr) |
+ S_008F3C_BORDER_COLOR_TYPE(radv_tex_bordercolor(border_color)));
if (device->physical_device->rad_info.chip_class >= GFX10) {
sampler->state[2] |= S_008F38_ANISO_OVERRIDE_GFX10(1);
@@ -7245,6 +7370,10 @@ void radv_DestroySampler(
if (!sampler)
return;
+
+ if (sampler->border_color_slot != RADV_BORDER_COLOR_COUNT)
+ radv_unregister_border_color(device, sampler->border_color_slot);
+
vk_object_base_finish(&sampler->base);
vk_free2(&device->vk.alloc, pAllocator, sampler);
}
diff --git a/src/amd/vulkan/radv_extensions.py b/src/amd/vulkan/radv_extensions.py
index 0549a09f377..23c32653651 100644
--- a/src/amd/vulkan/radv_extensions.py
+++ b/src/amd/vulkan/radv_extensions.py
@@ -122,6 +122,7 @@ EXTENSIONS = [
Extension('VK_EXT_calibrated_timestamps', 1, True),
Extension('VK_EXT_conditional_rendering', 1, True),
Extension('VK_EXT_conservative_rasterization', 1, 'device->rad_info.chip_class >= GFX9'),
+ Extension('VK_EXT_custom_border_color', 12, True),
Extension('VK_EXT_display_surface_counter', 1, 'VK_USE_PLATFORM_DISPLAY_KHR'),
Extension('VK_EXT_display_control', 1, 'VK_USE_PLATFORM_DISPLAY_KHR'),
Extension('VK_EXT_debug_report', 9, True),
diff --git a/src/amd/vulkan/radv_private.h b/src/amd/vulkan/radv_private.h
index 27d512a5f49..d655fa07d9b 100644
--- a/src/amd/vulkan/radv_private.h
+++ b/src/amd/vulkan/radv_private.h
@@ -781,6 +781,20 @@ struct radv_secure_compile_state {
char *uid;
};
+#define RADV_BORDER_COLOR_COUNT 4096
+#define RADV_BORDER_COLOR_BUFFER_SIZE (sizeof(VkClearColorValue) * RADV_BORDER_COLOR_COUNT)
+
+struct radv_device_border_color_data {
+ bool used[RADV_BORDER_COLOR_COUNT];
+
+ struct radeon_winsys_bo *bo;
+ VkClearColorValue *colors_gpu_ptr;
+
+ /* Mutex is required to guarantee vkCreateSampler thread safety
+ * given that we are writing to a buffer and checking color occupation */
+ pthread_mutex_t mutex;
+};
+
struct radv_device {
struct vk_device vk;
@@ -851,6 +865,8 @@ struct radv_device {
/* Whether anisotropy is forced with RADV_TEX_ANISO (-1 is disabled). */
int force_aniso;
+ struct radv_device_border_color_data border_color_data;
+
struct radv_secure_compile_state *sc_state;
/* Condition variable for legacy timelines, to notify waiters when a
@@ -2162,6 +2178,7 @@ struct radv_sampler {
struct vk_object_base base;
uint32_t state[4];
struct radv_sampler_ycbcr_conversion *ycbcr_sampler;
+ uint32_t border_color_slot;
};
struct radv_framebuffer {
diff --git a/src/amd/vulkan/si_cmd_buffer.c b/src/amd/vulkan/si_cmd_buffer.c
index 03fa89b74c7..ad71ee55ac2 100644
--- a/src/amd/vulkan/si_cmd_buffer.c
+++ b/src/amd/vulkan/si_cmd_buffer.c
@@ -456,6 +456,16 @@ si_emit_graphics(struct radv_device *device,
radeon_set_context_reg(cs, R_028C5C_VGT_OUT_DEALLOC_CNTL, 16);
}
+ if (device->border_color_data.bo) {
+ uint64_t border_color_va = radv_buffer_get_va(device->border_color_data.bo);
+
+ radeon_set_context_reg(cs, R_028080_TA_BC_BASE_ADDR, border_color_va >> 8);
+ if (physical_device->rad_info.chip_class >= GFX7) {
+ radeon_set_context_reg(cs, R_028084_TA_BC_BASE_ADDR_HI,
+ S_028084_ADDRESS(border_color_va >> 40));
+ }
+ }
+
if (physical_device->rad_info.chip_class >= GFX9) {
radeon_set_context_reg(cs, R_028C48_PA_SC_BINNER_CNTL_1,
S_028C48_MAX_ALLOC_COUNT(physical_device->rad_info.pbb_max_alloc_count - 1) |
More information about the mesa-commit
mailing list