- Wine-Devel - list.winehq.org

[PATCH 1/7] winecoreaudio: Use NtAllocateVirtualMemory() for local and tmp buffers.
by Huw Davies Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Huw Davies <huw(a)codeweavers.com> --- dlls/winecoreaudio.drv/mmdevdrv.c | 28 ++++++++++++++++++++-------- dlls/winecoreaudio.drv/unixlib.h | 2 +- 2 files changed, 21 insertions(+), 9 deletions(-) diff --git a/dlls/winecoreaudio.drv/mmdevdrv.c b/dlls/winecoreaudio.drv/mmdevdrv.c index 1e37a2a9026..642610b4b60 100644 --- a/dlls/winecoreaudio.drv/mmdevdrv.c +++ b/dlls/winecoreaudio.drv/mmdevdrv.c @@ -653,9 +653,13 @@ static ULONG WINAPI AudioClient_Release(IAudioClient3 *iface) LeaveCriticalSection(&g_sessions_lock); } HeapFree(GetProcessHeap(), 0, This->vols); - HeapFree(GetProcessHeap(), 0, This->stream->tmp_buffer); HeapFree(GetProcessHeap(), 0, This->stream->cap_buffer); - HeapFree(GetProcessHeap(), 0, This->stream->local_buffer); + if(This->stream->local_buffer) + NtFreeVirtualMemory(GetCurrentProcess(), (void **)&This->stream->local_buffer, + &This->stream->local_buffer_size, MEM_RELEASE); + if(This->stream->tmp_buffer) + NtFreeVirtualMemory(GetCurrentProcess(), (void **)&This->stream->tmp_buffer, + &This->stream->tmp_buffer_size, MEM_RELEASE); free(This->stream->wrap_buffer); HeapFree(GetProcessHeap(), 0, This->stream->resamp_buffer); CoTaskMemFree(This->stream->fmt); @@ -1335,7 +1339,9 @@ static HRESULT WINAPI AudioClient_Initialize(IAudioClient3 *iface, return osstatus_to_hresult(sc); } - This->stream->local_buffer = HeapAlloc(GetProcessHeap(), 0, This->stream->bufsize_frames * fmt->nBlockAlign); + This->stream->local_buffer_size = This->stream->bufsize_frames * fmt->nBlockAlign; + NtAllocateVirtualMemory(GetCurrentProcess(), (void **)&This->stream->local_buffer, 0, + &This->stream->local_buffer_size, MEM_COMMIT, PAGE_READWRITE); silence_buffer(This, This->stream->local_buffer, This->stream->bufsize_frames); if(This->dataflow == eCapture){ @@ -2292,9 +2298,12 @@ static HRESULT WINAPI AudioRenderClient_GetBuffer(IAudioRenderClient *iface, if(This->stream->wri_offs_frames + frames > This->stream->bufsize_frames){ if(This->stream->tmp_buffer_frames < frames){ - HeapFree(GetProcessHeap(), 0, This->stream->tmp_buffer); - This->stream->tmp_buffer = HeapAlloc(GetProcessHeap(), 0, frames * This->stream->fmt->nBlockAlign); - if(!This->stream->tmp_buffer){ + NtFreeVirtualMemory(GetCurrentProcess(), (void **)&This->stream->tmp_buffer, + &This->stream->tmp_buffer_size, MEM_RELEASE); + This->stream->tmp_buffer_size = frames * This->stream->fmt->nBlockAlign; + if(NtAllocateVirtualMemory(GetCurrentProcess(), (void **)&This->stream->tmp_buffer, 0, + &This->stream->tmp_buffer_size, MEM_COMMIT, PAGE_READWRITE)){ + This->stream->tmp_buffer_frames = 0; OSSpinLockUnlock(&This->stream->lock); return E_OUTOFMEMORY; } @@ -2449,8 +2458,11 @@ static HRESULT WINAPI AudioCaptureClient_GetBuffer(IAudioCaptureClient *iface, chunk_frames = This->stream->bufsize_frames - This->stream->lcl_offs_frames; if(chunk_frames < This->stream->period_frames){ chunk_bytes = chunk_frames * This->stream->fmt->nBlockAlign; - if(!This->stream->tmp_buffer) - This->stream->tmp_buffer = HeapAlloc(GetProcessHeap(), 0, This->stream->period_frames * This->stream->fmt->nBlockAlign); + if(!This->stream->tmp_buffer){ + This->stream->tmp_buffer_size = This->stream->period_frames * This->stream->fmt->nBlockAlign; + NtAllocateVirtualMemory(GetCurrentProcess(), (void **)&This->stream->tmp_buffer, 0, + &This->stream->tmp_buffer_size, MEM_COMMIT, PAGE_READWRITE); + } *data = This->stream->tmp_buffer; memcpy(*data, This->stream->local_buffer + This->stream->lcl_offs_frames * This->stream->fmt->nBlockAlign, chunk_bytes); memcpy((*data) + chunk_bytes, This->stream->local_buffer, This->stream->period_frames * This->stream->fmt->nBlockAlign - chunk_bytes); diff --git a/dlls/winecoreaudio.drv/unixlib.h b/dlls/winecoreaudio.drv/unixlib.h index 6a991f1de98..1b773b7f820 100644 --- a/dlls/winecoreaudio.drv/unixlib.h +++ b/dlls/winecoreaudio.drv/unixlib.h @@ -29,7 +29,6 @@ struct coreaudio_stream /* To be made private */ AUDCLNT_SHAREMODE share; BOOL playing; - SIZE_T local_buffer_size, tmp_buffer_size; UINT32 period_ms, period_frames; UINT32 bufsize_frames, resamp_bufsize_frames; UINT32 lcl_offs_frames, held_frames, wri_offs_frames, tmp_buffer_frames; @@ -39,6 +38,7 @@ struct coreaudio_stream /* To be made private */ INT32 getbuf_last; WAVEFORMATEX *fmt; BYTE *local_buffer, *cap_buffer, *wrap_buffer, *resamp_buffer, *tmp_buffer; + SIZE_T local_buffer_size, tmp_buffer_size; }; struct endpoint -- 2.23.0

2 1

[PATCH 1/6] wined3d: Move the "buffer_offset" field to struct wined3d_bo.
by Zebediah Figura Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- dlls/wined3d/adapter_vk.c | 14 +++++++------- dlls/wined3d/buffer.c | 4 ++-- dlls/wined3d/context_gl.c | 8 ++++---- dlls/wined3d/context_vk.c | 10 +++++----- dlls/wined3d/device.c | 6 +++--- dlls/wined3d/state.c | 4 ++-- dlls/wined3d/texture.c | 8 ++++---- dlls/wined3d/view.c | 8 ++++---- dlls/wined3d/wined3d_private.h | 3 +-- 9 files changed, 32 insertions(+), 33 deletions(-) diff --git a/dlls/wined3d/adapter_vk.c b/dlls/wined3d/adapter_vk.c index e6f5bf108c7..f9a31265cd3 100644 --- a/dlls/wined3d/adapter_vk.c +++ b/dlls/wined3d/adapter_vk.c @@ -1011,7 +1011,7 @@ static void *adapter_vk_map_bo_address(struct wined3d_context *context, vk_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.buffer = bo->vk_buffer; - vk_barrier.offset = bo->buffer_offset + (uintptr_t)data->addr; + vk_barrier.offset = bo->b.buffer_offset + (uintptr_t)data->addr; vk_barrier.size = size; VK_CALL(vkCmdPipelineBarrier(vk_command_buffer, VK_PIPELINE_STAGE_ALL_COMMANDS_BIT, VK_PIPELINE_STAGE_HOST_BIT, 0, 0, NULL, 1, &vk_barrier, 0, NULL)); @@ -1113,8 +1113,8 @@ void adapter_vk_copy_bo_address(struct wined3d_context *context, src_access_mask = vk_access_mask_from_buffer_usage(src_bo->usage); dst_access_mask = vk_access_mask_from_buffer_usage(dst_bo->usage); - region.srcOffset = src_bo->buffer_offset + (uintptr_t)src->addr; - region.dstOffset = dst_bo->buffer_offset + (uintptr_t)dst->addr; + region.srcOffset = src_bo->b.buffer_offset + (uintptr_t)src->addr; + region.dstOffset = dst_bo->b.buffer_offset + (uintptr_t)dst->addr; region.size = size; vk_barrier[0].sType = VK_STRUCTURE_TYPE_BUFFER_MEMORY_BARRIER; @@ -1781,7 +1781,7 @@ static void adapter_vk_draw_primitive(struct wined3d_device *device, for (i = 0; i < ARRAY_SIZE(context_vk->vk_so_counters); ++i) { context_vk->vk_so_counters[i] = bo->vk_buffer; - context_vk->vk_so_offsets[i] = bo->buffer_offset + i * sizeof(uint32_t) * 2; + context_vk->vk_so_offsets[i] = bo->b.buffer_offset + i * sizeof(uint32_t) * 2; } } @@ -1805,13 +1805,13 @@ static void adapter_vk_draw_primitive(struct wined3d_device *device, { stride = sizeof(VkDrawIndexedIndirectCommand); VK_CALL(vkCmdDrawIndexedIndirect(vk_command_buffer, bo->vk_buffer, - bo->buffer_offset + parameters->u.indirect.offset, size / stride, stride)); + bo->b.buffer_offset + parameters->u.indirect.offset, size / stride, stride)); } else { stride = sizeof(VkDrawIndirectCommand); VK_CALL(vkCmdDrawIndirect(vk_command_buffer, bo->vk_buffer, - bo->buffer_offset + parameters->u.indirect.offset, size / stride, stride)); + bo->b.buffer_offset + parameters->u.indirect.offset, size / stride, stride)); } } else @@ -1871,7 +1871,7 @@ static void adapter_vk_dispatch_compute(struct wined3d_device *device, wined3d_context_vk_reference_bo(context_vk, bo); VK_CALL(vkCmdDispatchIndirect(vk_command_buffer, bo->vk_buffer, - bo->buffer_offset + parameters->u.indirect.offset)); + bo->b.buffer_offset + parameters->u.indirect.offset)); } else { diff --git a/dlls/wined3d/buffer.c b/dlls/wined3d/buffer.c index 315ea05260e..d3aeb2e76f4 100644 --- a/dlls/wined3d/buffer.c +++ b/dlls/wined3d/buffer.c @@ -1521,7 +1521,7 @@ const VkDescriptorBufferInfo *wined3d_buffer_vk_get_buffer_info(struct wined3d_b return &buffer_vk->buffer_info; buffer_vk->buffer_info.buffer = bo->vk_buffer; - buffer_vk->buffer_info.offset = bo->buffer_offset; + buffer_vk->buffer_info.offset = bo->b.buffer_offset; buffer_vk->buffer_info.range = buffer_vk->b.resource.size; buffer_vk->b.bo_user.valid = true; @@ -1702,7 +1702,7 @@ void wined3d_buffer_vk_barrier(struct wined3d_buffer_vk *buffer_vk, vk_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.buffer = bo->vk_buffer; - vk_barrier.offset = bo->buffer_offset; + vk_barrier.offset = bo->b.buffer_offset; vk_barrier.size = buffer_vk->b.resource.size; VK_CALL(vkCmdPipelineBarrier(wined3d_context_vk_get_command_buffer(context_vk), vk_pipeline_stage_mask_from_bind_flags(src_bind_mask), diff --git a/dlls/wined3d/context_gl.c b/dlls/wined3d/context_gl.c index 2751a4a2182..69eb13f3027 100644 --- a/dlls/wined3d/context_gl.c +++ b/dlls/wined3d/context_gl.c @@ -2922,7 +2922,7 @@ bool wined3d_context_gl_create_bo(struct wined3d_context_gl *context_gl, GLsizei list_init(&bo->b.users); bo->command_fence_id = 0; bo->b.memory_offset = 0; - bo->buffer_offset = 0; + bo->b.buffer_offset = 0; bo->b.map_ptr = NULL; return true; @@ -3789,7 +3789,7 @@ static void wined3d_context_gl_bind_unordered_access_views(struct wined3d_contex if (view_gl->counter_bo.id) GL_EXTCALL(glBindBufferRange(GL_ATOMIC_COUNTER_BUFFER, i, view_gl->counter_bo.id, - view_gl->counter_bo.buffer_offset, view_gl->counter_bo.size)); + view_gl->counter_bo.b.buffer_offset, view_gl->counter_bo.size)); } checkGLcall("Bind unordered access views"); } @@ -4915,7 +4915,7 @@ void draw_primitive(struct wined3d_device *device, const struct wined3d_state *s if (!bo || !stream_info->all_vbo) idx_data = index_buffer->resource.heap_memory; else - idx_data = (void *)wined3d_bo_gl(bo)->buffer_offset; + idx_data = (void *)bo->buffer_offset; idx_data = (const BYTE *)idx_data + state->index_offset; if (state->index_format == WINED3DFMT_R16_UINT) @@ -5077,7 +5077,7 @@ static const void *get_vertex_attrib_pointer(const struct wined3d_stream_info_el const uint8_t *offset = element->data.addr + state->load_base_vertex_index * element->stride; if (element->data.buffer_object) - offset += wined3d_bo_gl(element->data.buffer_object)->buffer_offset; + offset += element->data.buffer_object->buffer_offset; return offset; } diff --git a/dlls/wined3d/context_vk.c b/dlls/wined3d/context_vk.c index ec3554ddc30..56b2da53060 100644 --- a/dlls/wined3d/context_vk.c +++ b/dlls/wined3d/context_vk.c @@ -456,8 +456,8 @@ static bool wined3d_context_vk_create_slab_bo(struct wined3d_context_vk *context *bo = slab->bo; bo->memory = NULL; bo->slab = slab; - bo->buffer_offset = idx * object_size; - bo->b.memory_offset = slab->bo.b.memory_offset + bo->buffer_offset; + bo->b.buffer_offset = idx * object_size; + bo->b.memory_offset = slab->bo.b.memory_offset + bo->b.buffer_offset; bo->size = size; list_init(&bo->b.users); bo->command_buffer_id = 0; @@ -465,7 +465,7 @@ static bool wined3d_context_vk_create_slab_bo(struct wined3d_context_vk *context TRACE("Using buffer 0x%s, memory 0x%s, offset 0x%s for bo %p.\n", wine_dbgstr_longlong(bo->vk_buffer), wine_dbgstr_longlong(bo->vk_memory), - wine_dbgstr_longlong(bo->buffer_offset), bo); + wine_dbgstr_longlong(bo->b.buffer_offset), bo); return true; } @@ -534,7 +534,7 @@ BOOL wined3d_context_vk_create_bo(struct wined3d_context_vk *context_vk, VkDevic } bo->b.map_ptr = NULL; - bo->buffer_offset = 0; + bo->b.buffer_offset = 0; bo->size = size; bo->usage = usage; bo->memory_type = adapter_vk->memory_properties.memoryTypes[memory_type_idx].propertyFlags; @@ -961,7 +961,7 @@ void wined3d_context_vk_destroy_bo(struct wined3d_context_vk *context_vk, const if (bo->b.map_ptr) wined3d_bo_slab_vk_unmap(slab_vk, context_vk); object_size = slab_vk->bo.size / 32; - idx = bo->buffer_offset / object_size; + idx = bo->b.buffer_offset / object_size; wined3d_context_vk_destroy_bo_slab_slice(context_vk, slab_vk, idx, bo->command_buffer_id); return; } diff --git a/dlls/wined3d/device.c b/dlls/wined3d/device.c index c5de58c29c9..7fb8a7ec3a6 100644 --- a/dlls/wined3d/device.c +++ b/dlls/wined3d/device.c @@ -657,9 +657,9 @@ bool wined3d_device_vk_create_null_resources(struct wined3d_device_vk *device_vk memory_type = VK_MEMORY_PROPERTY_DEVICE_LOCAL_BIT; if (!wined3d_context_vk_create_bo(context_vk, 16, usage, memory_type, &r->bo)) return false; - VK_CALL(vkCmdFillBuffer(vk_command_buffer, r->bo.vk_buffer, r->bo.buffer_offset, r->bo.size, 0x00000000u)); + VK_CALL(vkCmdFillBuffer(vk_command_buffer, r->bo.vk_buffer, r->bo.b.buffer_offset, r->bo.size, 0x00000000u)); r->buffer_info.buffer = r->bo.vk_buffer; - r->buffer_info.offset = r->bo.buffer_offset; + r->buffer_info.offset = r->bo.b.buffer_offset; r->buffer_info.range = r->bo.size; if (!wined3d_null_image_vk_init(&r->image_1d, context_vk, vk_command_buffer, VK_IMAGE_TYPE_1D, 1, 1)) @@ -735,7 +735,7 @@ bool wined3d_device_vk_create_null_views(struct wined3d_device_vk *device_vk, st buffer_create_info.flags = 0; buffer_create_info.buffer = r->bo.vk_buffer; buffer_create_info.format = VK_FORMAT_R32_UINT; - buffer_create_info.offset = r->bo.buffer_offset; + buffer_create_info.offset = r->bo.b.buffer_offset; buffer_create_info.range = r->bo.size; if ((vr = VK_CALL(vkCreateBufferView(device_vk->vk_device, diff --git a/dlls/wined3d/state.c b/dlls/wined3d/state.c index e0dba8701bc..c76e0c2b604 100644 --- a/dlls/wined3d/state.c +++ b/dlls/wined3d/state.c @@ -4600,7 +4600,7 @@ static void state_cb(struct wined3d_context *context, const struct wined3d_state buffer = buffer_state->buffer; bo_gl = wined3d_bo_gl(buffer->buffer_object); GL_EXTCALL(glBindBufferRange(GL_UNIFORM_BUFFER, base + i, - bo_gl->id, bo_gl->buffer_offset + buffer_state->offset, buffer_state->size)); + bo_gl->id, bo_gl->b.buffer_offset + buffer_state->offset, buffer_state->size)); buffer->bo_user.valid = true; } checkGLcall("bind constant buffers"); @@ -4677,7 +4677,7 @@ static void state_so(struct wined3d_context *context, const struct wined3d_state } size = buffer->resource.size - offset; GL_EXTCALL(glBindBufferRange(GL_TRANSFORM_FEEDBACK_BUFFER, i, - bo_gl->id, bo_gl->buffer_offset + offset, size)); + bo_gl->id, bo_gl->b.buffer_offset + offset, size)); buffer->bo_user.valid = true; } checkGLcall("bind transform feedback buffers"); diff --git a/dlls/wined3d/texture.c b/dlls/wined3d/texture.c index 7e2da350f5a..87ae569b549 100644 --- a/dlls/wined3d/texture.c +++ b/dlls/wined3d/texture.c @@ -4764,7 +4764,7 @@ static void wined3d_texture_vk_upload_data(struct wined3d_context *context, vk_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.buffer = src_bo->vk_buffer; - vk_barrier.offset = src_bo->buffer_offset + (size_t)src_bo_addr->addr; + vk_barrier.offset = src_bo->b.buffer_offset + (size_t)src_bo_addr->addr; vk_barrier.size = sub_resource->size; src_offset += (size_t)src_bo_addr->addr; @@ -4788,7 +4788,7 @@ static void wined3d_texture_vk_upload_data(struct wined3d_context *context, dst_texture_vk->layout, VK_IMAGE_LAYOUT_TRANSFER_DST_OPTIMAL, dst_texture_vk->image.vk_image, &vk_range); - region.bufferOffset = src_bo->buffer_offset + src_offset; + region.bufferOffset = src_bo->b.buffer_offset + src_offset; region.bufferRowLength = (src_row_pitch / src_format->block_byte_count) * src_format->block_width; if (src_row_pitch) region.bufferImageHeight = (src_slice_pitch / src_row_pitch) * src_format->block_height; @@ -4941,7 +4941,7 @@ static void wined3d_texture_vk_download_data(struct wined3d_context *context, vk_barrier.srcQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.dstQueueFamilyIndex = VK_QUEUE_FAMILY_IGNORED; vk_barrier.buffer = dst_bo->vk_buffer; - vk_barrier.offset = dst_bo->buffer_offset + (size_t)dst_bo_addr->addr; + vk_barrier.offset = dst_bo->b.buffer_offset + (size_t)dst_bo_addr->addr; vk_barrier.size = sub_resource->size; bo_stage_flags = vk_pipeline_stage_mask_from_buffer_usage(dst_bo->usage); @@ -4964,7 +4964,7 @@ static void wined3d_texture_vk_download_data(struct wined3d_context *context, src_texture_vk->layout, VK_IMAGE_LAYOUT_TRANSFER_SRC_OPTIMAL, src_texture_vk->image.vk_image, &vk_range); - region.bufferOffset = dst_bo->buffer_offset + dst_offset; + region.bufferOffset = dst_bo->b.buffer_offset + dst_offset; region.bufferRowLength = 0; region.bufferImageHeight = 0; region.imageSubresource.aspectMask = vk_range.aspectMask; diff --git a/dlls/wined3d/view.c b/dlls/wined3d/view.c index b902812aded..b981602f1a2 100644 --- a/dlls/wined3d/view.c +++ b/dlls/wined3d/view.c @@ -685,7 +685,7 @@ static VkBufferView wined3d_view_vk_create_vk_buffer_view(struct wined3d_context create_info.flags = 0; create_info.buffer = bo->vk_buffer; create_info.format = view_format_vk->vk_format; - create_info.offset = bo->buffer_offset + offset; + create_info.offset = bo->b.buffer_offset + offset; create_info.range = size; device_vk = wined3d_device_vk(buffer_vk->b.resource.device); @@ -2101,7 +2101,7 @@ void wined3d_unordered_access_view_vk_clear(struct wined3d_unordered_access_view buffer_info.buffer = constants_bo.vk_buffer; buffer_info.range = constants_bo.size; - buffer_info.offset = constants_bo.buffer_offset; + buffer_info.offset = constants_bo.b.buffer_offset; vk_info = context_vk->vk_info; @@ -2234,7 +2234,7 @@ static void wined3d_unordered_access_view_vk_cs_init(void *object) wined3d_context_vk_end_current_render_pass(context_vk); VK_CALL(vkCmdFillBuffer(wined3d_context_vk_get_command_buffer(context_vk), - uav_vk->counter_bo.vk_buffer, uav_vk->counter_bo.buffer_offset, sizeof(uint32_t), 0)); + uav_vk->counter_bo.vk_buffer, uav_vk->counter_bo.b.buffer_offset, sizeof(uint32_t), 0)); wined3d_context_vk_reference_bo(context_vk, &uav_vk->counter_bo); create_info.sType = VK_STRUCTURE_TYPE_BUFFER_VIEW_CREATE_INFO; @@ -2242,7 +2242,7 @@ static void wined3d_unordered_access_view_vk_cs_init(void *object) create_info.flags = 0; create_info.buffer = uav_vk->counter_bo.vk_buffer; create_info.format = VK_FORMAT_R32_UINT; - create_info.offset = uav_vk->counter_bo.buffer_offset; + create_info.offset = uav_vk->counter_bo.b.buffer_offset; create_info.range = sizeof(uint32_t); if ((vr = VK_CALL(vkCreateBufferView(device_vk->vk_device, &create_info, NULL, &uav_vk->vk_counter_view))) < 0) diff --git a/dlls/wined3d/wined3d_private.h b/dlls/wined3d/wined3d_private.h index 79b02c2448d..2fce585c6b1 100644 --- a/dlls/wined3d/wined3d_private.h +++ b/dlls/wined3d/wined3d_private.h @@ -1597,6 +1597,7 @@ struct wined3d_bo { struct list users; void *map_ptr; + size_t buffer_offset; size_t memory_offset; bool coherent; }; @@ -1606,7 +1607,6 @@ struct wined3d_bo_gl struct wined3d_bo b; GLuint id; - GLsizeiptr buffer_offset; GLsizeiptr size; GLenum binding; GLenum usage; @@ -1641,7 +1641,6 @@ struct wined3d_bo_vk VkDeviceMemory vk_memory; - VkDeviceSize buffer_offset; VkDeviceSize size; VkBufferUsageFlags usage; VkMemoryPropertyFlags memory_type; -- 2.33.0

3 12

[PATCH vkd3d] vkd3d-shader/hlsl: Add Gather methods and tests.
by Francisco Casas Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Francisco Casas <fcasas(a)codeweavers.com> --- Makefile.am | 2 + libs/vkd3d-shader/hlsl.c | 4 ++ libs/vkd3d-shader/hlsl.h | 4 ++ libs/vkd3d-shader/hlsl.y | 78 +++++++++++++++++++++++++ libs/vkd3d-shader/hlsl_sm4.c | 66 ++++++++++++++++++++- tests/hlsl-gather.shader_test | 106 ++++++++++++++++++++++++++++++++++ 6 files changed, 259 insertions(+), 1 deletion(-) create mode 100644 tests/hlsl-gather.shader_test diff --git a/Makefile.am b/Makefile.am index 3e083b0a..1a80152a 100644 --- a/Makefile.am +++ b/Makefile.am @@ -67,6 +67,7 @@ vkd3d_shader_tests = \ tests/hlsl-duplicate-modifiers.shader_test \ tests/hlsl-for.shader_test \ tests/hlsl-function-overload.shader_test \ + tests/hlsl-gather.shader_test \ tests/hlsl-invalid.shader_test \ tests/hlsl-majority-pragma.shader_test \ tests/hlsl-majority-typedef.shader_test \ @@ -292,6 +293,7 @@ XFAIL_TESTS = \ tests/hlsl-duplicate-modifiers.shader_test \ tests/hlsl-for.shader_test \ tests/hlsl-function-overload.shader_test \ + tests/hlsl-gather.shader_test \ tests/hlsl-majority-pragma.shader_test \ tests/hlsl-majority-typedef.shader_test \ tests/hlsl-nested-arrays.shader_test \ diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index 1eee4278..74b78f1b 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -1250,6 +1250,10 @@ static void dump_ir_resource_load(struct vkd3d_string_buffer *buffer, const stru { [HLSL_RESOURCE_LOAD] = "load_resource", [HLSL_RESOURCE_SAMPLE] = "sample", + [HLSL_RESOURCE_GATHER_RED] = "gather4r", + [HLSL_RESOURCE_GATHER_GREEN] = "gather4g", + [HLSL_RESOURCE_GATHER_BLUE] = "gather4b", + [HLSL_RESOURCE_GATHER_ALPHA] = "gather4a", }; vkd3d_string_buffer_printf(buffer, "%s(resource = ", type_names[load->load_type]); diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 365ef980..891f671a 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -377,6 +377,10 @@ enum hlsl_resource_load_type { HLSL_RESOURCE_LOAD, HLSL_RESOURCE_SAMPLE, + HLSL_RESOURCE_GATHER_RED, + HLSL_RESOURCE_GATHER_GREEN, + HLSL_RESOURCE_GATHER_BLUE, + HLSL_RESOURCE_GATHER_ALPHA, }; struct hlsl_ir_resource_load diff --git a/libs/vkd3d-shader/hlsl.y b/libs/vkd3d-shader/hlsl.y index b7e0409d..8a2c1dd9 100644 --- a/libs/vkd3d-shader/hlsl.y +++ b/libs/vkd3d-shader/hlsl.y @@ -1852,6 +1852,84 @@ static bool add_method_call(struct hlsl_ctx *ctx, struct list *instrs, struct hl list_add_tail(instrs, &load->node.entry); return true; } + else if (!strcmp(name, "Gather") || !strcmp(name, "GatherRed") || !strcmp(name, "GatherBlue") + || !strcmp(name, "GatherGreen") || !strcmp(name, "GatherAlpha")) + { + const unsigned int sampler_dim = sampler_dim_count(object_type->sampler_dim); + enum hlsl_resource_load_type load_type; + const struct hlsl_type *sampler_type; + struct hlsl_ir_resource_load *load; + struct hlsl_ir_load *sampler_load; + struct hlsl_type *result_type; + struct hlsl_ir_node *coords; + int read_channel; + + if(!strcmp(name, "Gather") || !strcmp(name, "GatherRed")){ + load_type = HLSL_RESOURCE_GATHER_RED; + read_channel = 0; + } + else if(!strcmp(name, "GatherGreen")){ + load_type = HLSL_RESOURCE_GATHER_GREEN; + read_channel = 1; + } + else if(!strcmp(name, "GatherBlue")){ + load_type = HLSL_RESOURCE_GATHER_BLUE; + read_channel = 2; + } + else if(!strcmp(name, "GatherAlpha")){ + load_type = HLSL_RESOURCE_GATHER_ALPHA; + read_channel = 3; + } else { + assert(!"Unexpected Gather method."); + } + + if (params->args_count != 3 && params->args_count != 4) + { + hlsl_error(ctx, *loc, VKD3D_SHADER_ERROR_HLSL_WRONG_PARAMETER_COUNT, + "Wrong number of arguments to method '%s': expected 3 or 4, but got %u.", name, params->args_count); + return false; + } + if (params->args_count >= 3) + FIXME("Ignoring offset parameter.\n"); + if (params->args_count >= 4) + FIXME("Ignoring 'status' output parameter.\n"); + + sampler_type = params->args[0]->data_type; + if (sampler_type->type != HLSL_CLASS_OBJECT || sampler_type->base_type != HLSL_TYPE_SAMPLER + || sampler_type->sampler_dim != HLSL_SAMPLER_DIM_GENERIC) + { + struct vkd3d_string_buffer *string; + + if ((string = hlsl_type_to_string(ctx, sampler_type))) + hlsl_error(ctx, *loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE, + "Wrong type for argument 0 of %s(): expected 'sampler', but got '%s'.", name, string->buffer); + hlsl_release_string_buffer(ctx, string); + return false; + } + + if(read_channel >= object_type->e.resource_format->dimx){ + hlsl_error(ctx, *loc, VKD3D_SHADER_ERROR_HLSL_INVALID_TYPE, + "Method %s() requires at least %d channels.", name, read_channel+1); + return false; + } + + result_type = hlsl_get_vector_type(ctx, object_type->e.resource_format->base_type, 4); + + /* Only HLSL_IR_LOAD can return an object. */ + sampler_load = hlsl_ir_load(params->args[0]); + + if (!(coords = add_implicit_conversion(ctx, instrs, params->args[1], + hlsl_get_vector_type(ctx, HLSL_TYPE_FLOAT, sampler_dim), loc))) + coords = params->args[1]; + + if (!(load = hlsl_new_resource_load(ctx, result_type, + load_type, object_load->src.var, object_load->src.offset.node, + sampler_load->src.var, sampler_load->src.offset.node, coords, loc))) + return false; + list_add_tail(instrs, &load->node.entry); + return true; + + } else { struct vkd3d_string_buffer *string; diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c index a95503ee..5f290129 100644 --- a/libs/vkd3d-shader/hlsl_sm4.c +++ b/libs/vkd3d-shader/hlsl_sm4.c @@ -753,7 +753,7 @@ struct sm4_instruction struct sm4_register reg; enum vkd3d_sm4_swizzle_type swizzle_type; unsigned int swizzle; - } srcs[2]; + } srcs[3]; unsigned int src_count; uint32_t idx[2]; @@ -1506,6 +1506,61 @@ static void write_sm4_loop(struct hlsl_ctx *ctx, write_sm4_instruction(buffer, &instr); } + +static void write_sm4_gather(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer, + const struct hlsl_type *resource_type, const struct hlsl_ir_node *dst, + const struct hlsl_deref *resource, const struct hlsl_deref *sampler, const struct hlsl_ir_node *coords, + enum hlsl_resource_load_type load_type) +{ + struct sm4_instruction instr; + unsigned int writemask; + + memset(&instr, 0, sizeof(instr)); + instr.opcode = VKD3D_SM4_OP_GATHER4; + + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, NULL, dst); + instr.dst_count = 1; + + sm4_register_from_node(&instr.srcs[0].reg, &writemask, &instr.srcs[0].swizzle_type, coords); + instr.srcs[0].swizzle = hlsl_swizzle_from_writemask(writemask); + + sm4_register_from_deref(ctx, &instr.srcs[1].reg, &writemask, &instr.srcs[1].swizzle_type, + resource, resource_type); + instr.srcs[1].swizzle = hlsl_map_swizzle(hlsl_swizzle_from_writemask(writemask), instr.dsts[0].writemask); + + sm4_register_from_deref(ctx, &instr.srcs[2].reg, &writemask, NULL, + sampler, sampler->var->data_type); + instr.srcs[2].reg.type = VKD3D_SM4_RT_SAMPLER; + instr.srcs[2].reg.dim = 1; + instr.srcs[2].swizzle_type = VKD3D_SM4_SWIZZLE_VEC4; + + switch (load_type){ + case HLSL_RESOURCE_GATHER_RED: + instr.srcs[2].swizzle = HLSL_SWIZZLE(X, X, X, X); + break; + case HLSL_RESOURCE_GATHER_GREEN: + instr.srcs[2].swizzle = HLSL_SWIZZLE(Y, Y, Y, Y); + break; + case HLSL_RESOURCE_GATHER_BLUE: + instr.srcs[2].swizzle = HLSL_SWIZZLE(Z, Z, Z, Z); + break; + case HLSL_RESOURCE_GATHER_ALPHA: + instr.srcs[2].swizzle = HLSL_SWIZZLE(W, W, W, W); + break; + default: + assert(!"Invalid hlsl_resource_load_type."); + break; + } + instr.src_count = 3; + + write_sm4_instruction(buffer, &instr); +} + + + + + + static void write_sm4_resource_load(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_resource_load *load) { @@ -1527,6 +1582,15 @@ static void write_sm4_resource_load(struct hlsl_ctx *ctx, case HLSL_RESOURCE_SAMPLE: hlsl_fixme(ctx, load->node.loc, "Resource sample instruction."); break; + + case HLSL_RESOURCE_GATHER_RED: + case HLSL_RESOURCE_GATHER_GREEN: + case HLSL_RESOURCE_GATHER_BLUE: + case HLSL_RESOURCE_GATHER_ALPHA: + assert(load->sampler.var); + write_sm4_gather(ctx, buffer, resource_type, &load->node, &load->resource, + &load->sampler, coords, load->load_type); + break; } } diff --git a/tests/hlsl-gather.shader_test b/tests/hlsl-gather.shader_test new file mode 100644 index 00000000..6731b77c --- /dev/null +++ b/tests/hlsl-gather.shader_test @@ -0,0 +1,106 @@ +[sampler 0] +filter linear linear linear +address clamp clamp clamp + +[texture 0] +size (3, 3) +0.0 0.0 0.0 0.4 0.1 0.0 0.5 0.0 0.2 0.0 0.0 0.4 +0.0 0.1 0.5 0.0 0.1 0.1 0.0 0.4 0.2 0.1 0.5 0.0 +0.0 0.2 0.0 0.4 0.1 0.2 0.5 0.0 0.2 0.2 0.0 0.4 + + + +[pixel shader] +sampler s; +Texture2D t; + +float4 main() : sv_target +{ + return t.Gather(s, float2(0.2, 0.2), int2(0 , 0)); +} + +[test] +draw quad +probe all rgba (0.0, 0.1, 0.1, 0.0) + + + +[pixel shader] +SamplerState s; +Texture2D t; + +float4 main() : sv_target +{ + return t.GatherGreen(s, float2(0.2, 0.2), int2(0 , 0)); +} + +[test] +draw quad +probe all rgba (0.1, 0.1, 0.0, 0.0) + + + +[pixel shader] +SamplerState s; +Texture2D t; + +float4 main() : sv_target +{ + return t.GatherGreen(s, float2(0.8, 0.8), int2(0 , 0)); +} + +[test] +draw quad +probe all rgba (0.2, 0.2, 0.1, 0.1) + + + + +[pixel shader] +SamplerState s; +Texture2D t; + +float4 main() : sv_target +{ + return t.GatherGreen(s, float2(0.2, 0.2), int2(0 , 0)); +} + +[test] +draw quad +probe all rgba (0.1, 0.1, 0.0, 0.0) + + + + +[pixel shader] +SamplerState s; +Texture2D t; + +float4 main() : sv_target +{ + return t.GatherBlue(s, float2(0.2, 0.8), int2(0 , 0)); +} + +[test] +draw quad +probe all rgba (0.0, 0.5, 0.0, 0.5) + + + + +[pixel shader] +SamplerState s; +Texture2D t; + +float4 main() : sv_target +{ + return t.GatherAlpha(s, float2(0.2, 0.8), int2(0 , 0)); +} + +[test] +draw quad +probe all rgba (0.4, 0.0, 0.4, 0.0) + + + + -- 2.25.1

2 1

[PATCH vkd3d v2 1/7] vkd3d-shader/hlsl: Write SM4 integer negation.
by Giovanni Mascellani Nov. 22, 2021

Nov. 22, 2021

--- libs/vkd3d-shader/hlsl_sm4.c | 4 ++++ 1 file changed, 4 insertions(+) diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c index a95503ee..acc11ce9 100644 --- a/libs/vkd3d-shader/hlsl_sm4.c +++ b/libs/vkd3d-shader/hlsl_sm4.c @@ -1376,6 +1376,10 @@ static void write_sm4_expr(struct hlsl_ctx *ctx, break; } + case HLSL_OP1_NEG: + write_sm4_unary_op(buffer, VKD3D_SM4_OP_INEG, &expr->node, arg1, 0); + break; + default: hlsl_fixme(ctx, expr->node.loc, "SM4 int \"%s\" expression.", debug_hlsl_expr_op(expr->op)); break; -- 2.34.0

2 15

[RFC, website] Remove stray file
by Julian Rüger Nov. 22, 2021

Nov. 22, 2021

Good evening to you all! I just noticed there is an empty file "freenode.uP4Y16UYccnE.html" in the top-level of wine-website git. https://source.winehq.org/git/website.git/tree Since freenode is dead and all, can this file be removed? Does anyone know what it's purpose was supposed to be? Best, Julian

2 1

[PATCH] ntdll: Place PEB before first TEB.
by Paul Gofman Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Paul Gofman <pgofman(a)codeweavers.com> --- dlls/ntdll/tests/wow64.c | 25 +++++++++++++++++++++++++ dlls/ntdll/thread.c | 4 +--- dlls/ntdll/unix/debug.c | 2 +- dlls/ntdll/unix/virtual.c | 15 ++++++++------- 4 files changed, 35 insertions(+), 11 deletions(-) diff --git a/dlls/ntdll/tests/wow64.c b/dlls/ntdll/tests/wow64.c index be2f2138933..3d2548843be 100644 --- a/dlls/ntdll/tests/wow64.c +++ b/dlls/ntdll/tests/wow64.c @@ -225,6 +225,14 @@ static void test_query_architectures(void) } } +static TEB *test_thread_teb_addr; + +static DWORD WINAPI test_teb_addr_thread(void *ptr) +{ + test_thread_teb_addr = NtCurrentTeb(); + return 0; +} + static void test_peb_teb(void) { PROCESS_BASIC_INFORMATION proc_info; @@ -232,6 +240,7 @@ static void test_peb_teb(void) PROCESS_INFORMATION pi; STARTUPINFOA si = {0}; NTSTATUS status; + HANDLE thread; void *redir; SIZE_T res; TEB teb; @@ -419,6 +428,22 @@ static void test_peb_teb(void) ok( !NtCurrentTeb()->GdiBatchCount, "GdiBatchCount set to %x\n", NtCurrentTeb()->GdiBatchCount ); ok( !NtCurrentTeb()->WowTebOffset || broken( NtCurrentTeb()->WowTebOffset == 1 ), /* vista */ "WowTebOffset set to %x\n", NtCurrentTeb()->WowTebOffset ); + + thread = CreateThread( NULL, 0, test_teb_addr_thread, NULL, 0, NULL ); + WaitForSingleObject( thread, INFINITE ); + if (test_thread_teb_addr > NtCurrentTeb()) + { + if (sizeof(void *) > sizeof(int) || is_wow64) + { + ok( (char *)NtCurrentTeb() - NtCurrentTeb()->WowTebOffset == (char *)NtCurrentTeb()->Peb + 0x1000, + "Got unexpected teb %p, peb %p.\n", NtCurrentTeb(), NtCurrentTeb()->Peb ); + } + } + else + { + /* Before Win10 1607. */ + win_skip( "Old teb placement.\n" ); + } } static void test_selectors(void) diff --git a/dlls/ntdll/thread.c b/dlls/ntdll/thread.c index 37dc7c8ab37..dcf226294e8 100644 --- a/dlls/ntdll/thread.c +++ b/dlls/ntdll/thread.c @@ -63,9 +63,7 @@ static inline struct debug_info *get_info(void) static void init_options(void) { - unsigned int offset = page_size * (sizeof(void *) / 4); - - debug_options = (struct __wine_debug_channel *)((char *)NtCurrentTeb()->Peb + offset); + debug_options = (struct __wine_debug_channel *)((char *)NtCurrentTeb()->Peb - 3 * page_size); while (debug_options[nb_debug_options].name[0]) nb_debug_options++; } diff --git a/dlls/ntdll/unix/debug.c b/dlls/ntdll/unix/debug.c index 26e7decc059..7744d421f6d 100644 --- a/dlls/ntdll/unix/debug.c +++ b/dlls/ntdll/unix/debug.c @@ -324,7 +324,7 @@ void dbg_init(void) if (nb_debug_options == -1) init_options(); - options = (struct __wine_debug_channel *)((char *)peb + (is_win64 ? 2 : 1) * page_size); + options = (struct __wine_debug_channel *)((char *)peb - 3 * page_size); memcpy( options, debug_options, nb_debug_options * sizeof(*options) ); free( debug_options ); debug_options = options; diff --git a/dlls/ntdll/unix/virtual.c b/dlls/ntdll/unix/virtual.c index 5873a3e2335..1dcde261bcf 100644 --- a/dlls/ntdll/unix/virtual.c +++ b/dlls/ntdll/unix/virtual.c @@ -2986,6 +2986,7 @@ static TEB *init_teb( void *ptr, BOOL is_wow ) */ TEB *virtual_alloc_first_teb(void) { + static const BOOL is_wow = FALSE; void *ptr; NTSTATUS status; SIZE_T data_size = page_size; @@ -3003,12 +3004,12 @@ TEB *virtual_alloc_first_teb(void) NtAllocateVirtualMemory( NtCurrentProcess(), &teb_block, is_win64 ? 0x7fffffff : 0, &total, MEM_RESERVE | MEM_TOP_DOWN, PAGE_READWRITE ); - teb_block_pos = 30; - ptr = (char *)teb_block + 30 * block_size; + teb_block_pos = 2; + ptr = (char *)teb_block; data_size = 2 * block_size; NtAllocateVirtualMemory( NtCurrentProcess(), (void **)&ptr, 0, &data_size, MEM_COMMIT, PAGE_READWRITE ); - peb = (PEB *)((char *)teb_block + 31 * block_size + (is_win64 ? 0 : page_size)); - return init_teb( ptr, FALSE ); + peb = (PEB *)((char *)teb_block + block_size - page_size - (is_win64 && is_wow ? page_size : 0)); + return init_teb( (char *)ptr + block_size, is_wow ); } @@ -3033,7 +3034,7 @@ NTSTATUS virtual_alloc_teb( TEB **ret_teb ) } else { - if (!teb_block_pos) + if (teb_block_pos == 32) { SIZE_T total = 32 * block_size; @@ -3044,9 +3045,9 @@ NTSTATUS virtual_alloc_teb( TEB **ret_teb ) return status; } teb_block = ptr; - teb_block_pos = 32; + teb_block_pos = 0; } - ptr = ((char *)teb_block + --teb_block_pos * block_size); + ptr = ((char *)teb_block + teb_block_pos++ * block_size); NtAllocateVirtualMemory( NtCurrentProcess(), (void **)&ptr, 0, &block_size, MEM_COMMIT, PAGE_READWRITE ); } -- 2.33.1

1 0

[website] German translation for release 6.22
by Julian Rüger Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Julian Rüger <jr98(a)gmx.net>

1 0

[website] German translation for WWN issue 407
by Julian Rüger Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Julian Rüger <jr98(a)gmx.net>

1 0

[website] German translation for release 6.21
by Julian Rüger Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Julian Rüger <jr98(a)gmx.net>

1 0

[website] German translation for WWN issue 406
by Julian Rüger Nov. 22, 2021

Nov. 22, 2021

Signed-off-by: Julian Rüger <jr98(a)gmx.net>

1 0