1 | /* |
2 | Simple DirectMedia Layer |
3 | Copyright (C) 1997-2025 Sam Lantinga <slouken@libsdl.org> |
4 | |
5 | This software is provided 'as-is', without any express or implied |
6 | warranty. In no event will the authors be held liable for any damages |
7 | arising from the use of this software. |
8 | |
9 | Permission is granted to anyone to use this software for any purpose, |
10 | including commercial applications, and to alter it and redistribute it |
11 | freely, subject to the following restrictions: |
12 | |
13 | 1. The origin of this software must not be misrepresented; you must not |
14 | claim that you wrote the original software. If you use this software |
15 | in a product, an acknowledgment in the product documentation would be |
16 | appreciated but is not required. |
17 | 2. Altered source versions must be plainly marked as such, and must not be |
18 | misrepresented as being the original software. |
19 | 3. This notice may not be removed or altered from any source distribution. |
20 | */ |
21 | #include "SDL_internal.h" |
22 | |
23 | #ifdef SDL_VIDEO_RENDER_GPU |
24 | |
25 | #include "../../video/SDL_pixels_c.h" |
26 | #include "../SDL_d3dmath.h" |
27 | #include "../SDL_sysrender.h" |
28 | #include "SDL_gpu_util.h" |
29 | #include "SDL_pipeline_gpu.h" |
30 | #include "SDL_shaders_gpu.h" |
31 | |
32 | typedef struct GPU_VertexShaderUniformData |
33 | { |
34 | Float4X4 mvp; |
35 | SDL_FColor color; |
36 | } GPU_VertexShaderUniformData; |
37 | |
38 | typedef struct GPU_FragmentShaderUniformData |
39 | { |
40 | float texel_width; |
41 | float texel_height; |
42 | float texture_width; |
43 | float texture_height; |
44 | } GPU_FragmentShaderUniformData; |
45 | |
46 | typedef struct GPU_RenderData |
47 | { |
48 | SDL_GPUDevice *device; |
49 | GPU_Shaders shaders; |
50 | GPU_PipelineCache pipeline_cache; |
51 | |
52 | struct |
53 | { |
54 | SDL_GPUTexture *texture; |
55 | SDL_GPUTextureFormat format; |
56 | Uint32 width; |
57 | Uint32 height; |
58 | } backbuffer; |
59 | |
60 | struct |
61 | { |
62 | SDL_GPUSwapchainComposition composition; |
63 | SDL_GPUPresentMode present_mode; |
64 | } swapchain; |
65 | |
66 | struct |
67 | { |
68 | SDL_GPUTransferBuffer *transfer_buf; |
69 | SDL_GPUBuffer *buffer; |
70 | Uint32 buffer_size; |
71 | } vertices; |
72 | |
73 | struct |
74 | { |
75 | SDL_GPURenderPass *render_pass; |
76 | SDL_Texture *render_target; |
77 | SDL_GPUCommandBuffer *command_buffer; |
78 | SDL_GPUColorTargetInfo color_attachment; |
79 | SDL_GPUViewport viewport; |
80 | SDL_Rect scissor; |
81 | SDL_FColor draw_color; |
82 | bool scissor_enabled; |
83 | bool scissor_was_enabled; |
84 | } state; |
85 | |
86 | SDL_GPUSampler *samplers[3][2]; |
87 | } GPU_RenderData; |
88 | |
89 | typedef struct GPU_TextureData |
90 | { |
91 | SDL_GPUTexture *texture; |
92 | SDL_GPUTextureFormat format; |
93 | GPU_FragmentShaderID shader; |
94 | void *pixels; |
95 | int pitch; |
96 | SDL_Rect locked_rect; |
97 | } GPU_TextureData; |
98 | |
99 | static bool GPU_SupportsBlendMode(SDL_Renderer *renderer, SDL_BlendMode blendMode) |
100 | { |
101 | SDL_BlendFactor srcColorFactor = SDL_GetBlendModeSrcColorFactor(blendMode); |
102 | SDL_BlendFactor srcAlphaFactor = SDL_GetBlendModeSrcAlphaFactor(blendMode); |
103 | SDL_BlendOperation colorOperation = SDL_GetBlendModeColorOperation(blendMode); |
104 | SDL_BlendFactor dstColorFactor = SDL_GetBlendModeDstColorFactor(blendMode); |
105 | SDL_BlendFactor dstAlphaFactor = SDL_GetBlendModeDstAlphaFactor(blendMode); |
106 | SDL_BlendOperation alphaOperation = SDL_GetBlendModeAlphaOperation(blendMode); |
107 | |
108 | if (GPU_ConvertBlendFactor(srcColorFactor) == SDL_GPU_BLENDFACTOR_INVALID || |
109 | GPU_ConvertBlendFactor(srcAlphaFactor) == SDL_GPU_BLENDFACTOR_INVALID || |
110 | GPU_ConvertBlendOperation(colorOperation) == SDL_GPU_BLENDOP_INVALID || |
111 | GPU_ConvertBlendFactor(dstColorFactor) == SDL_GPU_BLENDFACTOR_INVALID || |
112 | GPU_ConvertBlendFactor(dstAlphaFactor) == SDL_GPU_BLENDFACTOR_INVALID || |
113 | GPU_ConvertBlendOperation(alphaOperation) == SDL_GPU_BLENDOP_INVALID) { |
114 | return false; |
115 | } |
116 | |
117 | return true; |
118 | } |
119 | |
120 | static SDL_GPUTextureFormat PixFormatToTexFormat(SDL_PixelFormat pixel_format) |
121 | { |
122 | switch (pixel_format) { |
123 | case SDL_PIXELFORMAT_BGRA32: |
124 | case SDL_PIXELFORMAT_BGRX32: |
125 | return SDL_GPU_TEXTUREFORMAT_B8G8R8A8_UNORM; |
126 | case SDL_PIXELFORMAT_RGBA32: |
127 | case SDL_PIXELFORMAT_RGBX32: |
128 | return SDL_GPU_TEXTUREFORMAT_R8G8B8A8_UNORM; |
129 | |
130 | // YUV TODO |
131 | case SDL_PIXELFORMAT_YV12: |
132 | case SDL_PIXELFORMAT_IYUV: |
133 | case SDL_PIXELFORMAT_NV12: |
134 | case SDL_PIXELFORMAT_NV21: |
135 | case SDL_PIXELFORMAT_UYVY: |
136 | default: |
137 | return SDL_GPU_TEXTUREFORMAT_INVALID; |
138 | } |
139 | } |
140 | |
141 | static SDL_PixelFormat TexFormatToPixFormat(SDL_GPUTextureFormat tex_format) |
142 | { |
143 | switch (tex_format) { |
144 | case SDL_GPU_TEXTUREFORMAT_R8G8B8A8_UNORM: |
145 | return SDL_PIXELFORMAT_RGBA32; |
146 | case SDL_GPU_TEXTUREFORMAT_B8G8R8A8_UNORM: |
147 | return SDL_PIXELFORMAT_BGRA32; |
148 | case SDL_GPU_TEXTUREFORMAT_B5G6R5_UNORM: |
149 | return SDL_PIXELFORMAT_BGR565; |
150 | case SDL_GPU_TEXTUREFORMAT_B5G5R5A1_UNORM: |
151 | return SDL_PIXELFORMAT_BGRA5551; |
152 | case SDL_GPU_TEXTUREFORMAT_B4G4R4A4_UNORM: |
153 | return SDL_PIXELFORMAT_BGRA4444; |
154 | case SDL_GPU_TEXTUREFORMAT_R10G10B10A2_UNORM: |
155 | return SDL_PIXELFORMAT_ABGR2101010; |
156 | case SDL_GPU_TEXTUREFORMAT_R16G16B16A16_UNORM: |
157 | return SDL_PIXELFORMAT_RGBA64; |
158 | case SDL_GPU_TEXTUREFORMAT_R8G8B8A8_SNORM: |
159 | return SDL_PIXELFORMAT_RGBA32; |
160 | case SDL_GPU_TEXTUREFORMAT_R16G16B16A16_FLOAT: |
161 | return SDL_PIXELFORMAT_RGBA64_FLOAT; |
162 | case SDL_GPU_TEXTUREFORMAT_R32G32B32A32_FLOAT: |
163 | return SDL_PIXELFORMAT_RGBA128_FLOAT; |
164 | case SDL_GPU_TEXTUREFORMAT_R8G8B8A8_UINT: |
165 | return SDL_PIXELFORMAT_RGBA32; |
166 | case SDL_GPU_TEXTUREFORMAT_R16G16B16A16_UINT: |
167 | return SDL_PIXELFORMAT_RGBA64; |
168 | case SDL_GPU_TEXTUREFORMAT_R8G8B8A8_UNORM_SRGB: |
169 | return SDL_PIXELFORMAT_RGBA32; |
170 | case SDL_GPU_TEXTUREFORMAT_B8G8R8A8_UNORM_SRGB: |
171 | return SDL_PIXELFORMAT_BGRA32; |
172 | default: |
173 | return SDL_PIXELFORMAT_UNKNOWN; |
174 | } |
175 | } |
176 | |
177 | static bool GPU_CreateTexture(SDL_Renderer *renderer, SDL_Texture *texture, SDL_PropertiesID create_props) |
178 | { |
179 | GPU_RenderData *renderdata = (GPU_RenderData *)renderer->internal; |
180 | GPU_TextureData *data; |
181 | SDL_GPUTextureFormat format; |
182 | SDL_GPUTextureUsageFlags usage = SDL_GPU_TEXTUREUSAGE_SAMPLER; |
183 | |
184 | format = PixFormatToTexFormat(texture->format); |
185 | |
186 | if (format == SDL_GPU_TEXTUREFORMAT_INVALID) { |
187 | return SDL_SetError("Texture format %s not supported by SDL_GPU" , |
188 | SDL_GetPixelFormatName(texture->format)); |
189 | } |
190 | |
191 | data = (GPU_TextureData *)SDL_calloc(1, sizeof(*data)); |
192 | if (!data) { |
193 | return false; |
194 | } |
195 | |
196 | if (texture->access == SDL_TEXTUREACCESS_STREAMING) { |
197 | size_t size; |
198 | data->pitch = texture->w * SDL_BYTESPERPIXEL(texture->format); |
199 | size = (size_t)texture->h * data->pitch; |
200 | if (texture->format == SDL_PIXELFORMAT_YV12 || |
201 | texture->format == SDL_PIXELFORMAT_IYUV) { |
202 | // Need to add size for the U and V planes |
203 | size += 2 * ((texture->h + 1) / 2) * ((data->pitch + 1) / 2); |
204 | } |
205 | if (texture->format == SDL_PIXELFORMAT_NV12 || |
206 | texture->format == SDL_PIXELFORMAT_NV21) { |
207 | // Need to add size for the U/V plane |
208 | size += 2 * ((texture->h + 1) / 2) * ((data->pitch + 1) / 2); |
209 | } |
210 | data->pixels = SDL_calloc(1, size); |
211 | if (!data->pixels) { |
212 | SDL_free(data); |
213 | return false; |
214 | } |
215 | |
216 | // TODO allocate a persistent transfer buffer |
217 | } |
218 | |
219 | if (texture->access == SDL_TEXTUREACCESS_TARGET) { |
220 | usage |= SDL_GPU_TEXTUREUSAGE_COLOR_TARGET; |
221 | } |
222 | |
223 | texture->internal = data; |
224 | SDL_GPUTextureCreateInfo tci; |
225 | SDL_zero(tci); |
226 | tci.format = format; |
227 | tci.layer_count_or_depth = 1; |
228 | tci.num_levels = 1; |
229 | tci.usage = usage; |
230 | tci.width = texture->w; |
231 | tci.height = texture->h; |
232 | tci.sample_count = SDL_GPU_SAMPLECOUNT_1; |
233 | |
234 | data->format = format; |
235 | data->texture = SDL_CreateGPUTexture(renderdata->device, &tci); |
236 | |
237 | if (!data->texture) { |
238 | return false; |
239 | } |
240 | |
241 | if (texture->format == SDL_PIXELFORMAT_RGBA32 || texture->format == SDL_PIXELFORMAT_BGRA32) { |
242 | data->shader = FRAG_SHADER_TEXTURE_RGBA; |
243 | } else { |
244 | data->shader = FRAG_SHADER_TEXTURE_RGB; |
245 | } |
246 | |
247 | return true; |
248 | } |
249 | |
250 | static bool GPU_UpdateTexture(SDL_Renderer *renderer, SDL_Texture *texture, |
251 | const SDL_Rect *rect, const void *pixels, int pitch) |
252 | { |
253 | GPU_RenderData *renderdata = (GPU_RenderData *)renderer->internal; |
254 | GPU_TextureData *data = (GPU_TextureData *)texture->internal; |
255 | const Uint32 texturebpp = SDL_BYTESPERPIXEL(texture->format); |
256 | |
257 | size_t row_size, data_size; |
258 | |
259 | if (!SDL_size_mul_check_overflow(rect->w, texturebpp, &row_size) || |
260 | !SDL_size_mul_check_overflow(rect->h, row_size, &data_size)) { |
261 | return SDL_SetError("update size overflow" ); |
262 | } |
263 | |
264 | SDL_GPUTransferBufferCreateInfo tbci; |
265 | SDL_zero(tbci); |
266 | tbci.size = (Uint32)data_size; |
267 | tbci.usage = SDL_GPU_TRANSFERBUFFERUSAGE_UPLOAD; |
268 | |
269 | SDL_GPUTransferBuffer *tbuf = SDL_CreateGPUTransferBuffer(renderdata->device, &tbci); |
270 | |
271 | if (tbuf == NULL) { |
272 | return false; |
273 | } |
274 | |
275 | Uint8 *output = SDL_MapGPUTransferBuffer(renderdata->device, tbuf, false); |
276 | |
277 | if ((size_t)pitch == row_size) { |
278 | SDL_memcpy(output, pixels, data_size); |
279 | } else { |
280 | // FIXME is negative pitch supposed to work? |
281 | // If not, maybe use SDL_GPUTextureTransferInfo::pixels_per_row instead of this |
282 | const Uint8 *input = pixels; |
283 | |
284 | for (int i = 0; i < rect->h; ++i) { |
285 | SDL_memcpy(output, input, row_size); |
286 | output += row_size; |
287 | input += pitch; |
288 | } |
289 | } |
290 | |
291 | SDL_UnmapGPUTransferBuffer(renderdata->device, tbuf); |
292 | |
293 | SDL_GPUCommandBuffer *cbuf = renderdata->state.command_buffer; |
294 | SDL_GPUCopyPass *cpass = SDL_BeginGPUCopyPass(cbuf); |
295 | |
296 | SDL_GPUTextureTransferInfo tex_src; |
297 | SDL_zero(tex_src); |
298 | tex_src.transfer_buffer = tbuf; |
299 | tex_src.rows_per_layer = rect->h; |
300 | tex_src.pixels_per_row = rect->w; |
301 | |
302 | SDL_GPUTextureRegion tex_dst; |
303 | SDL_zero(tex_dst); |
304 | tex_dst.texture = data->texture; |
305 | tex_dst.x = rect->x; |
306 | tex_dst.y = rect->y; |
307 | tex_dst.w = rect->w; |
308 | tex_dst.h = rect->h; |
309 | tex_dst.d = 1; |
310 | |
311 | SDL_UploadToGPUTexture(cpass, &tex_src, &tex_dst, false); |
312 | SDL_EndGPUCopyPass(cpass); |
313 | SDL_ReleaseGPUTransferBuffer(renderdata->device, tbuf); |
314 | |
315 | return true; |
316 | } |
317 | |
318 | static bool GPU_LockTexture(SDL_Renderer *renderer, SDL_Texture *texture, |
319 | const SDL_Rect *rect, void **pixels, int *pitch) |
320 | { |
321 | GPU_TextureData *data = (GPU_TextureData *)texture->internal; |
322 | |
323 | data->locked_rect = *rect; |
324 | *pixels = |
325 | (void *)((Uint8 *)data->pixels + rect->y * data->pitch + |
326 | rect->x * SDL_BYTESPERPIXEL(texture->format)); |
327 | *pitch = data->pitch; |
328 | return true; |
329 | } |
330 | |
331 | static void GPU_UnlockTexture(SDL_Renderer *renderer, SDL_Texture *texture) |
332 | { |
333 | GPU_TextureData *data = (GPU_TextureData *)texture->internal; |
334 | const SDL_Rect *rect; |
335 | void *pixels; |
336 | |
337 | rect = &data->locked_rect; |
338 | pixels = |
339 | (void *)((Uint8 *)data->pixels + rect->y * data->pitch + |
340 | rect->x * SDL_BYTESPERPIXEL(texture->format)); |
341 | GPU_UpdateTexture(renderer, texture, rect, pixels, data->pitch); |
342 | } |
343 | |
344 | static bool GPU_SetRenderTarget(SDL_Renderer *renderer, SDL_Texture *texture) |
345 | { |
346 | GPU_RenderData *data = (GPU_RenderData *)renderer->internal; |
347 | |
348 | data->state.render_target = texture; |
349 | |
350 | return true; |
351 | } |
352 | |
353 | static bool GPU_QueueNoOp(SDL_Renderer *renderer, SDL_RenderCommand *cmd) |
354 | { |
355 | return true; // nothing to do in this backend. |
356 | } |
357 | |
358 | static SDL_FColor GetDrawCmdColor(SDL_Renderer *renderer, SDL_RenderCommand *cmd) |
359 | { |
360 | SDL_FColor color = cmd->data.color.color; |
361 | |
362 | if (SDL_RenderingLinearSpace(renderer)) { |
363 | SDL_ConvertToLinear(&color); |
364 | } |
365 | |
366 | color.r *= cmd->data.color.color_scale; |
367 | color.g *= cmd->data.color.color_scale; |
368 | color.b *= cmd->data.color.color_scale; |
369 | |
370 | return color; |
371 | } |
372 | |
373 | static bool GPU_QueueDrawPoints(SDL_Renderer *renderer, SDL_RenderCommand *cmd, const SDL_FPoint *points, int count) |
374 | { |
375 | float *verts = (float *)SDL_AllocateRenderVertices(renderer, count * 2 * sizeof(float), 0, &cmd->data.draw.first); |
376 | |
377 | if (!verts) { |
378 | return false; |
379 | } |
380 | |
381 | cmd->data.draw.count = count; |
382 | for (int i = 0; i < count; i++) { |
383 | *(verts++) = 0.5f + points[i].x; |
384 | *(verts++) = 0.5f + points[i].y; |
385 | } |
386 | |
387 | return true; |
388 | } |
389 | |
390 | static bool GPU_QueueGeometry(SDL_Renderer *renderer, SDL_RenderCommand *cmd, SDL_Texture *texture, |
391 | const float *xy, int xy_stride, const SDL_FColor *color, int color_stride, const float *uv, int uv_stride, |
392 | int num_vertices, const void *indices, int num_indices, int size_indices, |
393 | float scale_x, float scale_y) |
394 | { |
395 | int i; |
396 | int count = indices ? num_indices : num_vertices; |
397 | float *verts; |
398 | size_t sz = 2 * sizeof(float) + 4 * sizeof(float) + (texture ? 2 : 0) * sizeof(float); |
399 | const float color_scale = cmd->data.draw.color_scale; |
400 | bool convert_color = SDL_RenderingLinearSpace(renderer); |
401 | |
402 | verts = (float *)SDL_AllocateRenderVertices(renderer, count * sz, 0, &cmd->data.draw.first); |
403 | if (!verts) { |
404 | return false; |
405 | } |
406 | |
407 | cmd->data.draw.count = count; |
408 | size_indices = indices ? size_indices : 0; |
409 | |
410 | for (i = 0; i < count; i++) { |
411 | int j; |
412 | float *xy_; |
413 | SDL_FColor col_; |
414 | if (size_indices == 4) { |
415 | j = ((const Uint32 *)indices)[i]; |
416 | } else if (size_indices == 2) { |
417 | j = ((const Uint16 *)indices)[i]; |
418 | } else if (size_indices == 1) { |
419 | j = ((const Uint8 *)indices)[i]; |
420 | } else { |
421 | j = i; |
422 | } |
423 | |
424 | xy_ = (float *)((char *)xy + j * xy_stride); |
425 | |
426 | *(verts++) = xy_[0] * scale_x; |
427 | *(verts++) = xy_[1] * scale_y; |
428 | |
429 | col_ = *(SDL_FColor *)((char *)color + j * color_stride); |
430 | if (convert_color) { |
431 | SDL_ConvertToLinear(&col_); |
432 | } |
433 | |
434 | // FIXME: The Vulkan backend doesn't multiply by color_scale. GL does. I'm not sure which one is wrong. |
435 | // ANSWER: The color scale should be applied in linear space when using the scRGB colorspace. This is done in shaders in the Vulkan backend. |
436 | *(verts++) = col_.r * color_scale; |
437 | *(verts++) = col_.g * color_scale; |
438 | *(verts++) = col_.b * color_scale; |
439 | *(verts++) = col_.a; |
440 | |
441 | if (texture) { |
442 | float *uv_ = (float *)((char *)uv + j * uv_stride); |
443 | *(verts++) = uv_[0]; |
444 | *(verts++) = uv_[1]; |
445 | } |
446 | } |
447 | return true; |
448 | } |
449 | |
450 | static void GPU_InvalidateCachedState(SDL_Renderer *renderer) |
451 | { |
452 | GPU_RenderData *data = (GPU_RenderData *)renderer->internal; |
453 | |
454 | data->state.render_target = NULL; |
455 | data->state.scissor_enabled = false; |
456 | } |
457 | |
458 | static SDL_GPURenderPass *RestartRenderPass(GPU_RenderData *data) |
459 | { |
460 | if (data->state.render_pass) { |
461 | SDL_EndGPURenderPass(data->state.render_pass); |
462 | } |
463 | |
464 | data->state.render_pass = SDL_BeginGPURenderPass( |
465 | data->state.command_buffer, &data->state.color_attachment, 1, NULL); |
466 | |
467 | // *** FIXME *** |
468 | // This is busted. We should be able to know which load op to use. |
469 | // LOAD is incorrect behavior most of the time, unless we had to break a render pass. |
470 | // -cosmonaut |
471 | data->state.color_attachment.load_op = SDL_GPU_LOADOP_LOAD; |
472 | data->state.scissor_was_enabled = false; |
473 | |
474 | return data->state.render_pass; |
475 | } |
476 | |
477 | static void PushVertexUniforms(GPU_RenderData *data, SDL_RenderCommand *cmd) |
478 | { |
479 | GPU_VertexShaderUniformData uniforms; |
480 | SDL_zero(uniforms); |
481 | uniforms.mvp.m[0][0] = 2.0f / data->state.viewport.w; |
482 | uniforms.mvp.m[1][1] = -2.0f / data->state.viewport.h; |
483 | uniforms.mvp.m[2][2] = 1.0f; |
484 | uniforms.mvp.m[3][0] = -1.0f; |
485 | uniforms.mvp.m[3][1] = 1.0f; |
486 | uniforms.mvp.m[3][3] = 1.0f; |
487 | |
488 | uniforms.color = data->state.draw_color; |
489 | |
490 | SDL_PushGPUVertexUniformData(data->state.command_buffer, 0, &uniforms, sizeof(uniforms)); |
491 | } |
492 | |
493 | static void PushFragmentUniforms(GPU_RenderData *data, SDL_RenderCommand *cmd) |
494 | { |
495 | if (cmd->data.draw.texture && |
496 | cmd->data.draw.texture_scale_mode == SDL_SCALEMODE_PIXELART) { |
497 | SDL_Texture *texture = cmd->data.draw.texture; |
498 | GPU_FragmentShaderUniformData uniforms; |
499 | SDL_zero(uniforms); |
500 | uniforms.texture_width = texture->w; |
501 | uniforms.texture_height = texture->h; |
502 | uniforms.texel_width = 1.0f / uniforms.texture_width; |
503 | uniforms.texel_height = 1.0f / uniforms.texture_height; |
504 | SDL_PushGPUFragmentUniformData(data->state.command_buffer, 0, &uniforms, sizeof(uniforms)); |
505 | } |
506 | } |
507 | |
508 | static SDL_GPUSampler **SamplerPointer(GPU_RenderData *data, SDL_TextureAddressMode address_mode, SDL_ScaleMode scale_mode) |
509 | { |
510 | SDL_assert(scale_mode < SDL_arraysize(data->samplers)); |
511 | SDL_assert((address_mode - 1) < SDL_arraysize(data->samplers[0])); |
512 | return &data->samplers[scale_mode][address_mode - 1]; |
513 | } |
514 | |
515 | static void SetViewportAndScissor(GPU_RenderData *data) |
516 | { |
517 | SDL_SetGPUViewport(data->state.render_pass, &data->state.viewport); |
518 | |
519 | if (data->state.scissor_enabled) { |
520 | SDL_SetGPUScissor(data->state.render_pass, &data->state.scissor); |
521 | data->state.scissor_was_enabled = true; |
522 | } else if (data->state.scissor_was_enabled) { |
523 | SDL_Rect r; |
524 | r.x = (int)data->state.viewport.x; |
525 | r.y = (int)data->state.viewport.y; |
526 | r.w = (int)data->state.viewport.w; |
527 | r.h = (int)data->state.viewport.h; |
528 | SDL_SetGPUScissor(data->state.render_pass, &r); |
529 | data->state.scissor_was_enabled = false; |
530 | } |
531 | } |
532 | |
533 | static void Draw( |
534 | GPU_RenderData *data, SDL_RenderCommand *cmd, |
535 | Uint32 num_verts, |
536 | Uint32 offset, |
537 | SDL_GPUPrimitiveType prim) |
538 | { |
539 | if (!data->state.render_pass || data->state.color_attachment.load_op == SDL_GPU_LOADOP_CLEAR) { |
540 | RestartRenderPass(data); |
541 | } |
542 | |
543 | SDL_GPURenderPass *pass = data->state.render_pass; |
544 | SDL_GPURenderState *custom_state = cmd->data.draw.gpu_render_state; |
545 | SDL_GPUShader *custom_frag_shader = custom_state ? custom_state->fragment_shader : NULL; |
546 | GPU_VertexShaderID v_shader; |
547 | GPU_FragmentShaderID f_shader; |
548 | |
549 | if (prim == SDL_GPU_PRIMITIVETYPE_TRIANGLELIST) { |
550 | SDL_Texture *texture = cmd->data.draw.texture; |
551 | if (texture) { |
552 | v_shader = VERT_SHADER_TRI_TEXTURE; |
553 | if (texture->format == SDL_PIXELFORMAT_RGBA32 || texture->format == SDL_PIXELFORMAT_BGRA32) { |
554 | if (cmd->data.draw.texture_scale_mode == SDL_SCALEMODE_PIXELART) { |
555 | f_shader = FRAG_SHADER_TEXTURE_RGBA_PIXELART; |
556 | } else { |
557 | f_shader = FRAG_SHADER_TEXTURE_RGBA; |
558 | } |
559 | } else { |
560 | if (cmd->data.draw.texture_scale_mode == SDL_SCALEMODE_PIXELART) { |
561 | f_shader = FRAG_SHADER_TEXTURE_RGB_PIXELART; |
562 | } else { |
563 | f_shader = FRAG_SHADER_TEXTURE_RGB; |
564 | } |
565 | } |
566 | } else { |
567 | v_shader = VERT_SHADER_TRI_COLOR; |
568 | f_shader = FRAG_SHADER_COLOR; |
569 | } |
570 | } else { |
571 | v_shader = VERT_SHADER_LINEPOINT; |
572 | f_shader = FRAG_SHADER_COLOR; |
573 | } |
574 | |
575 | if (custom_frag_shader) { |
576 | f_shader = FRAG_SHADER_TEXTURE_CUSTOM; |
577 | data->shaders.frag_shaders[FRAG_SHADER_TEXTURE_CUSTOM] = custom_frag_shader; |
578 | } |
579 | |
580 | GPU_PipelineParameters pipe_params; |
581 | SDL_zero(pipe_params); |
582 | pipe_params.blend_mode = cmd->data.draw.blend; |
583 | pipe_params.vert_shader = v_shader; |
584 | pipe_params.frag_shader = f_shader; |
585 | pipe_params.primitive_type = prim; |
586 | pipe_params.custom_frag_shader = custom_frag_shader; |
587 | |
588 | if (data->state.render_target) { |
589 | pipe_params.attachment_format = ((GPU_TextureData *)data->state.render_target->internal)->format; |
590 | } else { |
591 | pipe_params.attachment_format = data->backbuffer.format; |
592 | } |
593 | |
594 | SDL_GPUGraphicsPipeline *pipe = GPU_GetPipeline(&data->pipeline_cache, &data->shaders, data->device, &pipe_params); |
595 | if (!pipe) { |
596 | return; |
597 | } |
598 | |
599 | SDL_BindGPUGraphicsPipeline(pass, pipe); |
600 | |
601 | Uint32 sampler_slot = 0; |
602 | if (cmd->data.draw.texture) { |
603 | GPU_TextureData *tdata = (GPU_TextureData *)cmd->data.draw.texture->internal; |
604 | SDL_GPUTextureSamplerBinding sampler_bind; |
605 | SDL_zero(sampler_bind); |
606 | sampler_bind.sampler = *SamplerPointer(data, cmd->data.draw.texture_address_mode, cmd->data.draw.texture_scale_mode); |
607 | sampler_bind.texture = tdata->texture; |
608 | SDL_BindGPUFragmentSamplers(pass, sampler_slot++, &sampler_bind, 1); |
609 | } |
610 | if (custom_state) { |
611 | if (custom_state->num_sampler_bindings > 0) { |
612 | SDL_BindGPUFragmentSamplers(pass, sampler_slot, custom_state->sampler_bindings, custom_state->num_sampler_bindings); |
613 | } |
614 | if (custom_state->num_storage_textures > 0) { |
615 | SDL_BindGPUFragmentStorageTextures(pass, 0, custom_state->storage_textures, custom_state->num_storage_textures); |
616 | } |
617 | if (custom_state->num_storage_buffers > 0) { |
618 | SDL_BindGPUFragmentStorageBuffers(pass, 0, custom_state->storage_buffers, custom_state->num_storage_buffers); |
619 | } |
620 | if (custom_state->num_uniform_buffers > 0) { |
621 | for (int i = 0; i < custom_state->num_uniform_buffers; i++) { |
622 | SDL_GPURenderStateUniformBuffer *ub = &custom_state->uniform_buffers[i]; |
623 | SDL_PushGPUFragmentUniformData(data->state.command_buffer, ub->slot_index, ub->data, ub->length); |
624 | } |
625 | } |
626 | } else { |
627 | PushFragmentUniforms(data, cmd); |
628 | } |
629 | |
630 | SDL_GPUBufferBinding buffer_bind; |
631 | SDL_zero(buffer_bind); |
632 | buffer_bind.buffer = data->vertices.buffer; |
633 | buffer_bind.offset = offset; |
634 | SDL_BindGPUVertexBuffers(pass, 0, &buffer_bind, 1); |
635 | PushVertexUniforms(data, cmd); |
636 | |
637 | SetViewportAndScissor(data); |
638 | |
639 | SDL_DrawGPUPrimitives(pass, num_verts, 1, 0, 0); |
640 | } |
641 | |
642 | static void ReleaseVertexBuffer(GPU_RenderData *data) |
643 | { |
644 | if (data->vertices.buffer) { |
645 | SDL_ReleaseGPUBuffer(data->device, data->vertices.buffer); |
646 | } |
647 | |
648 | if (data->vertices.transfer_buf) { |
649 | SDL_ReleaseGPUTransferBuffer(data->device, data->vertices.transfer_buf); |
650 | } |
651 | |
652 | data->vertices.buffer_size = 0; |
653 | } |
654 | |
655 | static bool InitVertexBuffer(GPU_RenderData *data, Uint32 size) |
656 | { |
657 | SDL_GPUBufferCreateInfo bci; |
658 | SDL_zero(bci); |
659 | bci.size = size; |
660 | bci.usage = SDL_GPU_BUFFERUSAGE_VERTEX; |
661 | |
662 | data->vertices.buffer = SDL_CreateGPUBuffer(data->device, &bci); |
663 | |
664 | if (!data->vertices.buffer) { |
665 | return false; |
666 | } |
667 | |
668 | SDL_GPUTransferBufferCreateInfo tbci; |
669 | SDL_zero(tbci); |
670 | tbci.size = size; |
671 | tbci.usage = SDL_GPU_TRANSFERBUFFERUSAGE_UPLOAD; |
672 | |
673 | data->vertices.transfer_buf = SDL_CreateGPUTransferBuffer(data->device, &tbci); |
674 | |
675 | if (!data->vertices.transfer_buf) { |
676 | return false; |
677 | } |
678 | |
679 | return true; |
680 | } |
681 | |
682 | static bool UploadVertices(GPU_RenderData *data, void *vertices, size_t vertsize) |
683 | { |
684 | if (vertsize == 0) { |
685 | return true; |
686 | } |
687 | |
688 | if (vertsize > data->vertices.buffer_size) { |
689 | ReleaseVertexBuffer(data); |
690 | if (!InitVertexBuffer(data, (Uint32)vertsize)) { |
691 | return false; |
692 | } |
693 | } |
694 | |
695 | void *staging_buf = SDL_MapGPUTransferBuffer(data->device, data->vertices.transfer_buf, true); |
696 | SDL_memcpy(staging_buf, vertices, vertsize); |
697 | SDL_UnmapGPUTransferBuffer(data->device, data->vertices.transfer_buf); |
698 | |
699 | SDL_GPUCopyPass *pass = SDL_BeginGPUCopyPass(data->state.command_buffer); |
700 | |
701 | if (!pass) { |
702 | return false; |
703 | } |
704 | |
705 | SDL_GPUTransferBufferLocation src; |
706 | SDL_zero(src); |
707 | src.transfer_buffer = data->vertices.transfer_buf; |
708 | |
709 | SDL_GPUBufferRegion dst; |
710 | SDL_zero(dst); |
711 | dst.buffer = data->vertices.buffer; |
712 | dst.size = (Uint32)vertsize; |
713 | |
714 | SDL_UploadToGPUBuffer(pass, &src, &dst, true); |
715 | SDL_EndGPUCopyPass(pass); |
716 | |
717 | return true; |
718 | } |
719 | |
720 | // *** FIXME *** |
721 | // We might be able to run these data uploads on a separate command buffer |
722 | // which would allow us to avoid breaking render passes. |
723 | // Honestly I'm a little skeptical of this entire approach, |
724 | // we already have a command buffer structure |
725 | // so it feels weird to be deferring the operations manually. |
726 | // We could also fairly easily run the geometry transformations |
727 | // on compute shaders instead of the CPU, which would be a HUGE performance win. |
728 | // -cosmonaut |
729 | static bool GPU_RunCommandQueue(SDL_Renderer *renderer, SDL_RenderCommand *cmd, void *vertices, size_t vertsize) |
730 | { |
731 | GPU_RenderData *data = (GPU_RenderData *)renderer->internal; |
732 | |
733 | if (!UploadVertices(data, vertices, vertsize)) { |
734 | return false; |
735 | } |
736 | |
737 | data->state.color_attachment.load_op = SDL_GPU_LOADOP_LOAD; |
738 | |
739 | if (renderer->target) { |
740 | GPU_TextureData *tdata = renderer->target->internal; |
741 | data->state.color_attachment.texture = tdata->texture; |
742 | } else { |
743 | data->state.color_attachment.texture = data->backbuffer.texture; |
744 | } |
745 | |
746 | if (!data->state.color_attachment.texture) { |
747 | return SDL_SetError("Render target texture is NULL" ); |
748 | } |
749 | |
750 | while (cmd) { |
751 | switch (cmd->command) { |
752 | case SDL_RENDERCMD_SETDRAWCOLOR: |
753 | { |
754 | data->state.draw_color = GetDrawCmdColor(renderer, cmd); |
755 | break; |
756 | } |
757 | |
758 | case SDL_RENDERCMD_SETVIEWPORT: |
759 | { |
760 | SDL_Rect *viewport = &cmd->data.viewport.rect; |
761 | data->state.viewport.x = viewport->x; |
762 | data->state.viewport.y = viewport->y; |
763 | data->state.viewport.w = viewport->w; |
764 | data->state.viewport.h = viewport->h; |
765 | break; |
766 | } |
767 | |
768 | case SDL_RENDERCMD_SETCLIPRECT: |
769 | { |
770 | const SDL_Rect *rect = &cmd->data.cliprect.rect; |
771 | data->state.scissor.x = (int)data->state.viewport.x + rect->x; |
772 | data->state.scissor.y = (int)data->state.viewport.y + rect->y; |
773 | data->state.scissor.w = rect->w; |
774 | data->state.scissor.h = rect->h; |
775 | data->state.scissor_enabled = cmd->data.cliprect.enabled; |
776 | break; |
777 | } |
778 | |
779 | case SDL_RENDERCMD_CLEAR: |
780 | { |
781 | data->state.color_attachment.clear_color = GetDrawCmdColor(renderer, cmd); |
782 | data->state.color_attachment.load_op = SDL_GPU_LOADOP_CLEAR; |
783 | break; |
784 | } |
785 | |
786 | case SDL_RENDERCMD_FILL_RECTS: // unused |
787 | break; |
788 | |
789 | case SDL_RENDERCMD_COPY: // unused |
790 | break; |
791 | |
792 | case SDL_RENDERCMD_COPY_EX: // unused |
793 | break; |
794 | |
795 | case SDL_RENDERCMD_DRAW_LINES: |
796 | { |
797 | Uint32 count = (Uint32)cmd->data.draw.count; |
798 | Uint32 offset = (Uint32)cmd->data.draw.first; |
799 | |
800 | if (count > 2) { |
801 | // joined lines cannot be grouped |
802 | Draw(data, cmd, count, offset, SDL_GPU_PRIMITIVETYPE_LINESTRIP); |
803 | } else { |
804 | // let's group non joined lines |
805 | SDL_RenderCommand *finalcmd = cmd; |
806 | SDL_RenderCommand *nextcmd = cmd->next; |
807 | SDL_BlendMode thisblend = cmd->data.draw.blend; |
808 | |
809 | while (nextcmd) { |
810 | const SDL_RenderCommandType nextcmdtype = nextcmd->command; |
811 | if (nextcmdtype != SDL_RENDERCMD_DRAW_LINES) { |
812 | break; // can't go any further on this draw call, different render command up next. |
813 | } else if (nextcmd->data.draw.count != 2) { |
814 | break; // can't go any further on this draw call, those are joined lines |
815 | } else if (nextcmd->data.draw.blend != thisblend) { |
816 | break; // can't go any further on this draw call, different blendmode copy up next. |
817 | } else { |
818 | finalcmd = nextcmd; // we can combine copy operations here. Mark this one as the furthest okay command. |
819 | count += (Uint32)nextcmd->data.draw.count; |
820 | } |
821 | nextcmd = nextcmd->next; |
822 | } |
823 | |
824 | Draw(data, cmd, count, offset, SDL_GPU_PRIMITIVETYPE_LINELIST); |
825 | cmd = finalcmd; // skip any copy commands we just combined in here. |
826 | } |
827 | break; |
828 | } |
829 | |
830 | case SDL_RENDERCMD_DRAW_POINTS: |
831 | case SDL_RENDERCMD_GEOMETRY: |
832 | { |
833 | /* as long as we have the same copy command in a row, with the |
834 | same texture, we can combine them all into a single draw call. */ |
835 | SDL_Texture *thistexture = cmd->data.draw.texture; |
836 | SDL_BlendMode thisblend = cmd->data.draw.blend; |
837 | SDL_ScaleMode thisscalemode = cmd->data.draw.texture_scale_mode; |
838 | SDL_TextureAddressMode thisaddressmode = cmd->data.draw.texture_address_mode; |
839 | const SDL_RenderCommandType thiscmdtype = cmd->command; |
840 | SDL_RenderCommand *finalcmd = cmd; |
841 | SDL_RenderCommand *nextcmd = cmd->next; |
842 | Uint32 count = (Uint32)cmd->data.draw.count; |
843 | Uint32 offset = (Uint32)cmd->data.draw.first; |
844 | |
845 | while (nextcmd) { |
846 | const SDL_RenderCommandType nextcmdtype = nextcmd->command; |
847 | if (nextcmdtype != thiscmdtype) { |
848 | break; // can't go any further on this draw call, different render command up next. |
849 | } else if (nextcmd->data.draw.texture != thistexture || |
850 | nextcmd->data.draw.texture_scale_mode != thisscalemode || |
851 | nextcmd->data.draw.texture_address_mode != thisaddressmode || |
852 | nextcmd->data.draw.blend != thisblend) { |
853 | // FIXME should we check address mode too? |
854 | break; // can't go any further on this draw call, different texture/blendmode copy up next. |
855 | } else { |
856 | finalcmd = nextcmd; // we can combine copy operations here. Mark this one as the furthest okay command. |
857 | count += (Uint32)nextcmd->data.draw.count; |
858 | } |
859 | nextcmd = nextcmd->next; |
860 | } |
861 | |
862 | SDL_GPUPrimitiveType prim = SDL_GPU_PRIMITIVETYPE_TRIANGLELIST; // SDL_RENDERCMD_GEOMETRY |
863 | if (thiscmdtype == SDL_RENDERCMD_DRAW_POINTS) { |
864 | prim = SDL_GPU_PRIMITIVETYPE_POINTLIST; |
865 | } |
866 | |
867 | Draw(data, cmd, count, offset, prim); |
868 | |
869 | cmd = finalcmd; // skip any copy commands we just combined in here. |
870 | break; |
871 | } |
872 | |
873 | case SDL_RENDERCMD_NO_OP: |
874 | break; |
875 | } |
876 | |
877 | cmd = cmd->next; |
878 | } |
879 | |
880 | if (data->state.color_attachment.load_op == SDL_GPU_LOADOP_CLEAR) { |
881 | RestartRenderPass(data); |
882 | } |
883 | |
884 | if (data->state.render_pass) { |
885 | SDL_EndGPURenderPass(data->state.render_pass); |
886 | data->state.render_pass = NULL; |
887 | } |
888 | |
889 | return true; |
890 | } |
891 | |
892 | static SDL_Surface *GPU_RenderReadPixels(SDL_Renderer *renderer, const SDL_Rect *rect) |
893 | { |
894 | GPU_RenderData *data = (GPU_RenderData *)renderer->internal; |
895 | SDL_GPUTexture *gpu_tex; |
896 | SDL_PixelFormat pixfmt; |
897 | |
898 | if (data->state.render_target) { |
899 | SDL_Texture *texture = data->state.render_target; |
900 | GPU_TextureData *texdata = texture->internal; |
901 | gpu_tex = texdata->texture; |
902 | pixfmt = texture->format; |
903 | } else { |
904 | gpu_tex = data->backbuffer.texture; |
905 | pixfmt = TexFormatToPixFormat(data->backbuffer.format); |
906 | |
907 | if (pixfmt == SDL_PIXELFORMAT_UNKNOWN) { |
908 | SDL_SetError("Unsupported backbuffer format" ); |
909 | return NULL; |
910 | } |
911 | } |
912 | |
913 | Uint32 bpp = SDL_BYTESPERPIXEL(pixfmt); |
914 | size_t row_size, image_size; |
915 | |
916 | if (!SDL_size_mul_check_overflow(rect->w, bpp, &row_size) || |
917 | !SDL_size_mul_check_overflow(rect->h, row_size, &image_size)) { |
918 | SDL_SetError("read size overflow" ); |
919 | return NULL; |
920 | } |
921 | |
922 | SDL_Surface *surface = SDL_CreateSurface(rect->w, rect->h, pixfmt); |
923 | |
924 | if (!surface) { |
925 | return NULL; |
926 | } |
927 | |
928 | SDL_GPUTransferBufferCreateInfo tbci; |
929 | SDL_zero(tbci); |
930 | tbci.size = (Uint32)image_size; |
931 | tbci.usage = SDL_GPU_TRANSFERBUFFERUSAGE_DOWNLOAD; |
932 | |
933 | SDL_GPUTransferBuffer *tbuf = SDL_CreateGPUTransferBuffer(data->device, &tbci); |
934 | |
935 | if (!tbuf) { |
936 | return NULL; |
937 | } |
938 | |
939 | SDL_GPUCopyPass *pass = SDL_BeginGPUCopyPass(data->state.command_buffer); |
940 | |
941 | SDL_GPUTextureRegion src; |
942 | SDL_zero(src); |
943 | src.texture = gpu_tex; |
944 | src.x = rect->x; |
945 | src.y = rect->y; |
946 | src.w = rect->w; |
947 | src.h = rect->h; |
948 | src.d = 1; |
949 | |
950 | SDL_GPUTextureTransferInfo dst; |
951 | SDL_zero(dst); |
952 | dst.transfer_buffer = tbuf; |
953 | dst.rows_per_layer = rect->h; |
954 | dst.pixels_per_row = rect->w; |
955 | |
956 | SDL_DownloadFromGPUTexture(pass, &src, &dst); |
957 | SDL_EndGPUCopyPass(pass); |
958 | |
959 | SDL_GPUFence *fence = SDL_SubmitGPUCommandBufferAndAcquireFence(data->state.command_buffer); |
960 | SDL_WaitForGPUFences(data->device, true, &fence, 1); |
961 | SDL_ReleaseGPUFence(data->device, fence); |
962 | data->state.command_buffer = SDL_AcquireGPUCommandBuffer(data->device); |
963 | |
964 | void *mapped_tbuf = SDL_MapGPUTransferBuffer(data->device, tbuf, false); |
965 | |
966 | if ((size_t)surface->pitch == row_size) { |
967 | SDL_memcpy(surface->pixels, mapped_tbuf, image_size); |
968 | } else { |
969 | Uint8 *input = mapped_tbuf; |
970 | Uint8 *output = surface->pixels; |
971 | |
972 | for (int row = 0; row < rect->h; ++row) { |
973 | SDL_memcpy(output, input, row_size); |
974 | output += surface->pitch; |
975 | input += row_size; |
976 | } |
977 | } |
978 | |
979 | SDL_UnmapGPUTransferBuffer(data->device, tbuf); |
980 | SDL_ReleaseGPUTransferBuffer(data->device, tbuf); |
981 | |
982 | return surface; |
983 | } |
984 | |
985 | static bool CreateBackbuffer(GPU_RenderData *data, Uint32 w, Uint32 h, SDL_GPUTextureFormat fmt) |
986 | { |
987 | SDL_GPUTextureCreateInfo tci; |
988 | SDL_zero(tci); |
989 | tci.width = w; |
990 | tci.height = h; |
991 | tci.format = fmt; |
992 | tci.layer_count_or_depth = 1; |
993 | tci.num_levels = 1; |
994 | tci.sample_count = SDL_GPU_SAMPLECOUNT_1; |
995 | tci.usage = SDL_GPU_TEXTUREUSAGE_COLOR_TARGET | SDL_GPU_TEXTUREUSAGE_SAMPLER; |
996 | |
997 | data->backbuffer.texture = SDL_CreateGPUTexture(data->device, &tci); |
998 | data->backbuffer.width = w; |
999 | data->backbuffer.height = h; |
1000 | data->backbuffer.format = fmt; |
1001 | |
1002 | if (!data->backbuffer.texture) { |
1003 | return false; |
1004 | } |
1005 | |
1006 | return true; |
1007 | } |
1008 | |
1009 | static bool GPU_RenderPresent(SDL_Renderer *renderer) |
1010 | { |
1011 | GPU_RenderData *data = (GPU_RenderData *)renderer->internal; |
1012 | |
1013 | SDL_GPUTexture *swapchain; |
1014 | Uint32 swapchain_texture_width, swapchain_texture_height; |
1015 | bool result = SDL_WaitAndAcquireGPUSwapchainTexture(data->state.command_buffer, renderer->window, &swapchain, &swapchain_texture_width, &swapchain_texture_height); |
1016 | |
1017 | if (!result) { |
1018 | SDL_LogError(SDL_LOG_CATEGORY_RENDER, "Failed to acquire swapchain texture: %s" , SDL_GetError()); |
1019 | } |
1020 | |
1021 | if (swapchain != NULL) { |
1022 | SDL_GPUBlitInfo blit_info; |
1023 | SDL_zero(blit_info); |
1024 | |
1025 | blit_info.source.texture = data->backbuffer.texture; |
1026 | blit_info.source.w = data->backbuffer.width; |
1027 | blit_info.source.h = data->backbuffer.height; |
1028 | blit_info.destination.texture = swapchain; |
1029 | blit_info.destination.w = swapchain_texture_width; |
1030 | blit_info.destination.h = swapchain_texture_height; |
1031 | blit_info.load_op = SDL_GPU_LOADOP_DONT_CARE; |
1032 | blit_info.filter = SDL_GPU_FILTER_LINEAR; |
1033 | |
1034 | SDL_BlitGPUTexture(data->state.command_buffer, &blit_info); |
1035 | |
1036 | SDL_SubmitGPUCommandBuffer(data->state.command_buffer); |
1037 | |
1038 | if (swapchain_texture_width != data->backbuffer.width || swapchain_texture_height != data->backbuffer.height) { |
1039 | SDL_ReleaseGPUTexture(data->device, data->backbuffer.texture); |
1040 | CreateBackbuffer(data, swapchain_texture_width, swapchain_texture_height, SDL_GetGPUSwapchainTextureFormat(data->device, renderer->window)); |
1041 | } |
1042 | } else { |
1043 | SDL_SubmitGPUCommandBuffer(data->state.command_buffer); |
1044 | } |
1045 | |
1046 | data->state.command_buffer = SDL_AcquireGPUCommandBuffer(data->device); |
1047 | |
1048 | return true; |
1049 | } |
1050 | |
1051 | static void GPU_DestroyTexture(SDL_Renderer *renderer, SDL_Texture *texture) |
1052 | { |
1053 | GPU_RenderData *renderdata = (GPU_RenderData *)renderer->internal; |
1054 | GPU_TextureData *data = (GPU_TextureData *)texture->internal; |
1055 | |
1056 | if (renderdata->state.render_target == texture) { |
1057 | renderdata->state.render_target = NULL; |
1058 | } |
1059 | |
1060 | if (!data) { |
1061 | return; |
1062 | } |
1063 | |
1064 | SDL_ReleaseGPUTexture(renderdata->device, data->texture); |
1065 | SDL_free(data->pixels); |
1066 | SDL_free(data); |
1067 | texture->internal = NULL; |
1068 | } |
1069 | |
1070 | static void GPU_DestroyRenderer(SDL_Renderer *renderer) |
1071 | { |
1072 | GPU_RenderData *data = (GPU_RenderData *)renderer->internal; |
1073 | |
1074 | if (!data) { |
1075 | return; |
1076 | } |
1077 | |
1078 | if (data->state.command_buffer) { |
1079 | SDL_SubmitGPUCommandBuffer(data->state.command_buffer); |
1080 | data->state.command_buffer = NULL; |
1081 | } |
1082 | |
1083 | for (Uint32 i = 0; i < sizeof(data->samplers) / sizeof(SDL_GPUSampler *); ++i) { |
1084 | SDL_ReleaseGPUSampler(data->device, ((SDL_GPUSampler **)data->samplers)[i]); |
1085 | } |
1086 | |
1087 | if (data->backbuffer.texture) { |
1088 | SDL_ReleaseGPUTexture(data->device, data->backbuffer.texture); |
1089 | } |
1090 | |
1091 | if (renderer->window) { |
1092 | SDL_ReleaseWindowFromGPUDevice(data->device, renderer->window); |
1093 | } |
1094 | |
1095 | ReleaseVertexBuffer(data); |
1096 | GPU_DestroyPipelineCache(&data->pipeline_cache); |
1097 | GPU_ReleaseShaders(&data->shaders, data->device); |
1098 | SDL_DestroyGPUDevice(data->device); |
1099 | |
1100 | SDL_free(data); |
1101 | } |
1102 | |
1103 | static bool ChoosePresentMode(SDL_GPUDevice *device, SDL_Window *window, const int vsync, SDL_GPUPresentMode *out_mode) |
1104 | { |
1105 | SDL_GPUPresentMode mode; |
1106 | |
1107 | switch (vsync) { |
1108 | case 0: |
1109 | mode = SDL_GPU_PRESENTMODE_MAILBOX; |
1110 | |
1111 | if (!SDL_WindowSupportsGPUPresentMode(device, window, mode)) { |
1112 | mode = SDL_GPU_PRESENTMODE_IMMEDIATE; |
1113 | |
1114 | if (!SDL_WindowSupportsGPUPresentMode(device, window, mode)) { |
1115 | mode = SDL_GPU_PRESENTMODE_VSYNC; |
1116 | } |
1117 | } |
1118 | |
1119 | // FIXME should we return an error if both mailbox and immediate fail? |
1120 | break; |
1121 | |
1122 | case 1: |
1123 | mode = SDL_GPU_PRESENTMODE_VSYNC; |
1124 | break; |
1125 | |
1126 | default: |
1127 | return SDL_Unsupported(); |
1128 | } |
1129 | |
1130 | *out_mode = mode; |
1131 | return true; |
1132 | } |
1133 | |
1134 | static bool GPU_SetVSync(SDL_Renderer *renderer, const int vsync) |
1135 | { |
1136 | GPU_RenderData *data = (GPU_RenderData *)renderer->internal; |
1137 | SDL_GPUPresentMode mode = SDL_GPU_PRESENTMODE_VSYNC; |
1138 | |
1139 | if (!ChoosePresentMode(data->device, renderer->window, vsync, &mode)) { |
1140 | return false; |
1141 | } |
1142 | |
1143 | if (mode != data->swapchain.present_mode) { |
1144 | // XXX returns bool instead of SDL-style error code |
1145 | if (SDL_SetGPUSwapchainParameters(data->device, renderer->window, data->swapchain.composition, mode)) { |
1146 | data->swapchain.present_mode = mode; |
1147 | return true; |
1148 | } else { |
1149 | return false; |
1150 | } |
1151 | } |
1152 | |
1153 | return true; |
1154 | } |
1155 | |
1156 | static bool InitSamplers(GPU_RenderData *data) |
1157 | { |
1158 | struct |
1159 | { |
1160 | struct |
1161 | { |
1162 | SDL_TextureAddressMode address_mode; |
1163 | SDL_ScaleMode scale_mode; |
1164 | } sdl; |
1165 | struct |
1166 | { |
1167 | SDL_GPUSamplerAddressMode address_mode; |
1168 | SDL_GPUFilter filter; |
1169 | SDL_GPUSamplerMipmapMode mipmap_mode; |
1170 | Uint32 anisotropy; |
1171 | } gpu; |
1172 | } configs[] = { |
1173 | { |
1174 | { SDL_TEXTURE_ADDRESS_CLAMP, SDL_SCALEMODE_NEAREST }, |
1175 | { SDL_GPU_SAMPLERADDRESSMODE_REPEAT, SDL_GPU_FILTER_NEAREST, SDL_GPU_SAMPLERMIPMAPMODE_NEAREST, 0 }, |
1176 | }, |
1177 | { |
1178 | { SDL_TEXTURE_ADDRESS_CLAMP, SDL_SCALEMODE_LINEAR }, |
1179 | { SDL_GPU_SAMPLERADDRESSMODE_REPEAT, SDL_GPU_FILTER_LINEAR, SDL_GPU_SAMPLERMIPMAPMODE_LINEAR, 0 }, |
1180 | }, |
1181 | { |
1182 | { SDL_TEXTURE_ADDRESS_CLAMP, SDL_SCALEMODE_PIXELART }, |
1183 | { SDL_GPU_SAMPLERADDRESSMODE_REPEAT, SDL_GPU_FILTER_NEAREST, SDL_GPU_SAMPLERMIPMAPMODE_NEAREST, 0 }, |
1184 | }, |
1185 | { |
1186 | { SDL_TEXTURE_ADDRESS_WRAP, SDL_SCALEMODE_NEAREST }, |
1187 | { SDL_GPU_SAMPLERADDRESSMODE_REPEAT, SDL_GPU_FILTER_NEAREST, SDL_GPU_SAMPLERMIPMAPMODE_NEAREST, 0 }, |
1188 | }, |
1189 | { |
1190 | { SDL_TEXTURE_ADDRESS_WRAP, SDL_SCALEMODE_LINEAR }, |
1191 | { SDL_GPU_SAMPLERADDRESSMODE_REPEAT, SDL_GPU_FILTER_LINEAR, SDL_GPU_SAMPLERMIPMAPMODE_LINEAR, 0 }, |
1192 | }, |
1193 | { |
1194 | { SDL_TEXTURE_ADDRESS_WRAP, SDL_SCALEMODE_PIXELART }, |
1195 | { SDL_GPU_SAMPLERADDRESSMODE_REPEAT, SDL_GPU_FILTER_NEAREST, SDL_GPU_SAMPLERMIPMAPMODE_NEAREST, 0 }, |
1196 | }, |
1197 | }; |
1198 | |
1199 | for (Uint32 i = 0; i < SDL_arraysize(configs); ++i) { |
1200 | SDL_GPUSamplerCreateInfo sci; |
1201 | SDL_zero(sci); |
1202 | sci.max_anisotropy = configs[i].gpu.anisotropy; |
1203 | sci.enable_anisotropy = configs[i].gpu.anisotropy > 0; |
1204 | sci.address_mode_u = sci.address_mode_v = sci.address_mode_w = configs[i].gpu.address_mode; |
1205 | sci.min_filter = sci.mag_filter = configs[i].gpu.filter; |
1206 | sci.mipmap_mode = configs[i].gpu.mipmap_mode; |
1207 | |
1208 | SDL_GPUSampler *sampler = SDL_CreateGPUSampler(data->device, &sci); |
1209 | |
1210 | if (sampler == NULL) { |
1211 | return false; |
1212 | } |
1213 | |
1214 | *SamplerPointer(data, configs[i].sdl.address_mode, configs[i].sdl.scale_mode) = sampler; |
1215 | } |
1216 | |
1217 | return true; |
1218 | } |
1219 | |
1220 | static bool GPU_CreateRenderer(SDL_Renderer *renderer, SDL_Window *window, SDL_PropertiesID create_props) |
1221 | { |
1222 | GPU_RenderData *data = NULL; |
1223 | |
1224 | SDL_SetupRendererColorspace(renderer, create_props); |
1225 | |
1226 | if (renderer->output_colorspace != SDL_COLORSPACE_SRGB) { |
1227 | // TODO |
1228 | return SDL_SetError("Unsupported output colorspace" ); |
1229 | } |
1230 | |
1231 | data = (GPU_RenderData *)SDL_calloc(1, sizeof(*data)); |
1232 | if (!data) { |
1233 | return false; |
1234 | } |
1235 | |
1236 | renderer->SupportsBlendMode = GPU_SupportsBlendMode; |
1237 | renderer->CreateTexture = GPU_CreateTexture; |
1238 | renderer->UpdateTexture = GPU_UpdateTexture; |
1239 | renderer->LockTexture = GPU_LockTexture; |
1240 | renderer->UnlockTexture = GPU_UnlockTexture; |
1241 | renderer->SetRenderTarget = GPU_SetRenderTarget; |
1242 | renderer->QueueSetViewport = GPU_QueueNoOp; |
1243 | renderer->QueueSetDrawColor = GPU_QueueNoOp; |
1244 | renderer->QueueDrawPoints = GPU_QueueDrawPoints; |
1245 | renderer->QueueDrawLines = GPU_QueueDrawPoints; // lines and points queue vertices the same way. |
1246 | renderer->QueueGeometry = GPU_QueueGeometry; |
1247 | renderer->InvalidateCachedState = GPU_InvalidateCachedState; |
1248 | renderer->RunCommandQueue = GPU_RunCommandQueue; |
1249 | renderer->RenderReadPixels = GPU_RenderReadPixels; |
1250 | renderer->RenderPresent = GPU_RenderPresent; |
1251 | renderer->DestroyTexture = GPU_DestroyTexture; |
1252 | renderer->DestroyRenderer = GPU_DestroyRenderer; |
1253 | renderer->SetVSync = GPU_SetVSync; |
1254 | renderer->internal = data; |
1255 | renderer->window = window; |
1256 | renderer->name = GPU_RenderDriver.name; |
1257 | |
1258 | bool debug = SDL_GetBooleanProperty(create_props, SDL_PROP_GPU_DEVICE_CREATE_DEBUGMODE_BOOLEAN, false); |
1259 | bool lowpower = SDL_GetBooleanProperty(create_props, SDL_PROP_GPU_DEVICE_CREATE_PREFERLOWPOWER_BOOLEAN, false); |
1260 | |
1261 | // Prefer environment variables/hints if they exist, otherwise defer to properties |
1262 | debug = SDL_GetHintBoolean(SDL_HINT_RENDER_GPU_DEBUG, debug); |
1263 | lowpower = SDL_GetHintBoolean(SDL_HINT_RENDER_GPU_LOW_POWER, lowpower); |
1264 | |
1265 | SDL_SetBooleanProperty(create_props, SDL_PROP_GPU_DEVICE_CREATE_DEBUGMODE_BOOLEAN, debug); |
1266 | SDL_SetBooleanProperty(create_props, SDL_PROP_GPU_DEVICE_CREATE_PREFERLOWPOWER_BOOLEAN, lowpower); |
1267 | |
1268 | GPU_FillSupportedShaderFormats(create_props); |
1269 | data->device = SDL_CreateGPUDeviceWithProperties(create_props); |
1270 | |
1271 | if (!data->device) { |
1272 | return false; |
1273 | } |
1274 | |
1275 | if (!GPU_InitShaders(&data->shaders, data->device)) { |
1276 | return false; |
1277 | } |
1278 | |
1279 | if (!GPU_InitPipelineCache(&data->pipeline_cache, data->device)) { |
1280 | return false; |
1281 | } |
1282 | |
1283 | // XXX what's a good initial size? |
1284 | if (!InitVertexBuffer(data, 1 << 16)) { |
1285 | return false; |
1286 | } |
1287 | |
1288 | if (!InitSamplers(data)) { |
1289 | return false; |
1290 | } |
1291 | |
1292 | if (!SDL_ClaimWindowForGPUDevice(data->device, window)) { |
1293 | return false; |
1294 | } |
1295 | |
1296 | data->swapchain.composition = SDL_GPU_SWAPCHAINCOMPOSITION_SDR; |
1297 | data->swapchain.present_mode = SDL_GPU_PRESENTMODE_VSYNC; |
1298 | |
1299 | int vsync = (int)SDL_GetNumberProperty(create_props, SDL_PROP_RENDERER_CREATE_PRESENT_VSYNC_NUMBER, 0); |
1300 | ChoosePresentMode(data->device, window, vsync, &data->swapchain.present_mode); |
1301 | |
1302 | SDL_SetGPUSwapchainParameters(data->device, window, data->swapchain.composition, data->swapchain.present_mode); |
1303 | |
1304 | SDL_SetGPUAllowedFramesInFlight(data->device, 1); |
1305 | |
1306 | SDL_AddSupportedTextureFormat(renderer, SDL_PIXELFORMAT_BGRA32); |
1307 | SDL_AddSupportedTextureFormat(renderer, SDL_PIXELFORMAT_RGBA32); |
1308 | SDL_AddSupportedTextureFormat(renderer, SDL_PIXELFORMAT_BGRX32); |
1309 | SDL_AddSupportedTextureFormat(renderer, SDL_PIXELFORMAT_RGBX32); |
1310 | |
1311 | SDL_SetNumberProperty(SDL_GetRendererProperties(renderer), SDL_PROP_RENDERER_MAX_TEXTURE_SIZE_NUMBER, 16384); |
1312 | |
1313 | data->state.draw_color.r = 1.0f; |
1314 | data->state.draw_color.g = 1.0f; |
1315 | data->state.draw_color.b = 1.0f; |
1316 | data->state.draw_color.a = 1.0f; |
1317 | data->state.viewport.min_depth = 0; |
1318 | data->state.viewport.max_depth = 1; |
1319 | data->state.command_buffer = SDL_AcquireGPUCommandBuffer(data->device); |
1320 | |
1321 | int w, h; |
1322 | SDL_GetWindowSizeInPixels(window, &w, &h); |
1323 | |
1324 | if (!CreateBackbuffer(data, w, h, SDL_GetGPUSwapchainTextureFormat(data->device, window))) { |
1325 | return false; |
1326 | } |
1327 | |
1328 | SDL_SetPointerProperty(SDL_GetRendererProperties(renderer), SDL_PROP_RENDERER_GPU_DEVICE_POINTER, data->device); |
1329 | |
1330 | return true; |
1331 | } |
1332 | |
1333 | SDL_RenderDriver GPU_RenderDriver = { |
1334 | GPU_CreateRenderer, "gpu" |
1335 | }; |
1336 | |
1337 | #endif // SDL_VIDEO_RENDER_GPU |
1338 | |