July 2023 - Wine-Commits - list.winehq.org

Matteo Bruni : d3dx9/tests: Get rid of unnecessary checks.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: wine Branch: master Commit: 64d9bc872b01baf5c481e8feea1410f59428895b URL: https://gitlab.winehq.org/wine/wine/-/commit/64d9bc872b01baf5c481e8feea1410… Author: Matteo Bruni <mbruni(a)codeweavers.com> Date: Tue Dec 6 00:06:35 2022 +0100 d3dx9/tests: Get rid of unnecessary checks. --- dlls/d3dx9_36/tests/core.c | 30 ++++++++++++------------------ 1 file changed, 12 insertions(+), 18 deletions(-) diff --git a/dlls/d3dx9_36/tests/core.c b/dlls/d3dx9_36/tests/core.c index c8925b2b186..0ef4c1af35b 100644 --- a/dlls/d3dx9_36/tests/core.c +++ b/dlls/d3dx9_36/tests/core.c @@ -1210,27 +1210,21 @@ static void test_D3DXCreateRenderToSurface(IDirect3DDevice9 *device) hr = D3DXCreateRenderToSurface(device, 256, 256, D3DFMT_A8R8G8B8, FALSE, D3DFMT_UNKNOWN, NULL /* out */); ok(hr == D3DERR_INVALIDCALL, "D3DXCreateRenderToSurface returned %#lx, expected %#lx\n", hr, D3DERR_INVALIDCALL); - for (i = 0; i < ARRAY_SIZE(tests); i++) + for (i = 0; i < ARRAY_SIZE(tests); ++i) { hr = D3DXCreateRenderToSurface(device, tests[i].Width, tests[i].Height, tests[i].Format, tests[i].DepthStencil, tests[i].DepthStencilFormat, &render); - ok(hr == D3D_OK, "%d: D3DXCreateRenderToSurface returned %#lx, expected %#lx\n", i, hr, D3D_OK); - if (SUCCEEDED(hr)) - { - hr = ID3DXRenderToSurface_GetDesc(render, &desc); - ok(hr == D3D_OK, "%d: GetDesc failed %#lx\n", i, hr); - if (SUCCEEDED(hr)) - { - ok(desc.Width == tests[i].Width, "%d: Got width %u, expected %u\n", i, desc.Width, tests[i].Width); - ok(desc.Height == tests[i].Height, "%d: Got height %u, expected %u\n", i, desc.Height, tests[i].Height); - ok(desc.Format == tests[i].Format, "%d: Got format %#x, expected %#x\n", i, desc.Format, tests[i].Format); - ok(desc.DepthStencil == tests[i].DepthStencil, "%d: Got depth stencil %d, expected %d\n", - i, desc.DepthStencil, tests[i].DepthStencil); - ok(desc.DepthStencilFormat == tests[i].DepthStencilFormat, "%d: Got depth stencil format %#x, expected %#x\n", - i, desc.DepthStencilFormat, tests[i].DepthStencilFormat); - } - ID3DXRenderToSurface_Release(render); - } + ok(hr == D3D_OK, "Unexpected hr %#lx.\n", hr); + hr = ID3DXRenderToSurface_GetDesc(render, &desc); + ok(hr == D3D_OK, "Unexpected hr %#lx.\n", hr); + ok(desc.Width == tests[i].Width, "%d: Got width %u, expected %u\n", i, desc.Width, tests[i].Width); + ok(desc.Height == tests[i].Height, "%d: Got height %u, expected %u\n", i, desc.Height, tests[i].Height); + ok(desc.Format == tests[i].Format, "%d: Got format %#x, expected %#x\n", i, desc.Format, tests[i].Format); + ok(desc.DepthStencil == tests[i].DepthStencil, "%d: Got depth stencil %d, expected %d\n", + i, desc.DepthStencil, tests[i].DepthStencil); + ok(desc.DepthStencilFormat == tests[i].DepthStencilFormat, "%d: Got depth stencil format %#x, expected %#x\n", + i, desc.DepthStencilFormat, tests[i].DepthStencilFormat); + ID3DXRenderToSurface_Release(render); } /* check device ref count */

1 0

Matteo Bruni : d3dx9/tests: Increase tolerance for a D3DXSHMultiply3 test.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: wine Branch: master Commit: a3ad4a5066c130857e270a84b209970e6737a86e URL: https://gitlab.winehq.org/wine/wine/-/commit/a3ad4a5066c130857e270a84b20997… Author: Matteo Bruni <mbruni(a)codeweavers.com> Date: Tue May 2 17:12:05 2023 +0200 d3dx9/tests: Increase tolerance for a D3DXSHMultiply3 test. Otherwise it fails for me on 32-bit Linux with my current compiler. --- dlls/d3dx9_36/tests/math.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dlls/d3dx9_36/tests/math.c b/dlls/d3dx9_36/tests/math.c index 8f4af067bc7..5d112bd4b92 100644 --- a/dlls/d3dx9_36/tests/math.c +++ b/dlls/d3dx9_36/tests/math.c @@ -4095,7 +4095,7 @@ static void test_D3DXSHMultiply3(void) D3DXSHMultiply3(c, c, b); for (i = 0; i < ARRAY_SIZE(expected_aliased); ++i) { - equal = compare_float(c[i], expected_aliased[i], 34); + equal = compare_float(c[i], expected_aliased[i], 64); ok(equal, "Expected[%u] = %.8e, received = %.8e.\n", i, expected_aliased[i], c[i]); } }

1 0

Eric Pouech : d3dx9/tests: Clean up some ok() messages.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: wine Branch: master Commit: 8031d00fed2936b4483c0537d30ec1c6279eec4a URL: https://gitlab.winehq.org/wine/wine/-/commit/8031d00fed2936b4483c0537d30ec1… Author: Eric Pouech <eric.pouech(a)gmail.com> Date: Wed Feb 2 09:38:00 2022 +0100 d3dx9/tests: Clean up some ok() messages. --- dlls/d3dx9_36/tests/asm.c | 46 +++++++++++++++++++------------------------- dlls/d3dx9_36/tests/shader.c | 4 ++-- 2 files changed, 22 insertions(+), 28 deletions(-)

1 0

Zebediah Figura : wined3d: Create fences when the command buffer is allocated.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: wine Branch: master Commit: 2740af1448103b5b3813d8b67eaee5eb584fe525 URL: https://gitlab.winehq.org/wine/wine/-/commit/2740af1448103b5b3813d8b67eaee5… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Tue May 23 17:01:02 2023 -0500 wined3d: Create fences when the command buffer is allocated. This shouldn't really make a difference, but does make the code a bit conceptually simpler. --- dlls/wined3d/context_vk.c | 30 +++++++++++++----------------- 1 file changed, 13 insertions(+), 17 deletions(-) diff --git a/dlls/wined3d/context_vk.c b/dlls/wined3d/context_vk.c index bf2e527f566..097f8c7032c 100644 --- a/dlls/wined3d/context_vk.c +++ b/dlls/wined3d/context_vk.c @@ -1114,8 +1114,7 @@ static void free_command_buffer(struct wined3d_context_vk *context_vk, struct wi struct wined3d_device_vk *device_vk = wined3d_device_vk(context_vk->c.device); const struct wined3d_vk_info *vk_info = context_vk->vk_info; - if (buffer->vk_fence) - VK_CALL(vkDestroyFence(device_vk->vk_device, buffer->vk_fence, NULL)); + VK_CALL(vkDestroyFence(device_vk->vk_device, buffer->vk_fence, NULL)); VK_CALL(vkFreeCommandBuffers(device_vk->vk_device, context_vk->vk_command_pool, 1, &buffer->vk_command_buffer)); } @@ -1772,6 +1771,16 @@ VkCommandBuffer wined3d_context_vk_get_command_buffer(struct wined3d_context_vk else { VkCommandBufferAllocateInfo command_buffer_info; + VkFenceCreateInfo fence_desc; + + fence_desc.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; + fence_desc.pNext = NULL; + fence_desc.flags = 0; + if ((vr = VK_CALL(vkCreateFence(device_vk->vk_device, &fence_desc, NULL, &buffer->vk_fence))) < 0) + { + ERR("Failed to create fence, vr %s.\n", wined3d_debug_vkresult(vr)); + return VK_NULL_HANDLE; + } command_buffer_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; command_buffer_info.pNext = NULL; @@ -1782,10 +1791,9 @@ VkCommandBuffer wined3d_context_vk_get_command_buffer(struct wined3d_context_vk &command_buffer_info, &buffer->vk_command_buffer))) < 0) { WARN("Failed to allocate Vulkan command buffer, vr %s.\n", wined3d_debug_vkresult(vr)); + VK_CALL(vkDestroyFence(device_vk->vk_device, buffer->vk_fence, NULL)); return VK_NULL_HANDLE; } - - buffer->vk_fence = VK_NULL_HANDLE; } begin_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; @@ -1826,7 +1834,6 @@ void wined3d_context_vk_submit_command_buffer(struct wined3d_context_vk *context struct wined3d_query_pool_vk *pool_vk, *pool_vk_next; struct wined3d_command_buffer_vk *buffer; struct wined3d_query_vk *query_vk; - VkFenceCreateInfo fence_desc; VkSubmitInfo submit_info; VkResult vr; @@ -1870,18 +1877,7 @@ void wined3d_context_vk_submit_command_buffer(struct wined3d_context_vk *context VK_CALL(vkEndCommandBuffer(buffer->vk_command_buffer)); - if (buffer->vk_fence) - { - VK_CALL(vkResetFences(device_vk->vk_device, 1, &buffer->vk_fence)); - } - else - { - fence_desc.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; - fence_desc.pNext = NULL; - fence_desc.flags = 0; - if ((vr = VK_CALL(vkCreateFence(device_vk->vk_device, &fence_desc, NULL, &buffer->vk_fence))) < 0) - ERR("Failed to create fence, vr %s.\n", wined3d_debug_vkresult(vr)); - } + VK_CALL(vkResetFences(device_vk->vk_device, 1, &buffer->vk_fence)); submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO; submit_info.pNext = NULL;

1 0

Zebediah Figura : wined3d: Reuse command buffers and fences.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: wine Branch: master Commit: c42b4fec1f81078ff5aabed33b9fe753d33c71fc URL: https://gitlab.winehq.org/wine/wine/-/commit/c42b4fec1f81078ff5aabed33b9fe7… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Tue May 23 16:55:42 2023 -0500 wined3d: Reuse command buffers and fences. Allocating and freeing fences in particular can be expensive, as much as 250 µs per call (observed on an NVidia machine). --- dlls/wined3d/context_vk.c | 79 ++++++++++++++++++++++++++++++------------ dlls/wined3d/wined3d_private.h | 2 +- 2 files changed, 58 insertions(+), 23 deletions(-) diff --git a/dlls/wined3d/context_vk.c b/dlls/wined3d/context_vk.c index e70e70f28ec..bf2e527f566 100644 --- a/dlls/wined3d/context_vk.c +++ b/dlls/wined3d/context_vk.c @@ -1109,19 +1109,31 @@ void wined3d_context_vk_destroy_bo(struct wined3d_context_vk *context_vk, const wined3d_context_vk_destroy_vk_memory(context_vk, bo->vk_memory, bo->command_buffer_id); } -static void wined3d_context_vk_remove_command_buffer(struct wined3d_context_vk *context_vk, - unsigned int submit_index) +static void free_command_buffer(struct wined3d_context_vk *context_vk, struct wined3d_command_buffer_vk *buffer) { - struct wined3d_command_buffer_vk *buffer = &context_vk->submitted.buffers[submit_index]; struct wined3d_device_vk *device_vk = wined3d_device_vk(context_vk->c.device); const struct wined3d_vk_info *vk_info = context_vk->vk_info; - VK_CALL(vkDestroyFence(device_vk->vk_device, buffer->vk_fence, NULL)); + if (buffer->vk_fence) + VK_CALL(vkDestroyFence(device_vk->vk_device, buffer->vk_fence, NULL)); VK_CALL(vkFreeCommandBuffers(device_vk->vk_device, context_vk->vk_command_pool, 1, &buffer->vk_command_buffer)); +} + +static void wined3d_context_vk_remove_command_buffer(struct wined3d_context_vk *context_vk, + unsigned int submit_index) +{ + struct wined3d_command_buffer_vk *buffer = &context_vk->submitted.buffers[submit_index]; if (buffer->id > context_vk->completed_command_buffer_id) context_vk->completed_command_buffer_id = buffer->id; + + if (wined3d_array_reserve((void **)&context_vk->completed.buffers, &context_vk->completed.buffers_size, + context_vk->completed.buffer_count + 1, sizeof(*context_vk->completed.buffers))) + context_vk->completed.buffers[context_vk->completed.buffer_count++] = *buffer; + else + free_command_buffer(context_vk, buffer); + *buffer = context_vk->submitted.buffers[--context_vk->submitted.buffer_count]; } @@ -1685,13 +1697,14 @@ void wined3d_context_vk_cleanup(struct wined3d_context_vk *context_vk) if (buffer->vk_command_buffer) { - VK_CALL(vkFreeCommandBuffers(device_vk->vk_device, - context_vk->vk_command_pool, 1, &buffer->vk_command_buffer)); + free_command_buffer(context_vk, buffer); buffer->vk_command_buffer = VK_NULL_HANDLE; } wined3d_context_vk_wait_command_buffer(context_vk, buffer->id - 1); context_vk->completed_command_buffer_id = buffer->id; + for (i = 0; i < context_vk->completed.buffer_count; ++i) + free_command_buffer(context_vk, &context_vk->completed.buffers[i]); heap_free(context_vk->compute.bindings.bindings); heap_free(context_vk->graphics.bindings.bindings); @@ -1713,6 +1726,7 @@ void wined3d_context_vk_cleanup(struct wined3d_context_vk *context_vk) wined3d_context_vk_destroy_query_pools(context_vk, &context_vk->free_stream_output_statistics_query_pools); wine_rb_destroy(&context_vk->bo_slab_available, wined3d_context_vk_destroy_bo_slab, context_vk); heap_free(context_vk->submitted.buffers); + heap_free(context_vk->completed.buffers); heap_free(context_vk->retired.objects); wined3d_shader_descriptor_writes_vk_cleanup(&context_vk->descriptor_writes); @@ -1727,7 +1741,6 @@ VkCommandBuffer wined3d_context_vk_get_command_buffer(struct wined3d_context_vk { struct wined3d_device_vk *device_vk = wined3d_device_vk(context_vk->c.device); const struct wined3d_vk_info *vk_info = context_vk->vk_info; - VkCommandBufferAllocateInfo command_buffer_info; struct wined3d_command_buffer_vk *buffer; VkCommandBufferBeginInfo begin_info; struct wined3d_query_vk *query_vk; @@ -1748,16 +1761,31 @@ VkCommandBuffer wined3d_context_vk_get_command_buffer(struct wined3d_context_vk } } - command_buffer_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; - command_buffer_info.pNext = NULL; - command_buffer_info.commandPool = context_vk->vk_command_pool; - command_buffer_info.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY; - command_buffer_info.commandBufferCount = 1; - if ((vr = VK_CALL(vkAllocateCommandBuffers(device_vk->vk_device, - &command_buffer_info, &buffer->vk_command_buffer))) < 0) + if (context_vk->completed.buffer_count) { - WARN("Failed to allocate Vulkan command buffer, vr %s.\n", wined3d_debug_vkresult(vr)); - return VK_NULL_HANDLE; + const struct wined3d_command_buffer_vk *old_buffer; + + old_buffer = &context_vk->completed.buffers[--context_vk->completed.buffer_count]; + buffer->vk_command_buffer = old_buffer->vk_command_buffer; + buffer->vk_fence = old_buffer->vk_fence; + } + else + { + VkCommandBufferAllocateInfo command_buffer_info; + + command_buffer_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_ALLOCATE_INFO; + command_buffer_info.pNext = NULL; + command_buffer_info.commandPool = context_vk->vk_command_pool; + command_buffer_info.level = VK_COMMAND_BUFFER_LEVEL_PRIMARY; + command_buffer_info.commandBufferCount = 1; + if ((vr = VK_CALL(vkAllocateCommandBuffers(device_vk->vk_device, + &command_buffer_info, &buffer->vk_command_buffer))) < 0) + { + WARN("Failed to allocate Vulkan command buffer, vr %s.\n", wined3d_debug_vkresult(vr)); + return VK_NULL_HANDLE; + } + + buffer->vk_fence = VK_NULL_HANDLE; } begin_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; @@ -1842,11 +1870,18 @@ void wined3d_context_vk_submit_command_buffer(struct wined3d_context_vk *context VK_CALL(vkEndCommandBuffer(buffer->vk_command_buffer)); - fence_desc.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; - fence_desc.pNext = NULL; - fence_desc.flags = 0; - if ((vr = VK_CALL(vkCreateFence(device_vk->vk_device, &fence_desc, NULL, &buffer->vk_fence))) < 0) - ERR("Failed to create fence, vr %s.\n", wined3d_debug_vkresult(vr)); + if (buffer->vk_fence) + { + VK_CALL(vkResetFences(device_vk->vk_device, 1, &buffer->vk_fence)); + } + else + { + fence_desc.sType = VK_STRUCTURE_TYPE_FENCE_CREATE_INFO; + fence_desc.pNext = NULL; + fence_desc.flags = 0; + if ((vr = VK_CALL(vkCreateFence(device_vk->vk_device, &fence_desc, NULL, &buffer->vk_fence))) < 0) + ERR("Failed to create fence, vr %s.\n", wined3d_debug_vkresult(vr)); + } submit_info.sType = VK_STRUCTURE_TYPE_SUBMIT_INFO; submit_info.pNext = NULL; @@ -3825,7 +3860,7 @@ HRESULT wined3d_context_vk_init(struct wined3d_context_vk *context_vk, struct wi command_pool_info.sType = VK_STRUCTURE_TYPE_COMMAND_POOL_CREATE_INFO; command_pool_info.pNext = NULL; - command_pool_info.flags = VK_COMMAND_POOL_CREATE_TRANSIENT_BIT; + command_pool_info.flags = VK_COMMAND_POOL_CREATE_TRANSIENT_BIT | VK_COMMAND_POOL_CREATE_RESET_COMMAND_BUFFER_BIT; command_pool_info.queueFamilyIndex = device_vk->vk_queue_family_index; if ((vr = VK_CALL(vkCreateCommandPool(device_vk->vk_device, &command_pool_info, NULL, &context_vk->vk_command_pool))) < 0) diff --git a/dlls/wined3d/wined3d_private.h b/dlls/wined3d/wined3d_private.h index efdcff41e58..e67c2e17791 100644 --- a/dlls/wined3d/wined3d_private.h +++ b/dlls/wined3d/wined3d_private.h @@ -2729,7 +2729,7 @@ struct wined3d_context_vk struct wined3d_command_buffer_vk *buffers; SIZE_T buffers_size; SIZE_T buffer_count; - } submitted; + } submitted, completed; struct wined3d_shader_descriptor_writes_vk descriptor_writes;

1 0

Zebediah Figura : wined3d: Set VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: wine Branch: master Commit: e0769f42ec0394c681623594c316e08d31b0e980 URL: https://gitlab.winehq.org/wine/wine/-/commit/e0769f42ec0394c681623594c316e0… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Tue May 23 15:31:22 2023 -0500 wined3d: Set VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT. --- dlls/wined3d/context_vk.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dlls/wined3d/context_vk.c b/dlls/wined3d/context_vk.c index 8c021a66bc7..e70e70f28ec 100644 --- a/dlls/wined3d/context_vk.c +++ b/dlls/wined3d/context_vk.c @@ -1762,7 +1762,7 @@ VkCommandBuffer wined3d_context_vk_get_command_buffer(struct wined3d_context_vk begin_info.sType = VK_STRUCTURE_TYPE_COMMAND_BUFFER_BEGIN_INFO; begin_info.pNext = NULL; - begin_info.flags = 0; + begin_info.flags = VK_COMMAND_BUFFER_USAGE_ONE_TIME_SUBMIT_BIT; begin_info.pInheritanceInfo = NULL; if ((vr = VK_CALL(vkBeginCommandBuffer(buffer->vk_command_buffer, &begin_info))) < 0) {

1 0

Zebediah Figura : vkd3d-shader/hlsl: Do not emit DEF instructions for uniform constants.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: vkd3d Branch: master Commit: 53e9ad3e4ce58dc3a91fa9e2073e3dad123ad43a URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/53e9ad3e4ce58dc3a91fa9e2073e3… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri Feb 24 16:42:26 2023 -0600 vkd3d-shader/hlsl: Do not emit DEF instructions for uniform constants. --- libs/vkd3d-shader/d3dbc.c | 5 +++-- libs/vkd3d-shader/hlsl.h | 6 +++++- libs/vkd3d-shader/hlsl_codegen.c | 40 +++++++++++++++++++++++++++------------- 3 files changed, 35 insertions(+), 16 deletions(-) diff --git a/libs/vkd3d-shader/d3dbc.c b/libs/vkd3d-shader/d3dbc.c index 92c78422..369112ce 100644 --- a/libs/vkd3d-shader/d3dbc.c +++ b/libs/vkd3d-shader/d3dbc.c @@ -1818,12 +1818,13 @@ static void write_sm1_constant_defs(struct hlsl_ctx *ctx, struct vkd3d_bytecode_ for (i = 0; i < ctx->constant_defs.count; ++i) { + const struct hlsl_constant_register *constant_reg = &ctx->constant_defs.regs[i]; uint32_t token = D3DSIO_DEF; const struct sm1_dst_register reg = { .type = D3DSPR_CONST, .writemask = VKD3DSP_WRITEMASK_ALL, - .reg = i, + .reg = constant_reg->index, }; if (ctx->profile->major_version > 1) @@ -1832,7 +1833,7 @@ static void write_sm1_constant_defs(struct hlsl_ctx *ctx, struct vkd3d_bytecode_ write_sm1_dst_register(buffer, &reg); for (x = 0; x < 4; ++x) - put_f32(buffer, ctx->constant_defs.values[i].f[x]); + put_f32(buffer, constant_reg->value.f[x]); } } diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index f102657d..7a6abf81 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -810,7 +810,11 @@ struct hlsl_ctx * Only used for SM1 profiles. */ struct hlsl_constant_defs { - struct hlsl_vec4 *values; + struct hlsl_constant_register + { + uint32_t index; + struct hlsl_vec4 value; + } *regs; size_t count, size; } constant_defs; /* Number of temp. registers required for the shader to run, i.e. the largest temp register diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index ce673d1d..8927e291 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -3254,10 +3254,33 @@ static void allocate_temp_registers_recurse(struct hlsl_ctx *ctx, } } +static void record_constant(struct hlsl_ctx *ctx, unsigned int component_index, float f) +{ + struct hlsl_constant_defs *defs = &ctx->constant_defs; + struct hlsl_constant_register *reg; + size_t i; + + for (i = 0; i < defs->count; ++i) + { + reg = &defs->regs[i]; + if (reg->index == (component_index / 4)) + { + reg->value.f[component_index % 4] = f; + return; + } + } + + if (!hlsl_array_reserve(ctx, (void **)&defs->regs, &defs->size, defs->count + 1, sizeof(*defs->regs))) + return; + reg = &defs->regs[defs->count++]; + memset(reg, 0, sizeof(*reg)); + reg->index = component_index / 4; + reg->value.f[component_index % 4] = f; +} + static void allocate_const_registers_recurse(struct hlsl_ctx *ctx, struct hlsl_block *block, struct register_allocator *allocator) { - struct hlsl_constant_defs *defs = &ctx->constant_defs; struct hlsl_ir_node *instr; LIST_FOR_EACH_ENTRY(instr, &block->instrs, struct hlsl_ir_node, entry) @@ -3268,21 +3291,11 @@ static void allocate_const_registers_recurse(struct hlsl_ctx *ctx, { struct hlsl_ir_constant *constant = hlsl_ir_constant(instr); const struct hlsl_type *type = instr->data_type; - unsigned int x, i, end_reg; + unsigned int x, i; constant->reg = allocate_numeric_registers_for_type(ctx, allocator, 1, UINT_MAX, type); TRACE("Allocated constant @%u to %s.\n", instr->index, debug_register('c', constant->reg, type)); - if (!hlsl_array_reserve(ctx, (void **)&defs->values, &defs->size, - constant->reg.id + 1, sizeof(*defs->values))) - return; - end_reg = constant->reg.id + 1; - if (end_reg > defs->count) - { - memset(&defs->values[defs->count], 0, sizeof(*defs->values) * (end_reg - defs->count)); - defs->count = end_reg; - } - assert(type->class <= HLSL_CLASS_LAST_NUMERIC); assert(type->dimy == 1); assert(constant->reg.writemask); @@ -3322,7 +3335,8 @@ static void allocate_const_registers_recurse(struct hlsl_ctx *ctx, default: vkd3d_unreachable(); } - defs->values[constant->reg.id].f[x] = f; + + record_constant(ctx, constant->reg.id * 4 + x, f); } break;

1 0

Zebediah Figura : vkd3d-shader/hlsl: Simplify allocate_const_registers_recurse() to stop accounting for constants larger than a vector.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: vkd3d Branch: master Commit: e5afbda34a0635899c4c3b37decffba9b41c0b77 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/e5afbda34a0635899c4c3b37decff… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Fri Feb 24 16:24:47 2023 -0600 vkd3d-shader/hlsl: Simplify allocate_const_registers_recurse() to stop accounting for constants larger than a vector. That cannot happen with the current code. --- libs/vkd3d-shader/hlsl_codegen.c | 81 +++++++++++++++++++--------------------- 1 file changed, 38 insertions(+), 43 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index e265a36e..ce673d1d 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -3268,16 +3268,15 @@ static void allocate_const_registers_recurse(struct hlsl_ctx *ctx, { struct hlsl_ir_constant *constant = hlsl_ir_constant(instr); const struct hlsl_type *type = instr->data_type; - unsigned int x, y, i, writemask, end_reg; - unsigned int reg_size = type->reg_size[HLSL_REGSET_NUMERIC]; + unsigned int x, i, end_reg; constant->reg = allocate_numeric_registers_for_type(ctx, allocator, 1, UINT_MAX, type); TRACE("Allocated constant @%u to %s.\n", instr->index, debug_register('c', constant->reg, type)); if (!hlsl_array_reserve(ctx, (void **)&defs->values, &defs->size, - constant->reg.id + reg_size / 4, sizeof(*defs->values))) + constant->reg.id + 1, sizeof(*defs->values))) return; - end_reg = constant->reg.id + reg_size / 4; + end_reg = constant->reg.id + 1; if (end_reg > defs->count) { memset(&defs->values[defs->count], 0, sizeof(*defs->values) * (end_reg - defs->count)); @@ -3285,49 +3284,45 @@ static void allocate_const_registers_recurse(struct hlsl_ctx *ctx, } assert(type->class <= HLSL_CLASS_LAST_NUMERIC); + assert(type->dimy == 1); + assert(constant->reg.writemask); - if (!(writemask = constant->reg.writemask)) - writemask = (1u << type->dimx) - 1; - - for (y = 0; y < type->dimy; ++y) + for (x = 0, i = 0; x < 4; ++x) { - for (x = 0, i = 0; x < 4; ++x) + const union hlsl_constant_value_component *value; + float f; + + if (!(constant->reg.writemask & (1u << x))) + continue; + value = &constant->value.u[i++]; + + switch (type->base_type) { - const union hlsl_constant_value_component *value; - float f; - - if (!(writemask & (1u << x))) - continue; - value = &constant->value.u[i++]; - - switch (type->base_type) - { - case HLSL_TYPE_BOOL: - f = !!value->u; - break; - - case HLSL_TYPE_FLOAT: - case HLSL_TYPE_HALF: - f = value->f; - break; - - case HLSL_TYPE_INT: - f = value->i; - break; - - case HLSL_TYPE_UINT: - f = value->u; - break; - - case HLSL_TYPE_DOUBLE: - FIXME("Double constant.\n"); - return; - - default: - vkd3d_unreachable(); - } - defs->values[constant->reg.id + y].f[x] = f; + case HLSL_TYPE_BOOL: + f = !!value->u; + break; + + case HLSL_TYPE_FLOAT: + case HLSL_TYPE_HALF: + f = value->f; + break; + + case HLSL_TYPE_INT: + f = value->i; + break; + + case HLSL_TYPE_UINT: + f = value->u; + break; + + case HLSL_TYPE_DOUBLE: + FIXME("Double constant.\n"); + return; + + default: + vkd3d_unreachable(); } + defs->values[constant->reg.id].f[x] = f; } break;

1 0

Zebediah Figura : vkd3d-shader/d3dbc: Set the source count for DEF instructions to 1 (Valgrind).
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: vkd3d Branch: master Commit: e390bc35e2c9b0a2110370f916033eea2366317e URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/e390bc35e2c9b0a2110370f916033… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Wed Jun 28 16:14:27 2023 -0500 vkd3d-shader/d3dbc: Set the source count for DEF instructions to 1 (Valgrind). We emit one source with multiple components. --- libs/vkd3d-shader/d3dbc.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/libs/vkd3d-shader/d3dbc.c b/libs/vkd3d-shader/d3dbc.c index e8a6236b..92c78422 100644 --- a/libs/vkd3d-shader/d3dbc.c +++ b/libs/vkd3d-shader/d3dbc.c @@ -260,9 +260,9 @@ static const struct vkd3d_sm1_opcode_info vs_opcode_table[] = /* Declarations */ {VKD3D_SM1_OP_DCL, 0, 2, VKD3DSIH_DCL}, /* Constant definitions */ - {VKD3D_SM1_OP_DEF, 1, 4, VKD3DSIH_DEF}, + {VKD3D_SM1_OP_DEF, 1, 1, VKD3DSIH_DEF}, {VKD3D_SM1_OP_DEFB, 1, 1, VKD3DSIH_DEFB}, - {VKD3D_SM1_OP_DEFI, 1, 4, VKD3DSIH_DEFI}, + {VKD3D_SM1_OP_DEFI, 1, 1, VKD3DSIH_DEFI}, /* Control flow */ {VKD3D_SM1_OP_REP, 0, 1, VKD3DSIH_REP, {2, 0}, {~0u, ~0u}}, {VKD3D_SM1_OP_ENDREP, 0, 0, VKD3DSIH_ENDREP, {2, 0}, {~0u, ~0u}}, @@ -327,9 +327,9 @@ static const struct vkd3d_sm1_opcode_info ps_opcode_table[] = /* Declarations */ {VKD3D_SM1_OP_DCL, 0, 2, VKD3DSIH_DCL}, /* Constant definitions */ - {VKD3D_SM1_OP_DEF, 1, 4, VKD3DSIH_DEF}, + {VKD3D_SM1_OP_DEF, 1, 1, VKD3DSIH_DEF}, {VKD3D_SM1_OP_DEFB, 1, 1, VKD3DSIH_DEFB}, - {VKD3D_SM1_OP_DEFI, 1, 4, VKD3DSIH_DEFI}, + {VKD3D_SM1_OP_DEFI, 1, 1, VKD3DSIH_DEFI}, /* Control flow */ {VKD3D_SM1_OP_REP, 0, 1, VKD3DSIH_REP, {2, 1}, {~0u, ~0u}}, {VKD3D_SM1_OP_ENDREP, 0, 0, VKD3DSIH_ENDREP, {2, 1}, {~0u, ~0u}},

1 0

Zebediah Figura : vkd3d-shader: Record a global temporary count per sm4 shader.
by Alexandre Julliard July 4, 2023

July 4, 2023

Module: vkd3d Branch: master Commit: e489098878997fc89997712033a48f27eda7cf50 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/e489098878997fc89997712033a48… Author: Zebediah Figura <zfigura(a)codeweavers.com> Date: Wed Jun 28 14:49:42 2023 -0500 vkd3d-shader: Record a global temporary count per sm4 shader. Store it in the shader_desc, and declare temps from that when compiling SPIR-V, instead of parsing dcl_instructions. As part of this change, we declare a single, global temps array (with Private scope instead of Function) which is as large as the maximum of all dcl_temps instructions. It is not clear to me whether this will improve, hurt, or have no significant effect on the lower-level compiler. An alternative is to still redeclare a new temps array every time (although still with a smaller size). --- libs/vkd3d-shader/spirv.c | 21 +++++++++------------ libs/vkd3d-shader/tpf.c | 2 ++ libs/vkd3d-shader/vkd3d_shader_private.h | 2 ++ 3 files changed, 13 insertions(+), 12 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 3542b5fa..5535a650 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -3217,7 +3217,7 @@ static bool spirv_compiler_get_register_info(const struct spirv_compiler *compil { assert(reg->idx[0].offset < compiler->temp_count); register_info->id = compiler->temp_id + reg->idx[0].offset; - register_info->storage_class = SpvStorageClassFunction; + register_info->storage_class = SpvStorageClassPrivate; register_info->descriptor_array = NULL; register_info->member_idx = 0; register_info->component_type = VKD3D_SHADER_COMPONENT_FLOAT; @@ -5258,8 +5258,7 @@ static void spirv_compiler_emit_dcl_global_flags(struct spirv_compiler *compiler WARN("Unhandled global flags %#x.\n", flags); } -static void spirv_compiler_emit_dcl_temps(struct spirv_compiler *compiler, - const struct vkd3d_shader_instruction *instruction) +static void spirv_compiler_emit_temps(struct spirv_compiler *compiler, uint32_t count) { struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; size_t function_location; @@ -5270,11 +5269,11 @@ static void spirv_compiler_emit_dcl_temps(struct spirv_compiler *compiler, vkd3d_spirv_begin_function_stream_insertion(builder, function_location); assert(!compiler->temp_count); - compiler->temp_count = instruction->declaration.count; + compiler->temp_count = count; for (i = 0; i < compiler->temp_count; ++i) { - id = spirv_compiler_emit_variable(compiler, &builder->function_stream, - SpvStorageClassFunction, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE); + id = spirv_compiler_emit_variable(compiler, &builder->global_stream, + SpvStorageClassPrivate, VKD3D_SHADER_COMPONENT_FLOAT, VKD3D_VEC4_SIZE); if (!i) compiler->temp_id = id; assert(id == compiler->temp_id + i); @@ -6236,9 +6235,6 @@ static void spirv_compiler_leave_shader_phase(struct spirv_compiler *compiler) vkd3d_spirv_build_op_function_end(builder); - compiler->temp_id = 0; - compiler->temp_count = 0; - if (is_in_control_point_phase(compiler)) { if (compiler->epilogue_function_id) @@ -9103,9 +9099,6 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler, case VKD3DSIH_DCL_GLOBAL_FLAGS: spirv_compiler_emit_dcl_global_flags(compiler, instruction); break; - case VKD3DSIH_DCL_TEMPS: - spirv_compiler_emit_dcl_temps(compiler, instruction); - break; case VKD3DSIH_DCL_INDEXABLE_TEMP: spirv_compiler_emit_dcl_indexable_temp(compiler, instruction); break; @@ -9426,6 +9419,7 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler, spirv_compiler_emit_cut_stream(compiler, instruction); break; case VKD3DSIH_DCL_HS_MAX_TESSFACTOR: + case VKD3DSIH_DCL_TEMPS: case VKD3DSIH_HS_DECLS: case VKD3DSIH_NOP: /* nothing to do */ @@ -9448,6 +9442,9 @@ static int spirv_compiler_generate_spirv(struct spirv_compiler *compiler, enum vkd3d_result result = VKD3D_OK; unsigned int i; + if (parser->shader_desc.temp_count) + spirv_compiler_emit_temps(compiler, parser->shader_desc.temp_count); + compiler->location.column = 0; compiler->location.line = 1; diff --git a/libs/vkd3d-shader/tpf.c b/libs/vkd3d-shader/tpf.c index a465e0b7..290fdcb3 100644 --- a/libs/vkd3d-shader/tpf.c +++ b/libs/vkd3d-shader/tpf.c @@ -989,6 +989,8 @@ static void shader_sm4_read_declaration_count(struct vkd3d_shader_instruction *i uint32_t opcode_token, const uint32_t *tokens, unsigned int token_count, struct vkd3d_shader_sm4_parser *priv) { ins->declaration.count = *tokens; + if (opcode == VKD3D_SM4_OP_DCL_TEMPS) + priv->p.shader_desc.temp_count = max(priv->p.shader_desc.temp_count, *tokens); } static void shader_sm4_read_declaration_dst(struct vkd3d_shader_instruction *ins, uint32_t opcode, diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 398139c3..85fca964 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -829,6 +829,8 @@ struct vkd3d_shader_desc struct shader_signature input_signature; struct shader_signature output_signature; struct shader_signature patch_constant_signature; + + uint32_t temp_count; }; struct vkd3d_shader_register_semantic

1 0