From ea77472d75ae73524baf89069e38c76189ef00c1 Mon Sep 17 00:00:00 2001
From: Caleb Cornett <[EMAIL REDACTED]>
Date: Sun, 23 Feb 2025 13:34:28 -0500
Subject: [PATCH] Document/ignore GPU features without universal support
---
include/SDL3/SDL_gpu.h | 20 +++++++++++------
src/gpu/d3d12/SDL_gpu_d3d12.c | 8 ++-----
src/gpu/metal/SDL_gpu_metal.m | 11 +---------
src/gpu/vulkan/SDL_gpu_vulkan.c | 39 ++-------------------------------
4 files changed, 18 insertions(+), 60 deletions(-)
diff --git a/include/SDL3/SDL_gpu.h b/include/SDL3/SDL_gpu.h
index dd4238e521bba..b350c7c4cc42f 100644
--- a/include/SDL3/SDL_gpu.h
+++ b/include/SDL3/SDL_gpu.h
@@ -1495,6 +1495,9 @@ typedef struct SDL_GPUIndirectDispatchCommand
/**
* A structure specifying the parameters of a sampler.
*
+ * Note that mip_lod_bias is a no-op for the Metal driver.
+ * For Metal, LOD bias must be applied via shader instead.
+ *
* \since This function is available since SDL 3.2.0.
*
* \sa SDL_CreateGPUSampler
@@ -1547,7 +1550,7 @@ typedef struct SDL_GPUVertexBufferDescription
Uint32 slot; /**< The binding slot of the vertex buffer. */
Uint32 pitch; /**< The byte pitch between consecutive elements of the vertex buffer. */
SDL_GPUVertexInputRate input_rate; /**< Whether attribute addressing is a function of the vertex index or instance index. */
- Uint32 instance_step_rate; /**< The number of instances to draw using the same per-instance data before advancing in the instance buffer by one element. Ignored unless input_rate is SDL_GPU_VERTEXINPUTRATE_INSTANCE */
+ Uint32 instance_step_rate; /**< Ignored, reserved for future use. */
} SDL_GPUVertexBufferDescription;
/**
@@ -1717,10 +1720,13 @@ typedef struct SDL_GPUTransferBufferCreateInfo
* A structure specifying the parameters of the graphics pipeline rasterizer
* state.
*
- * NOTE: Some backend APIs (D3D11/12) will enable depth clamping even if
- * enable_depth_clip is true. If you rely on this clamp+clip behavior,
- * consider enabling depth clip and then manually clamping depth in your
- * fragment shaders on Metal and Vulkan.
+ * Note that SDL_GPU_FILLMODE_LINE is not supported on many Android devices.
+ * For those devices, the fill mode will automatically fall back to FILL.
+ *
+ * Also note that the D3D12 driver will enable depth clamping even if
+ * enable_depth_clip is true. If you need this clamp+clip behavior, consider
+ * enabling depth clip and then manually clamping depth in your fragment
+ * shaders on Metal and Vulkan.
*
* \since This struct is available since SDL 3.2.0.
*
@@ -1751,8 +1757,8 @@ typedef struct SDL_GPURasterizerState
typedef struct SDL_GPUMultisampleState
{
SDL_GPUSampleCount sample_count; /**< The number of samples to be used in rasterization. */
- Uint32 sample_mask; /**< Determines which samples get updated in the render targets. Treated as 0xFFFFFFFF if enable_mask is false. */
- bool enable_mask; /**< Enables sample masking. */
+ Uint32 sample_mask; /**< Ignored, reserved for future use. */
+ bool enable_mask; /**< Ignored, reserved for future use. */
Uint8 padding1;
Uint8 padding2;
Uint8 padding3;
diff --git a/src/gpu/d3d12/SDL_gpu_d3d12.c b/src/gpu/d3d12/SDL_gpu_d3d12.c
index f4fa85bfdbb4a..9f5bb95295c14 100644
--- a/src/gpu/d3d12/SDL_gpu_d3d12.c
+++ b/src/gpu/d3d12/SDL_gpu_d3d12.c
@@ -2928,7 +2928,7 @@ static bool D3D12_INTERNAL_ConvertVertexInputState(SDL_GPUVertexInputState verte
desc[i].AlignedByteOffset = attribute.offset;
desc[i].InputSlotClass = SDLToD3D12_InputRate[vertexInputState.vertex_buffer_descriptions[attribute.buffer_slot].input_rate];
desc[i].InstanceDataStepRate = (vertexInputState.vertex_buffer_descriptions[attribute.buffer_slot].input_rate == SDL_GPU_VERTEXINPUTRATE_INSTANCE)
- ? vertexInputState.vertex_buffer_descriptions[attribute.buffer_slot].instance_step_rate
+ ? 1
: 0;
}
@@ -3010,11 +3010,7 @@ static SDL_GPUGraphicsPipeline *D3D12_CreateGraphicsPipeline(
return NULL;
}
- Uint32 sampleMask = createinfo->multisample_state.enable_mask ?
- createinfo->multisample_state.sample_mask :
- 0xFFFFFFFF;
-
- psoDesc.SampleMask = sampleMask;
+ psoDesc.SampleMask = 0xFFFFFFFF;
psoDesc.SampleDesc.Count = SDLToD3D12_SampleCount[createinfo->multisample_state.sample_count];
psoDesc.SampleDesc.Quality = (createinfo->multisample_state.sample_count > SDL_GPU_SAMPLECOUNT_1) ? D3D12_STANDARD_MULTISAMPLE_PATTERN : 0;
diff --git a/src/gpu/metal/SDL_gpu_metal.m b/src/gpu/metal/SDL_gpu_metal.m
index 3020492d91334..365b344806d82 100644
--- a/src/gpu/metal/SDL_gpu_metal.m
+++ b/src/gpu/metal/SDL_gpu_metal.m
@@ -478,8 +478,6 @@ static MTLDepthClipMode SDLToMetal_DepthClipMode(
{
id<MTLRenderPipelineState> handle;
- Uint32 sample_mask;
-
SDL_GPURasterizerState rasterizerState;
SDL_GPUPrimitiveType primitiveType;
@@ -1181,9 +1179,7 @@ static void METAL_ReleaseGraphicsPipeline(
for (Uint32 i = 0; i < createinfo->vertex_input_state.num_vertex_buffers; i += 1) {
binding = METAL_FIRST_VERTEX_BUFFER_SLOT + createinfo->vertex_input_state.vertex_buffer_descriptions[i].slot;
vertexDescriptor.layouts[binding].stepFunction = SDLToMetal_StepFunction[createinfo->vertex_input_state.vertex_buffer_descriptions[i].input_rate];
- vertexDescriptor.layouts[binding].stepRate = (createinfo->vertex_input_state.vertex_buffer_descriptions[i].input_rate == SDL_GPU_VERTEXINPUTRATE_INSTANCE)
- ? createinfo->vertex_input_state.vertex_buffer_descriptions[i].instance_step_rate
- : 1;
+ vertexDescriptor.layouts[binding].stepRate = 1;
vertexDescriptor.layouts[binding].stride = createinfo->vertex_input_state.vertex_buffer_descriptions[i].pitch;
}
@@ -1202,13 +1198,8 @@ static void METAL_ReleaseGraphicsPipeline(
SET_ERROR_AND_RETURN("Creating render pipeline failed: %s", [[error description] UTF8String], NULL);
}
- Uint32 sampleMask = createinfo->multisample_state.enable_mask ?
- createinfo->multisample_state.sample_mask :
- 0xFFFFFFFF;
-
result = SDL_calloc(1, sizeof(MetalGraphicsPipeline));
result->handle = pipelineState;
- result->sample_mask = sampleMask;
result->depth_stencil_state = depthStencilState;
result->rasterizerState = createinfo->rasterizer_state;
result->primitiveType = createinfo->primitive_type;
diff --git a/src/gpu/vulkan/SDL_gpu_vulkan.c b/src/gpu/vulkan/SDL_gpu_vulkan.c
index cf832ad2b4756..5ef56b7e5fede 100644
--- a/src/gpu/vulkan/SDL_gpu_vulkan.c
+++ b/src/gpu/vulkan/SDL_gpu_vulkan.c
@@ -56,8 +56,6 @@ typedef struct VulkanExtensions
// Core since 1.2, but requires annoying paperwork to implement
Uint8 KHR_driver_properties;
- // EXT, probably not going to be Core
- Uint8 EXT_vertex_attribute_divisor;
// Only required for special implementations (i.e. MoltenVK)
Uint8 KHR_portability_subset;
// Only required for decoding HDR ASTC textures
@@ -6202,11 +6200,8 @@ static SDL_GPUGraphicsPipeline *VULKAN_CreateGraphicsPipeline(
VkPipelineShaderStageCreateInfo shaderStageCreateInfos[2];
VkPipelineVertexInputStateCreateInfo vertexInputStateCreateInfo;
- VkPipelineVertexInputDivisorStateCreateInfoEXT divisorStateCreateInfo;
VkVertexInputBindingDescription *vertexInputBindingDescriptions = SDL_stack_alloc(VkVertexInputBindingDescription, createinfo->vertex_input_state.num_vertex_buffers);
VkVertexInputAttributeDescription *vertexInputAttributeDescriptions = SDL_stack_alloc(VkVertexInputAttributeDescription, createinfo->vertex_input_state.num_vertex_attributes);
- VkVertexInputBindingDivisorDescriptionEXT *divisorDescriptions = SDL_stack_alloc(VkVertexInputBindingDivisorDescriptionEXT, createinfo->vertex_input_state.num_vertex_buffers);
- Uint32 divisorDescriptionCount = 0;
VkPipelineInputAssemblyStateCreateInfo inputAssemblyStateCreateInfo;
@@ -6289,10 +6284,6 @@ static SDL_GPUGraphicsPipeline *VULKAN_CreateGraphicsPipeline(
vertexInputBindingDescriptions[i].binding = createinfo->vertex_input_state.vertex_buffer_descriptions[i].slot;
vertexInputBindingDescriptions[i].inputRate = SDLToVK_VertexInputRate[createinfo->vertex_input_state.vertex_buffer_descriptions[i].input_rate];
vertexInputBindingDescriptions[i].stride = createinfo->vertex_input_state.vertex_buffer_descriptions[i].pitch;
-
- if (createinfo->vertex_input_state.vertex_buffer_descriptions[i].input_rate == SDL_GPU_VERTEXINPUTRATE_INSTANCE) {
- divisorDescriptionCount += 1;
- }
}
for (i = 0; i < createinfo->vertex_input_state.num_vertex_attributes; i += 1) {
@@ -6310,26 +6301,6 @@ static SDL_GPUGraphicsPipeline *VULKAN_CreateGraphicsPipeline(
vertexInputStateCreateInfo.vertexAttributeDescriptionCount = createinfo->vertex_input_state.num_vertex_attributes;
vertexInputStateCreateInfo.pVertexAttributeDescriptions = vertexInputAttributeDescriptions;
- if (divisorDescriptionCount > 0) {
- divisorDescriptionCount = 0;
-
- for (i = 0; i < createinfo->vertex_input_state.num_vertex_buffers; i += 1) {
- if (createinfo->vertex_input_state.vertex_buffer_descriptions[i].input_rate == SDL_GPU_VERTEXINPUTRATE_INSTANCE) {
- divisorDescriptions[divisorDescriptionCount].binding = createinfo->vertex_input_state.vertex_buffer_descriptions[i].slot;
- divisorDescriptions[divisorDescriptionCount].divisor = createinfo->vertex_input_state.vertex_buffer_descriptions[i].instance_step_rate;
-
- divisorDescriptionCount += 1;
- }
- }
-
- divisorStateCreateInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_VERTEX_INPUT_DIVISOR_STATE_CREATE_INFO_EXT;
- divisorStateCreateInfo.pNext = NULL;
- divisorStateCreateInfo.vertexBindingDivisorCount = divisorDescriptionCount;
- divisorStateCreateInfo.pVertexBindingDivisors = divisorDescriptions;
-
- vertexInputStateCreateInfo.pNext = &divisorStateCreateInfo;
- }
-
// Topology
inputAssemblyStateCreateInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_INPUT_ASSEMBLY_STATE_CREATE_INFO;
@@ -6376,9 +6347,7 @@ static SDL_GPUGraphicsPipeline *VULKAN_CreateGraphicsPipeline(
// Multisample
- Uint32 sampleMask = createinfo->multisample_state.enable_mask ?
- createinfo->multisample_state.sample_mask :
- 0xFFFFFFFF;
+ Uint32 sampleMask = 0xFFFFFFFF;
multisampleStateCreateInfo.sType = VK_STRUCTURE_TYPE_PIPELINE_MULTISAMPLE_STATE_CREATE_INFO;
multisampleStateCreateInfo.pNext = NULL;
@@ -6476,7 +6445,6 @@ static SDL_GPUGraphicsPipeline *VULKAN_CreateGraphicsPipeline(
SDL_stack_free(vertexInputBindingDescriptions);
SDL_stack_free(vertexInputAttributeDescriptions);
SDL_stack_free(colorBlendAttachmentStates);
- SDL_stack_free(divisorDescriptions);
SDL_free(graphicsPipeline);
SET_STRING_ERROR_AND_RETURN("Failed to initialize pipeline resource layout!", NULL);
}
@@ -6515,7 +6483,6 @@ static SDL_GPUGraphicsPipeline *VULKAN_CreateGraphicsPipeline(
SDL_stack_free(vertexInputBindingDescriptions);
SDL_stack_free(vertexInputAttributeDescriptions);
SDL_stack_free(colorBlendAttachmentStates);
- SDL_stack_free(divisorDescriptions);
renderer->vkDestroyRenderPass(
renderer->logicalDevice,
@@ -10898,7 +10865,7 @@ static inline Uint8 CheckDeviceExtensions(
supports->ext = 1; \
}
CHECK(KHR_swapchain)
- else CHECK(KHR_maintenance1) else CHECK(KHR_driver_properties) else CHECK(EXT_vertex_attribute_divisor) else CHECK(KHR_portability_subset) else CHECK(EXT_texture_compression_astc_hdr)
+ else CHECK(KHR_maintenance1) else CHECK(KHR_driver_properties) CHECK(KHR_portability_subset) else CHECK(EXT_texture_compression_astc_hdr)
#undef CHECK
}
@@ -10912,7 +10879,6 @@ static inline Uint32 GetDeviceExtensionCount(VulkanExtensions *supports)
supports->KHR_swapchain +
supports->KHR_maintenance1 +
supports->KHR_driver_properties +
- supports->EXT_vertex_attribute_divisor +
supports->KHR_portability_subset +
supports->EXT_texture_compression_astc_hdr);
}
@@ -10929,7 +10895,6 @@ static inline void CreateDeviceExtensionArray(
CHECK(KHR_swapchain)
CHECK(KHR_maintenance1)
CHECK(KHR_driver_properties)
- CHECK(EXT_vertex_attribute_divisor)
CHECK(KHR_portability_subset)
CHECK(EXT_texture_compression_astc_hdr)
#undef CHECK