rework CreateTransferBuffer

include/SDL3/SDL_gpu.h

@@ -177,14 +177,6 @@ typedef enum SDL_GpuBufferUsageFlagBits
 
 typedef Uint32 SDL_GpuBufferUsageFlags;
 
-typedef enum SDL_GpuTransferBufferMapFlagBits
-{
-    SDL_GPU_TRANSFER_MAP_READ = 0x00000001,
-    SDL_GPU_TRANSFER_MAP_WRITE = 0x00000002
-} SDL_GpuTransferBufferMapFlagBits;
-
-typedef Uint32 SDL_GpuTransferBufferMapFlags;
-
 typedef enum SDL_GpuShaderStage
 {
     SDL_GPU_SHADERSTAGE_VERTEX,
@@ -336,12 +328,6 @@ typedef enum SDL_GpuBorderColor
     SDL_GPU_BORDERCOLOR_INT_OPAQUE_WHITE
 } SDL_GpuBorderColor;
 
-typedef enum SDL_GpuTransferUsage
-{
-    SDL_GPU_TRANSFERUSAGE_BUFFER,
-    SDL_GPU_TRANSFERUSAGE_TEXTURE
-} SDL_GpuTransferUsage;
-
 /*
  * VSYNC:
  *   Waits for vblank before presenting.
@@ -950,8 +936,7 @@ extern SDL_DECLSPEC SDL_GpuBuffer *SDLCALL SDL_GpuCreateBuffer(
  * Creates a transfer buffer to be used when uploading to or downloading from graphics resources.
  *
  * \param device a GPU Context
- * \param usage specifies whether the transfer buffer will transfer buffers or textures
- * \param mapFlags specify read-write options for the transfer buffer
+ * \param uploadOnly specifies that the transfer buffer will only be used for uploads, allows optimizations on certain backends
  * \param sizeInBytes the size of the transfer buffer
  * \returns a transfer buffer on success, or NULL on failure
  *
@@ -965,8 +950,7 @@ extern SDL_DECLSPEC SDL_GpuBuffer *SDLCALL SDL_GpuCreateBuffer(
  */
 extern SDL_DECLSPEC SDL_GpuTransferBuffer *SDLCALL SDL_GpuCreateTransferBuffer(
     SDL_GpuDevice *device,
-    SDL_GpuTransferUsage usage,
-    SDL_GpuTransferBufferMapFlags mapFlags,
+    SDL_bool uploadOnly,
     Uint32 sizeInBytes);
 
 /* Debug Naming */

src/dynapi/SDL_dynapi_procs.h

@@ -1060,7 +1060,7 @@ SDL_DYNAPI_PROC(SDL_GpuSampler*,SDL_GpuCreateSampler,(SDL_GpuDevice *a, SDL_GpuS
 SDL_DYNAPI_PROC(SDL_GpuShader*,SDL_GpuCreateShader,(SDL_GpuDevice *a, SDL_GpuShaderCreateInfo *b),(a,b),return)
 SDL_DYNAPI_PROC(SDL_GpuTexture*,SDL_GpuCreateTexture,(SDL_GpuDevice *a, SDL_GpuTextureCreateInfo *b),(a,b),return)
 SDL_DYNAPI_PROC(SDL_GpuBuffer*,SDL_GpuCreateBuffer,(SDL_GpuDevice *a, SDL_GpuBufferUsageFlags b, Uint32 c),(a,b,c),return)
-SDL_DYNAPI_PROC(SDL_GpuTransferBuffer*,SDL_GpuCreateTransferBuffer,(SDL_GpuDevice *a, SDL_GpuTransferUsage b, SDL_GpuTransferBufferMapFlags c, Uint32 d),(a,b,c,d),return)
+SDL_DYNAPI_PROC(SDL_GpuTransferBuffer*,SDL_GpuCreateTransferBuffer,(SDL_GpuDevice *a, SDL_bool b, Uint32 c),(a,b,c),return)
 SDL_DYNAPI_PROC(void,SDL_GpuSetBufferName,(SDL_GpuDevice *a, SDL_GpuBuffer *b, const char *c),(a,b,c),)
 SDL_DYNAPI_PROC(void,SDL_GpuSetTextureName,(SDL_GpuDevice *a, SDL_GpuTexture *b, const char *c),(a,b,c),)
 SDL_DYNAPI_PROC(void,SDL_GpuSetStringMarker,(SDL_GpuCommandBuffer *a, const char *b),(a,b),)

src/gpu/SDL_gpu.c

@@ -444,15 +444,13 @@ SDL_GpuBuffer *SDL_GpuCreateBuffer(
 
 SDL_GpuTransferBuffer *SDL_GpuCreateTransferBuffer(
     SDL_GpuDevice *device,
-    SDL_GpuTransferUsage usage,
-    SDL_GpuTransferBufferMapFlags mapFlags,
+    SDL_bool uploadOnly,
     Uint32 sizeInBytes)
 {
     NULL_ASSERT(device)
     return device->CreateTransferBuffer(
         device->driverData,
-        usage,
-        mapFlags,
+        uploadOnly,
         sizeInBytes);
 }
 

src/gpu/SDL_gpu_driver.h

@@ -231,8 +231,7 @@ struct SDL_GpuDevice
 
     SDL_GpuTransferBuffer *(*CreateTransferBuffer)(
         SDL_GpuRenderer *driverData,
-        SDL_GpuTransferUsage usage,
-        SDL_GpuTransferBufferMapFlags mapFlags,
+        SDL_bool uploadOnly,
         Uint32 sizeInBytes);
 
     /* Debug Naming */

src/gpu/d3d11/SDL_gpu_d3d11.c

@@ -614,9 +614,6 @@ typedef struct D3D11TransferBuffer
 
 typedef struct D3D11TransferBufferContainer
 {
-    SDL_GpuTransferUsage usage;
-    SDL_GpuTransferBufferMapFlags mapFlags;
-
     D3D11TransferBuffer *activeBuffer;
 
     /* These are all the buffers that have been used by this container.
@@ -1885,7 +1882,7 @@ static D3D11Texture *D3D11_INTERNAL_CreateTexture(
     SDL_GpuTextureCreateInfo *textureCreateInfo,
     D3D11_SUBRESOURCE_DATA *initialData)
 {
-    Uint8 isSampler, isColorTarget, isDepthStencil, isMultisample, needSubresourceSRV, needSubresourceUAV;
+    Uint8 isSampler, isColorTarget, isDepthStencil, isMultisample, isStaging, needSubresourceSRV, needSubresourceUAV;
     DXGI_FORMAT format;
     ID3D11Resource *textureHandle;
     ID3D11ShaderResourceView *srv = NULL;
@@ -1901,6 +1898,7 @@ static D3D11Texture *D3D11_INTERNAL_CreateTexture(
     needSubresourceUAV =
         (textureCreateInfo->usageFlags & SDL_GPU_TEXTUREUSAGE_COMPUTE_STORAGE_WRITE_BIT);
     isMultisample = textureCreateInfo->sampleCount > SDL_GPU_SAMPLECOUNT_1;
+    isStaging = textureCreateInfo->usageFlags == 0;
 
     format = SDLToD3D11_TextureFormat[textureCreateInfo->format];
     if (isDepthStencil) {
@@ -1927,13 +1925,13 @@ static D3D11Texture *D3D11_INTERNAL_CreateTexture(
         desc2D.Width = textureCreateInfo->width;
         desc2D.Height = textureCreateInfo->height;
         desc2D.ArraySize = textureCreateInfo->isCube ? 6 : textureCreateInfo->layerCount;
-        desc2D.CPUAccessFlags = 0;
+        desc2D.CPUAccessFlags = isStaging ? D3D11_CPU_ACCESS_WRITE : 0;
         desc2D.Format = format;
         desc2D.MipLevels = textureCreateInfo->levelCount;
         desc2D.MiscFlags = textureCreateInfo->isCube ? D3D11_RESOURCE_MISC_TEXTURECUBE : 0;
         desc2D.SampleDesc.Count = 1;
         desc2D.SampleDesc.Quality = 0;
-        desc2D.Usage = D3D11_USAGE_DEFAULT;
+        desc2D.Usage = isStaging ? D3D11_USAGE_STAGING : D3D11_USAGE_DEFAULT;
 
         res = ID3D11Device_CreateTexture2D(
             renderer->device,
@@ -1991,11 +1989,11 @@ static D3D11Texture *D3D11_INTERNAL_CreateTexture(
         desc3D.Width = textureCreateInfo->width;
         desc3D.Height = textureCreateInfo->height;
         desc3D.Depth = textureCreateInfo->depth;
-        desc3D.CPUAccessFlags = 0;
+        desc3D.CPUAccessFlags = isStaging ? D3D11_CPU_ACCESS_WRITE : 0;
         desc3D.Format = format;
         desc3D.MipLevels = textureCreateInfo->levelCount;
         desc3D.MiscFlags = 0;
-        desc3D.Usage = D3D11_USAGE_DEFAULT;
+        desc3D.Usage = isStaging ? D3D11_USAGE_STAGING : D3D11_USAGE_DEFAULT;
 
         res = ID3D11Device_CreateTexture3D(
             renderer->device,
@@ -2588,8 +2586,6 @@ static D3D11Buffer *D3D11_INTERNAL_PrepareBufferForWrite(
 
 static D3D11TransferBuffer *D3D11_INTERNAL_CreateTransferBuffer(
     D3D11Renderer *renderer,
-    SDL_GpuTransferUsage usage,
-    SDL_GpuTransferBufferMapFlags mapFlags,
     Uint32 sizeInBytes)
 {
     D3D11TransferBuffer *transferBuffer = SDL_malloc(sizeof(D3D11TransferBuffer));
@@ -2605,24 +2601,19 @@ static D3D11TransferBuffer *D3D11_INTERNAL_CreateTransferBuffer(
 /* This actually returns a container handle so we can rotate buffers on Cycle. */
 static SDL_GpuTransferBuffer *D3D11_CreateTransferBuffer(
     SDL_GpuRenderer *driverData,
-    SDL_GpuTransferUsage usage,
-    SDL_GpuTransferBufferMapFlags mapFlags,
+    SDL_bool uploadOnly, /* ignored on D3D11 */
     Uint32 sizeInBytes)
 {
     D3D11Renderer *renderer = (D3D11Renderer *)driverData;
     D3D11TransferBufferContainer *container = (D3D11TransferBufferContainer *)SDL_malloc(sizeof(D3D11TransferBufferContainer));
 
-    container->usage = usage;
-    container->mapFlags = mapFlags;
     container->bufferCapacity = 1;
     container->bufferCount = 1;
     container->buffers = SDL_malloc(
         container->bufferCapacity * sizeof(D3D11TransferBuffer *));
 
     container->buffers[0] = D3D11_INTERNAL_CreateTransferBuffer(
         renderer,
-        usage,
-        mapFlags,
         sizeInBytes);
 
     container->activeBuffer = container->buffers[0];
@@ -2654,8 +2645,6 @@ static void D3D11_INTERNAL_CycleActiveTransferBuffer(
 
     container->buffers[container->bufferCount] = D3D11_INTERNAL_CreateTransferBuffer(
         renderer,
-        container->usage,
-        container->mapFlags,
         size);
     container->bufferCount += 1;
 

src/gpu/metal/SDL_gpu_metal.m

@@ -430,7 +430,8 @@ static MTLColorWriteMask SDLToMetal_ColorWriteMask(
     Uint32 bufferCount;
     MetalBuffer **buffers;
 
-    SDL_GpuTransferBufferMapFlags transferMapFlags;
+    SDL_bool isPrivate;
+    SDL_bool isWriteOnly;
     char *debugName;
 } MetalBufferContainer;
 
@@ -1356,7 +1357,8 @@ static void METAL_INTERNAL_CycleActiveTexture(
 static MetalBufferContainer *METAL_INTERNAL_CreateBufferContainer(
     MetalRenderer *renderer,
     Uint32 sizeInBytes,
-    SDL_GpuTransferBufferMapFlags transferMapFlags)
+    SDL_bool isPrivate,
+    SDL_bool isWriteOnly)
 {
     MetalBufferContainer *container = SDL_malloc(sizeof(MetalBufferContainer));
     MTLResourceOptions resourceOptions;
@@ -1366,15 +1368,19 @@ static void METAL_INTERNAL_CycleActiveTexture(
     container->bufferCount = 1;
     container->buffers = SDL_malloc(
         container->bufferCapacity * sizeof(MetalBuffer *));
-    container->transferMapFlags = transferMapFlags;
+    container->isPrivate = isPrivate;
+    container->isWriteOnly = isWriteOnly;
     container->debugName = NULL;
 
-    if (transferMapFlags == 0) {
+    if (isPrivate) {
         resourceOptions = MTLResourceStorageModePrivate;
-    } else if ((transferMapFlags & SDL_GPU_TRANSFER_MAP_WRITE) && !(transferMapFlags & SDL_GPU_TRANSFER_MAP_READ)) {
-        resourceOptions = MTLResourceCPUCacheModeWriteCombined;
-    } else {
-        resourceOptions = MTLResourceCPUCacheModeDefaultCache;
+    } else
+    {
+        if (isWriteOnly) {
+            resourceOptions = MTLResourceCPUCacheModeWriteCombined;
+        } else {
+            resourceOptions = MTLResourceCPUCacheModeDefaultCache;
+        }
     }
 
     container->buffers[0] = METAL_INTERNAL_CreateBuffer(
@@ -1395,20 +1401,21 @@ static void METAL_INTERNAL_CycleActiveTexture(
     return (SDL_GpuBuffer *)METAL_INTERNAL_CreateBufferContainer(
         (MetalRenderer *)driverData,
         sizeInBytes,
-        0);
+        SDL_TRUE,
+        SDL_FALSE);
 }
 
 static SDL_GpuTransferBuffer *METAL_CreateTransferBuffer(
     SDL_GpuRenderer *driverData,
-    SDL_GpuTransferUsage usage,
-    SDL_GpuTransferBufferMapFlags mapFlags,
+    SDL_bool uploadOnly,
     Uint32 sizeInBytes)
 {
     (void)usage;
     return (SDL_GpuTransferBuffer *)METAL_INTERNAL_CreateBufferContainer(
         (MetalRenderer *)driverData,
         sizeInBytes,
-        mapFlags);
+        SDL_FALSE,
+        uploadOnly);
 }
 
 static MetalUniformBuffer *METAL_INTERNAL_CreateUniformBuffer(
@@ -1419,7 +1426,8 @@ static void METAL_INTERNAL_CycleActiveTexture(
     uniformBuffer->container = METAL_INTERNAL_CreateBufferContainer(
         renderer,
         sizeInBytes,
-        SDL_GPU_TRANSFER_MAP_WRITE);
+        SDL_FALSE,
+        SDL_TRUE);
     uniformBuffer->offset = 0;
     return uniformBuffer;
 }

src/gpu/vulkan/SDL_gpu_vulkan.c

@@ -6765,8 +6765,7 @@ static VulkanUniformBuffer *VULKAN_INTERNAL_CreateUniformBuffer(
 
 static SDL_GpuTransferBuffer *VULKAN_CreateTransferBuffer(
     SDL_GpuRenderer *driverData,
-    SDL_GpuTransferUsage usage,             /* ignored on Vulkan */
-    SDL_GpuTransferBufferMapFlags mapFlags, /* ignored on Vulkan */
+    SDL_bool uploadOnly, /* ignored on Vulkan */
     Uint32 sizeInBytes)
 {
     return (SDL_GpuTransferBuffer *)VULKAN_INTERNAL_CreateBufferContainer(

test/testgpu_spinning_cube.c

@@ -455,8 +455,7 @@ init_render_state(void)
 
     buf_transfer = SDL_GpuCreateTransferBuffer(
         gpu_device,
-        SDL_GPU_TRANSFERUSAGE_BUFFER,
-        SDL_GPU_TRANSFER_MAP_WRITE,
+        SDL_TRUE,
         sizeof(vertex_data)
     );
     CHECK_CREATE(buf_transfer, "Vertex transfer buffer")