May 2022 - Wine-commits - list.winehq.org

Nikolay Sivov : dwrite/opentype: Fix incorrect MarkGlyphSets table handling.
by Alexandre Julliard 02 May '22

02 May '22

Module: wine Branch: master Commit: d2d91fba5553c3c0d44c9e3ebf665961199def69 URL: https://source.winehq.org/git/wine.git/?a=commit;h=d2d91fba5553c3c0d44c9e3e… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Mon May 2 09:22:17 2022 +0300 dwrite/opentype: Fix incorrect MarkGlyphSets table handling. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/dwrite/opentype.c | 16 ++++++++++++---- 1 file changed, 12 insertions(+), 4 deletions(-) diff --git a/dlls/dwrite/opentype.c b/dlls/dwrite/opentype.c index 7df153e681d..f07b010baae 100644 --- a/dlls/dwrite/opentype.c +++ b/dlls/dwrite/opentype.c @@ -475,6 +475,13 @@ struct ot_gdef_classdef_format2 struct ot_gdef_class_range ranges[1]; }; +struct ot_gdef_markglyphsets +{ + uint16_t format; + uint16_t count; + uint32_t offsets[1]; +}; + struct gpos_gsub_header { uint16_t major_version; @@ -3762,7 +3769,7 @@ static BOOL opentype_match_coverage_func(UINT16 glyph, UINT16 glyph_data, const static BOOL opentype_layout_mark_set_covers(const struct scriptshaping_cache *cache, unsigned int set_index, UINT16 glyph) { - unsigned int format, offset = cache->gdef.markglyphsetdef, coverage_offset, set_count; + unsigned int format, offset = cache->gdef.markglyphsetdef, coverage_offset, count; if (!offset) return FALSE; @@ -3771,11 +3778,12 @@ static BOOL opentype_layout_mark_set_covers(const struct scriptshaping_cache *ca if (format == 1) { - set_count = table_read_be_word(&cache->gdef.table, offset + 2); - if (!set_count || set_index >= set_count) + count = table_read_be_word(&cache->gdef.table, offset + FIELD_OFFSET(struct ot_gdef_markglyphsets, count)); + if (!count || set_index >= count) return FALSE; - coverage_offset = table_read_be_dword(&cache->gdef.table, offset + 2 + set_index * sizeof(coverage_offset)); + coverage_offset = table_read_be_dword(&cache->gdef.table, offset + + FIELD_OFFSET(struct ot_gdef_markglyphsets, offsets[set_index])); return opentype_layout_is_glyph_covered(&cache->gdef.table, offset + coverage_offset, glyph) != GLYPH_NOT_COVERED; } else

1 0

Nikolay Sivov : dwrite/opentype: Do not include base advance to the mark advance offset.
by Alexandre Julliard 02 May '22

02 May '22

Module: wine Branch: master Commit: 979868aed5084a8cb96f015dba46b217ead58965 URL: https://source.winehq.org/git/wine.git/?a=commit;h=979868aed5084a8cb96f015d… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Mon May 2 09:22:18 2022 +0300 dwrite/opentype: Do not include base advance to the mark advance offset. This is handled later using attachment chains, after all positional features are applied. Signed-off-by: Nikolay Sivov <nsivov(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- dlls/dwrite/opentype.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/dlls/dwrite/opentype.c b/dlls/dwrite/opentype.c index f07b010baae..50e83769715 100644 --- a/dlls/dwrite/opentype.c +++ b/dlls/dwrite/opentype.c @@ -4320,7 +4320,7 @@ static BOOL opentype_layout_apply_mark_array(struct scriptshaping_context *conte if (context->is_rtl) context->offsets[context->cur].advanceOffset = mark_x - base_x; else - context->offsets[context->cur].advanceOffset = -context->advances[glyph_pos] + base_x - mark_x; + context->offsets[context->cur].advanceOffset = base_x - mark_x; context->offsets[context->cur].ascenderOffset = base_y - mark_y; opentype_set_glyph_attach_type(context, context->cur, GLYPH_ATTACH_MARK); context->glyph_infos[context->cur].attach_chain = (int)glyph_pos - (int)context->cur;

1 0

Francisco Casas : vkd3d-shader/hlsl: Detect missing loads on rhs when splitting copies of non-numeric types.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: a58c0e7c9d21a3ef2bc48884c2d374fdc3dc5e1d URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=a58c0e7c9d21a3ef2bc4888… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Thu Apr 28 15:31:58 2022 +0200 vkd3d-shader/hlsl: Detect missing loads on rhs when splitting copies of non-numeric types. Otherwise we can get failed assertions: assert(node->type == HLSL_IR_LOAD); because broadcasts to these types are not implemented yet. Signed-off-by: Francisco Casas <fcasas(a)codeweavers.com> Signed-off-by: Giovanni Mascellani <gmascellani(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl_codegen.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index e945b94d..f7396a96 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -637,6 +637,12 @@ static bool split_array_copies(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, element_type = type->e.array.type; element_size = hlsl_type_get_array_element_reg_size(element_type); + if (rhs->type != HLSL_IR_LOAD) + { + hlsl_fixme(ctx, &instr->loc, "Array store rhs is not HLSL_IR_LOAD. Broadcast may be missing."); + return false; + } + for (i = 0; i < type->e.array.elements_count; ++i) { if (!split_copy(ctx, store, hlsl_ir_load(rhs), i * element_size, element_type)) @@ -667,6 +673,12 @@ static bool split_struct_copies(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr if (type->type != HLSL_CLASS_STRUCT) return false; + if (rhs->type != HLSL_IR_LOAD) + { + hlsl_fixme(ctx, &instr->loc, "Struct store rhs is not HLSL_IR_LOAD. Broadcast may be missing."); + return false; + } + LIST_FOR_EACH_ENTRY(field, type->e.elements, struct hlsl_struct_field, entry) { if (!split_copy(ctx, store, hlsl_ir_load(rhs), field->reg_offset, field->type))

1 0

Francisco Casas : tests: Test complex broadcasts.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: 4b4537b874303b7dae0f3f2cc1c3225a0bdba9ca URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=4b4537b874303b7dae0f3f2… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Thu Apr 28 15:31:59 2022 +0200 tests: Test complex broadcasts. Signed-off-by: Francisco Casas <fcasas(a)codeweavers.com> Signed-off-by: Giovanni Mascellani <gmascellani(a)codeweavers.com> Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- Makefile.am | 1 + tests/cast-broadcast.shader_test | 24 ++++++++++++++++++++++++ 2 files changed, 25 insertions(+) diff --git a/Makefile.am b/Makefile.am index 26aa3f84..fa3c2a6d 100644 --- a/Makefile.am +++ b/Makefile.am @@ -56,6 +56,7 @@ vkd3d_shader_tests = \ tests/arithmetic-int.shader_test \ tests/arithmetic-uint.shader_test \ tests/bitwise.shader_test \ + tests/cast-broadcast.shader_test \ tests/cast-to-float.shader_test \ tests/cast-to-half.shader_test \ tests/cast-to-int.shader_test \ diff --git a/tests/cast-broadcast.shader_test b/tests/cast-broadcast.shader_test new file mode 100644 index 00000000..02d14c0b --- /dev/null +++ b/tests/cast-broadcast.shader_test @@ -0,0 +1,24 @@ +[pixel shader] + +struct foo +{ + float3 aa; + float4 bb; +}; + +struct bar +{ + struct foo aa; + int2 bb; + int4 cc[8]; +}; + +float4 main() : SV_TARGET +{ + struct bar p = (struct bar)42; + return p.aa.bb + p.cc[5]; +} + +[test] +todo draw quad +todo probe all rgba (84.0, 84.0, 84.0, 84.0)

1 0

Francisco Casas : tests: Test initialization of implicit size arrays.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: 1240a947d227d0440d75286bf25d86b937091a4c URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=1240a947d227d0440d75286… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Thu Apr 28 15:32:00 2022 +0200 tests: Test initialization of implicit size arrays. Signed-off-by: Francisco Casas <fcasas(a)codeweavers.com> Signed-off-by: Giovanni Mascellani <gmascellani(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- Makefile.am | 1 + tests/hlsl-initializer-implicit-array.shader_test | 18 ++++++++++++++++++ 2 files changed, 19 insertions(+) diff --git a/Makefile.am b/Makefile.am index fa3c2a6d..2742e77c 100644 --- a/Makefile.am +++ b/Makefile.am @@ -77,6 +77,7 @@ vkd3d_shader_tests = \ tests/hlsl-gather.shader_test \ tests/hlsl-initializer-flatten.shader_test \ tests/hlsl-initializer-invalid-arg-count.shader_test \ + tests/hlsl-initializer-implicit-array.shader_test \ tests/hlsl-initializer-local-array.shader_test \ tests/hlsl-initializer-nested.shader_test \ tests/hlsl-initializer-numeric.shader_test \ diff --git a/tests/hlsl-initializer-implicit-array.shader_test b/tests/hlsl-initializer-implicit-array.shader_test new file mode 100644 index 00000000..d2b94da4 --- /dev/null +++ b/tests/hlsl-initializer-implicit-array.shader_test @@ -0,0 +1,18 @@ +[pixel shader] +float4 main() : SV_TARGET +{ + float4 arr[] = {10, 20, 30, 40, 50, 60, 70, 80}; + return arr[1]; +} + +[test] +todo draw quad +todo probe all rgba (50, 60, 70, 80) + + +[pixel shader fail] +float4 main() : SV_TARGET +{ + float4 arr[] = {10, 20, 30, 40, 50, 60, 70}; + return arr[0]; +}

1 0

Giovanni Mascellani : vkd3d-shader/hlsl: Allow storing to matrices.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: 38d7913f5f573bbf7a5fd6af672c552219d005dd URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=38d7913f5f573bbf7a5fd6a… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Thu Apr 28 15:32:04 2022 +0200 vkd3d-shader/hlsl: Allow storing to matrices. Signed-off-by: Giovanni Mascellani <gmascellani(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl_sm4.c | 6 ------ tests/hlsl-initializer-matrix.shader_test | 2 +- tests/hlsl-matrix-indexing.shader_test | 2 +- tests/hlsl-numeric-types.shader_test | 4 ++-- 4 files changed, 4 insertions(+), 10 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c index 8afa9333..a5fc094b 100644 --- a/libs/vkd3d-shader/hlsl_sm4.c +++ b/libs/vkd3d-shader/hlsl_sm4.c @@ -2047,12 +2047,6 @@ static void write_sm4_store(struct hlsl_ctx *ctx, struct sm4_instruction instr; unsigned int writemask; - if (store->lhs.var->data_type->type == HLSL_CLASS_MATRIX) - { - hlsl_fixme(ctx, &store->node.loc, "Store to a matrix variable."); - return; - } - memset(&instr, 0, sizeof(instr)); instr.opcode = VKD3D_SM4_OP_MOV; diff --git a/tests/hlsl-initializer-matrix.shader_test b/tests/hlsl-initializer-matrix.shader_test index 790d402d..ea9de9c0 100644 --- a/tests/hlsl-initializer-matrix.shader_test +++ b/tests/hlsl-initializer-matrix.shader_test @@ -68,5 +68,5 @@ float4 main() : SV_TARGET } [test] -todo draw quad +draw quad probe all rgba (21, 22, 31, 32) diff --git a/tests/hlsl-matrix-indexing.shader_test b/tests/hlsl-matrix-indexing.shader_test index 9336a67a..716b43d4 100644 --- a/tests/hlsl-matrix-indexing.shader_test +++ b/tests/hlsl-matrix-indexing.shader_test @@ -74,5 +74,5 @@ float4 main() : SV_TARGET } [test] -todo draw quad +draw quad probe all rgba (1.0, 5.0, 7.0, 12.0) diff --git a/tests/hlsl-numeric-types.shader_test b/tests/hlsl-numeric-types.shader_test index 58612191..7504f95a 100644 --- a/tests/hlsl-numeric-types.shader_test +++ b/tests/hlsl-numeric-types.shader_test @@ -56,7 +56,7 @@ float4 main() : sv_target } [test] -todo draw quad +draw quad probe all rgba (5.0, 6.0, 7.0, 8.0) [pixel shader] @@ -68,7 +68,7 @@ float4 main() : sv_target } [test] -todo draw quad +draw quad probe all rgba (5.0, 6.0, 7.0, 0.0) [pixel shader fail]

1 0

Giovanni Mascellani : vkd3d-shader/hlsl: Handle branches in copy propagation.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: 6167b3a4e28c6b7be5c7ae02062660dcfa29ca0f URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=6167b3a4e28c6b7be5c7ae0… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Thu Apr 28 15:32:05 2022 +0200 vkd3d-shader/hlsl: Handle branches in copy propagation. Signed-off-by: Giovanni Mascellani <gmascellani(a)codeweavers.com> Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d-shader/hlsl_codegen.c | 179 ++++++++++++++++++++++++++++++++++----- 1 file changed, 159 insertions(+), 20 deletions(-) Diff: https://source.winehq.org/git/vkd3d.git/?a=commitdiff;h=6167b3a4e28c6b7be5c…

1 0

Conor McCarthy : vkd3d: Introduce an internal refcount to d3d12_fence to replace the thread waiting mechanism.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: 34e7b87966beeb7c4148b0ab535096fcf6cd1b9e URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=34e7b87966beeb7c4148b0a… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Apr 29 23:21:07 2022 +1000 vkd3d: Introduce an internal refcount to d3d12_fence to replace the thread waiting mechanism. Simplifies the preservation of fence objects until worker threads are done with them, and will be needed when threaded queue submission is added. Signed-off-by: Conor McCarthy <cmccarthy(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/command.c | 72 +++++++++++++++------------------------------- libs/vkd3d/vkd3d_private.h | 4 +-- 2 files changed, 24 insertions(+), 52 deletions(-) diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index 09171fe4..7a373b34 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -20,6 +20,8 @@ #include "vkd3d_private.h" +static void d3d12_fence_incref(struct d3d12_fence *fence); +static void d3d12_fence_decref(struct d3d12_fence *fence); static HRESULT d3d12_fence_signal(struct d3d12_fence *fence, uint64_t value, VkFence vk_fence); HRESULT vkd3d_queue_create(struct d3d12_device *device, @@ -295,7 +297,7 @@ static HRESULT vkd3d_enqueue_gpu_fence(struct vkd3d_fence_worker *worker, waiting_fence->queue_sequence_number = queue_sequence_number; ++worker->enqueued_fence_count; - InterlockedIncrement(&fence->pending_worker_operation_count); + d3d12_fence_incref(fence); vkd3d_cond_signal(&worker->cond); vkd3d_mutex_unlock(&worker->mutex); @@ -303,37 +305,6 @@ static HRESULT vkd3d_enqueue_gpu_fence(struct vkd3d_fence_worker *worker, return S_OK; } -static void vkd3d_fence_worker_remove_fence(struct vkd3d_fence_worker *worker, struct d3d12_fence *fence) -{ - LONG count; - int rc; - - if (!(count = InterlockedAdd(&fence->pending_worker_operation_count, 0))) - return; - - WARN("Waiting for %u pending fence operations (fence %p).\n", count, fence); - - if ((rc = vkd3d_mutex_lock(&worker->mutex))) - { - ERR("Failed to lock mutex, error %d.\n", rc); - return; - } - - while ((count = InterlockedAdd(&fence->pending_worker_operation_count, 0))) - { - TRACE("Still waiting for %u pending fence operations (fence %p).\n", count, fence); - - worker->pending_fence_destruction = true; - vkd3d_cond_signal(&worker->cond); - - vkd3d_cond_wait(&worker->fence_destruction_cond, &worker->mutex); - } - - TRACE("Removed fence %p.\n", fence); - - vkd3d_mutex_unlock(&worker->mutex); -} - static void vkd3d_fence_worker_move_enqueued_fences_locked(struct vkd3d_fence_worker *worker) { unsigned int i; @@ -432,7 +403,7 @@ static void vkd3d_wait_for_gpu_timeline_semaphores(struct vkd3d_fence_worker *wo if (FAILED(hr = d3d12_fence_signal(current->fence, counter_value, VK_NULL_HANDLE))) ERR("Failed to signal D3D12 fence, hr %#x.\n", hr); - InterlockedDecrement(&current->fence->pending_worker_operation_count); + d3d12_fence_decref(current->fence); continue; } @@ -480,7 +451,7 @@ static void vkd3d_wait_for_gpu_fences(struct vkd3d_fence_worker *worker) if (FAILED(hr = d3d12_fence_signal(current->fence, current->value, vk_fence))) ERR("Failed to signal D3D12 fence, hr %#x.\n", hr); - InterlockedDecrement(&current->fence->pending_worker_operation_count); + d3d12_fence_decref(current->fence); vkd3d_queue_update_sequence_number(current->queue, current->queue_sequence_number, device); continue; @@ -518,12 +489,6 @@ static void *vkd3d_fence_worker_main(void *arg) break; } - if (worker->pending_fence_destruction) - { - vkd3d_cond_broadcast(&worker->fence_destruction_cond); - worker->pending_fence_destruction = false; - } - if (worker->enqueued_fence_count) { vkd3d_fence_worker_move_enqueued_fences_locked(worker); @@ -560,7 +525,6 @@ HRESULT vkd3d_fence_worker_start(struct vkd3d_fence_worker *worker, TRACE("worker %p.\n", worker); worker->should_exit = false; - worker->pending_fence_destruction = false; worker->device = device; worker->enqueued_fence_count = 0; @@ -1026,22 +990,35 @@ static ULONG STDMETHODCALLTYPE d3d12_fence_AddRef(ID3D12Fence *iface) return refcount; } +static void d3d12_fence_incref(struct d3d12_fence *fence) +{ + InterlockedIncrement(&fence->internal_refcount); +} + static ULONG STDMETHODCALLTYPE d3d12_fence_Release(ID3D12Fence *iface) { struct d3d12_fence *fence = impl_from_ID3D12Fence(iface); ULONG refcount = InterlockedDecrement(&fence->refcount); - int rc; TRACE("%p decreasing refcount to %u.\n", fence, refcount); if (!refcount) + d3d12_fence_decref(fence); + + return refcount; +} + +static void d3d12_fence_decref(struct d3d12_fence *fence) +{ + ULONG internal_refcount = InterlockedDecrement(&fence->internal_refcount); + int rc; + + if (!internal_refcount) { struct d3d12_device *device = fence->device; vkd3d_private_store_destroy(&fence->private_store); - vkd3d_fence_worker_remove_fence(&device->fence_worker, fence); - d3d12_fence_destroy_vk_objects(fence); vkd3d_free(fence->events); @@ -1052,8 +1029,6 @@ static ULONG STDMETHODCALLTYPE d3d12_fence_Release(ID3D12Fence *iface) d3d12_device_release(device); } - - return refcount; } static HRESULT STDMETHODCALLTYPE d3d12_fence_GetPrivateData(ID3D12Fence *iface, @@ -1380,6 +1355,7 @@ static HRESULT d3d12_fence_init(struct d3d12_fence *fence, struct d3d12_device * int rc; fence->ID3D12Fence_iface.lpVtbl = &d3d12_fence_vtbl; + fence->internal_refcount = 1; fence->refcount = 1; fence->value = initial_value; @@ -1419,8 +1395,6 @@ static HRESULT d3d12_fence_init(struct d3d12_fence *fence, struct d3d12_device * memset(fence->old_vk_fences, 0, sizeof(fence->old_vk_fences)); - fence->pending_worker_operation_count = 0; - if (FAILED(hr = vkd3d_private_store_init(&fence->private_store))) { vkd3d_mutex_destroy(&fence->mutex); @@ -6496,7 +6470,7 @@ static HRESULT vkd3d_enqueue_timeline_semaphore(struct vkd3d_fence_worker *worke waiting_fence->queue = queue; ++worker->enqueued_fence_count; - InterlockedIncrement(&fence->pending_worker_operation_count); + d3d12_fence_incref(fence); vkd3d_cond_signal(&worker->cond); vkd3d_mutex_unlock(&worker->mutex); diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index 56060b6d..a0163c8d 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -346,7 +346,6 @@ struct vkd3d_fence_worker struct vkd3d_cond cond; struct vkd3d_cond fence_destruction_cond; bool should_exit; - bool pending_fence_destruction; LONG enqueued_fence_count; struct vkd3d_enqueued_fence @@ -532,6 +531,7 @@ struct vkd3d_pending_fence_wait struct d3d12_fence { ID3D12Fence ID3D12Fence_iface; + LONG internal_refcount; LONG refcount; uint64_t value; @@ -555,8 +555,6 @@ struct d3d12_fence struct list semaphores; unsigned int semaphore_count; - LONG pending_worker_operation_count; - VkFence old_vk_fences[VKD3D_MAX_VK_SYNC_OBJECTS]; struct d3d12_device *device;

1 0

Conor McCarthy : vkd3d: Create one fence worker thread per command queue.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: 488722b9b52abdd5f4c057ba9eb6e3f92b897f0a URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=488722b9b52abdd5f4c057b… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Apr 29 23:21:08 2022 +1000 vkd3d: Create one fence worker thread per command queue. Simplifies the handling of GPU waits, and in vkd3d-proton is reported to increase performance when support for multiple Vulkan queues is enabled, because it avoids the problem of fences being signaled while they sit in the pending buffer waiting to be moved to the wait buffer. Based on a vkd3d-proton patch by Philip Rebohle. Signed-off-by: Conor McCarthy <cmccarthy(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/command.c | 274 ++++++++++++++------------------------------- libs/vkd3d/device.c | 8 +- libs/vkd3d/vkd3d_private.h | 29 ++--- 3 files changed, 96 insertions(+), 215 deletions(-) Diff: https://source.winehq.org/git/vkd3d.git/?a=commitdiff;h=488722b9b52abdd5f4c…

1 0

Conor McCarthy : vkd3d: Replace the signaled semaphore list with a resizable array.
by Alexandre Julliard 02 May '22

02 May '22

Module: vkd3d Branch: master Commit: 07e38212ec950d91d59fc5f3220d05314c470c49 URL: https://source.winehq.org/git/vkd3d.git/?a=commit;h=07e38212ec950d91d59fc5f… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Apr 29 23:21:09 2022 +1000 vkd3d: Replace the signaled semaphore list with a resizable array. Order does not need to be preserved here, and another function will add to this array when mapped timeline semaphores are implemented. Signed-off-by: Conor McCarthy <cmccarthy(a)codeweavers.com> Signed-off-by: Henri Verbeet <hverbeet(a)codeweavers.com> Signed-off-by: Alexandre Julliard <julliard(a)winehq.org> --- libs/vkd3d/command.c | 44 +++++++++++++++++++++++++++----------------- libs/vkd3d/vkd3d_private.h | 4 ++-- 2 files changed, 29 insertions(+), 19 deletions(-) diff --git a/libs/vkd3d/command.c b/libs/vkd3d/command.c index 7690760c..55e6be58 100644 --- a/libs/vkd3d/command.c +++ b/libs/vkd3d/command.c @@ -590,32 +590,34 @@ static void d3d12_fence_garbage_collect_vk_semaphores_locked(struct d3d12_fence { struct d3d12_device *device = fence->device; const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs; - struct vkd3d_signaled_semaphore *current, *p; - unsigned int semaphore_count; + struct vkd3d_signaled_semaphore *current; + unsigned int i, semaphore_count; semaphore_count = fence->semaphore_count; if (!destroy_all && semaphore_count < VKD3D_MAX_VK_SYNC_OBJECTS) return; - LIST_FOR_EACH_ENTRY_SAFE(current, p, &fence->semaphores, struct vkd3d_signaled_semaphore, entry) + i = 0; + while (i < fence->semaphore_count) { if (!destroy_all && fence->semaphore_count < VKD3D_MAX_VK_SYNC_OBJECTS) break; + current = &fence->semaphores[i]; /* The semaphore doesn't have a pending signal operation if the fence * was signaled. */ if ((current->vk_fence || current->is_acquired) && !destroy_all) + { + ++i; continue; + } if (current->vk_fence) WARN("Destroying potentially pending semaphore.\n"); assert(!current->is_acquired); VK_CALL(vkDestroySemaphore(device->vk_device, current->vk_semaphore, NULL)); - list_remove(&current->entry); - vkd3d_free(current); - - --fence->semaphore_count; + fence->semaphores[i] = fence->semaphores[--fence->semaphore_count]; } if (semaphore_count != fence->semaphore_count) @@ -656,6 +658,7 @@ static struct vkd3d_signaled_semaphore *d3d12_fence_acquire_vk_semaphore(struct struct vkd3d_signaled_semaphore *semaphore; struct vkd3d_signaled_semaphore *current; uint64_t semaphore_value; + unsigned int i; int rc; TRACE("fence %p, value %#"PRIx64".\n", fence, value); @@ -669,8 +672,9 @@ static struct vkd3d_signaled_semaphore *d3d12_fence_acquire_vk_semaphore(struct semaphore = NULL; semaphore_value = ~(uint64_t)0; - LIST_FOR_EACH_ENTRY(current, &fence->semaphores, struct vkd3d_signaled_semaphore, entry) + for (i = 0; i < fence->semaphore_count; ++i) { + current = &fence->semaphores[i]; /* Prefer a semaphore with the smallest value. */ if (!current->is_acquired && current->value >= value && semaphore_value >= current->value) { @@ -703,10 +707,7 @@ static void d3d12_fence_remove_vk_semaphore(struct d3d12_fence *fence, struct vk assert(semaphore->is_acquired); - list_remove(&semaphore->entry); - vkd3d_free(semaphore); - - --fence->semaphore_count; + *semaphore = fence->semaphores[--fence->semaphore_count]; vkd3d_mutex_unlock(&fence->mutex); } @@ -751,14 +752,20 @@ static HRESULT d3d12_fence_add_vk_semaphore(struct d3d12_fence *fence, d3d12_fence_garbage_collect_vk_semaphores_locked(fence, false); + if (!vkd3d_array_reserve((void**)&fence->semaphores, &fence->semaphores_size, + fence->semaphore_count + 1, sizeof(*fence->semaphores))) + { + ERR("Failed to add semaphore.\n"); + vkd3d_mutex_unlock(&fence->mutex); + return false; + } + + semaphore = &fence->semaphores[fence->semaphore_count++]; semaphore->value = value; semaphore->vk_semaphore = vk_semaphore; semaphore->vk_fence = vk_fence; semaphore->is_acquired = false; - list_add_tail(&fence->semaphores, &semaphore->entry); - ++fence->semaphore_count; - vkd3d_mutex_unlock(&fence->mutex); return hr; @@ -821,8 +828,9 @@ static HRESULT d3d12_fence_signal(struct d3d12_fence *fence, uint64_t value, VkF { const struct vkd3d_vk_device_procs *vk_procs = &device->vk_procs; - LIST_FOR_EACH_ENTRY(current, &fence->semaphores, struct vkd3d_signaled_semaphore, entry) + for (i = 0; i < fence->semaphore_count; ++i) { + current = &fence->semaphores[i]; if (current->vk_fence == vk_fence) current->vk_fence = VK_NULL_HANDLE; } @@ -910,6 +918,7 @@ static void d3d12_fence_decref(struct d3d12_fence *fence) d3d12_fence_destroy_vk_objects(fence); vkd3d_free(fence->events); + vkd3d_free(fence->semaphores); if ((rc = vkd3d_mutex_destroy(&fence->mutex))) ERR("Failed to destroy mutex, error %d.\n", rc); vkd3d_cond_destroy(&fence->null_event_cond); @@ -1278,7 +1287,8 @@ static HRESULT d3d12_fence_init(struct d3d12_fence *fence, struct d3d12_device * fence->pending_timeline_value = initial_value; fence->gpu_wait_count = 0; - list_init(&fence->semaphores); + fence->semaphores = NULL; + fence->semaphores_size = 0; fence->semaphore_count = 0; memset(fence->old_vk_fences, 0, sizeof(fence->old_vk_fences)); diff --git a/libs/vkd3d/vkd3d_private.h b/libs/vkd3d/vkd3d_private.h index 350382cd..4e03145d 100644 --- a/libs/vkd3d/vkd3d_private.h +++ b/libs/vkd3d/vkd3d_private.h @@ -501,7 +501,6 @@ HRESULT vkd3d_set_private_data_interface(struct vkd3d_private_store *store, cons struct vkd3d_signaled_semaphore { - struct list entry; uint64_t value; VkSemaphore vk_semaphore; VkFence vk_fence; @@ -539,7 +538,8 @@ struct d3d12_fence struct vkd3d_pending_fence_wait gpu_waits[VKD3D_MAX_FENCE_WAITING_QUEUES]; unsigned int gpu_wait_count; - struct list semaphores; + struct vkd3d_signaled_semaphore *semaphores; + size_t semaphores_size; unsigned int semaphore_count; VkFence old_vk_fences[VKD3D_MAX_VK_SYNC_OBJECTS];

1 0