November 2021 - Wine-Devel - list.winehq.org

[PATCH vkd3d 5/5] vkd3d-shader/hlsl: Write SM4 sample instructions.

by Zebediah Figura

Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- libs/vkd3d-shader/hlsl_sm4.c | 56 ++++++++++++++++++++++++++++++++++-- 1 file changed, 54 insertions(+), 2 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c index e597425ae..da31d5ae2 100644 --- a/libs/vkd3d-shader/hlsl_sm4.c +++ b/libs/vkd3d-shader/hlsl_sm4.c @@ -749,7 +749,7 @@ struct sm4_instruction { struct sm4_register reg; unsigned int swizzle; - } srcs[2]; + } srcs[3]; unsigned int src_count; uint32_t idx[2]; @@ -761,6 +761,7 @@ static unsigned int sm4_swizzle_type(enum vkd3d_sm4_register_type type) switch (type) { case VKD3D_SM4_RT_IMMCONST: + case VKD3D_SM4_RT_SAMPLER: return VKD3D_SM4_SWIZZLE_NONE; case VKD3D_SM4_RT_CONSTBUFFER: @@ -790,6 +791,14 @@ static void sm4_register_from_deref(struct hlsl_ctx *ctx, struct sm4_register *r reg->idx_count = 1; *writemask = VKD3DSP_WRITEMASK_ALL; } + else if (data_type->type == HLSL_CLASS_OBJECT && data_type->base_type == HLSL_TYPE_SAMPLER) + { + reg->type = VKD3D_SM4_RT_SAMPLER; + reg->dim = VKD3D_SM4_DIMENSION_NONE; + reg->idx[0] = var->reg.id; + reg->idx_count = 1; + *writemask = VKD3DSP_WRITEMASK_ALL; + } else { unsigned int offset = hlsl_offset_from_deref(deref) + var->buffer_offset; @@ -1227,6 +1236,32 @@ static void write_sm4_ld(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buf write_sm4_instruction(buffer, &instr); } +static void write_sm4_sample(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) +{ + struct sm4_instruction instr; + unsigned int writemask; + + memset(&instr, 0, sizeof(instr)); + instr.opcode = VKD3D_SM4_OP_SAMPLE; + + sm4_register_from_node(&instr.dsts[0].reg, &instr.dsts[0].writemask, dst); + instr.dst_count = 1; + + sm4_register_from_node(&instr.srcs[0].reg, &writemask, coords); + instr.srcs[0].swizzle = hlsl_swizzle_from_writemask(writemask); + + sm4_register_from_deref(ctx, &instr.srcs[1].reg, &writemask, 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, sampler, sampler->var->data_type); + + instr.src_count = 3; + + write_sm4_instruction(buffer, &instr); +} + static void write_sm4_expr(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_expr *expr) { @@ -1497,6 +1532,21 @@ static void write_sm4_resource_load(struct hlsl_ctx *ctx, const struct hlsl_type *resource_type = load->resource.var->data_type; const struct hlsl_ir_node *coords = load->coords.node; + if (load->sampler.var) + { + const struct hlsl_type *sampler_type = load->sampler.var->data_type; + + assert(sampler_type->type == HLSL_CLASS_OBJECT); + assert(sampler_type->base_type == HLSL_TYPE_SAMPLER); + assert(sampler_type->sampler_dim == HLSL_SAMPLER_DIM_GENERIC); + + if (!load->sampler.var->is_uniform) + { + hlsl_fixme(ctx, load->node.loc, "Sample using non-uniform sampler variable."); + return; + } + } + if (!load->resource.var->is_uniform) { hlsl_fixme(ctx, load->node.loc, "Load from non-uniform resource variable."); @@ -1510,7 +1560,9 @@ static void write_sm4_resource_load(struct hlsl_ctx *ctx, break; case HLSL_RESOURCE_SAMPLE: - hlsl_fixme(ctx, load->node.loc, "Resource sample instruction."); + if (!load->sampler.var) + hlsl_fixme(ctx, load->node.loc, "SM4 combined sample expression.\n"); + write_sm4_sample(ctx, buffer, resource_type, &load->node, &load->resource, &load->sampler, coords); break; } } -- 2.33.1

Nov. 9, 2021

[PATCH vkd3d 4/5] vkd3d-shader/hlsl: Write SM4 sampler declarations.

by Zebediah Figura

Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- libs/vkd3d-shader/hlsl_sm4.c | 18 +++++++++++++++++- 1 file changed, 17 insertions(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/hlsl_sm4.c b/libs/vkd3d-shader/hlsl_sm4.c index d48a41d11..e597425ae 100644 --- a/libs/vkd3d-shader/hlsl_sm4.c +++ b/libs/vkd3d-shader/hlsl_sm4.c @@ -974,6 +974,20 @@ static void write_sm4_dcl_constant_buffer(struct vkd3d_bytecode_buffer *buffer, write_sm4_instruction(buffer, &instr); } +static void write_sm4_dcl_sampler(struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_var *var) +{ + const struct sm4_instruction instr = + { + .opcode = VKD3D_SM4_OP_DCL_SAMPLER, + + .dsts[0].reg.type = VKD3D_SM4_RT_SAMPLER, + .dsts[0].reg.idx = {var->reg.id}, + .dsts[0].reg.idx_count = 1, + .dst_count = 1, + }; + write_sm4_instruction(buffer, &instr); +} + static void write_sm4_dcl_texture(struct vkd3d_bytecode_buffer *buffer, const struct hlsl_ir_var *var) { const struct sm4_instruction instr = @@ -1647,7 +1661,9 @@ static void write_sm4_shdr(struct hlsl_ctx *ctx, if (!var->reg.allocated || var->data_type->type != HLSL_CLASS_OBJECT) continue; - if (var->data_type->base_type == HLSL_TYPE_TEXTURE) + if (var->data_type->base_type == HLSL_TYPE_SAMPLER) + write_sm4_dcl_sampler(&buffer, var); + else if (var->data_type->base_type == HLSL_TYPE_TEXTURE) write_sm4_dcl_texture(&buffer, var); } -- 2.33.1

Nov. 9, 2021

[PATCH vkd3d 3/5] vkd3d-shader/hlsl: Handle samplers in write_sm1_uniform().

by Zebediah Figura

Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- libs/vkd3d-shader/hlsl_sm1.c | 14 ++++++++++++-- 1 file changed, 12 insertions(+), 2 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_sm1.c b/libs/vkd3d-shader/hlsl_sm1.c index c3d27eea1..875f521f7 100644 --- a/libs/vkd3d-shader/hlsl_sm1.c +++ b/libs/vkd3d-shader/hlsl_sm1.c @@ -352,8 +352,18 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffe if (!var->semantic.name && var->reg.allocated) { put_u32(buffer, 0); /* name */ - put_u32(buffer, vkd3d_make_u32(D3DXRS_FLOAT4, var->reg.id)); - put_u32(buffer, var->data_type->reg_size / 4); + if (var->data_type->type == HLSL_CLASS_OBJECT + && (var->data_type->base_type == HLSL_TYPE_SAMPLER + || var->data_type->base_type == HLSL_TYPE_TEXTURE)) + { + put_u32(buffer, vkd3d_make_u32(D3DXRS_SAMPLER, var->reg.id)); + put_u32(buffer, 1); + } + else + { + put_u32(buffer, vkd3d_make_u32(D3DXRS_FLOAT4, var->reg.id)); + put_u32(buffer, var->data_type->reg_size / 4); + } put_u32(buffer, 0); /* type */ put_u32(buffer, 0); /* FIXME: default value */ } -- 2.33.1

Nov. 9, 2021

[PATCH vkd3d 2/5] vkd3d-shader/hlsl: Allocate samplers.

by Zebediah Figura

Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- Yes, this is very similar to the allocation path for textures, and it will be similar to the allocation path for UAVs. I was unable to deduplicate it in a way I was happy with... libs/vkd3d-shader/hlsl_codegen.c | 59 ++++++++++++++++++++++++++++++++ 1 file changed, 59 insertions(+) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 24b8205c1..e358bd7c9 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -1202,6 +1202,64 @@ static void allocate_buffers(struct hlsl_ctx *ctx) } } +static const struct hlsl_ir_var *get_reserved_sampler(struct hlsl_ctx *ctx, uint32_t index) +{ + const struct hlsl_ir_var *var; + + LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, const struct hlsl_ir_var, extern_entry) + { + if (var->last_read && var->reg_reservation.type == 's' && var->reg_reservation.index == index) + return var; + } + return NULL; +} + +static void allocate_samplers(struct hlsl_ctx *ctx) +{ + struct hlsl_ir_var *var; + uint32_t index = 0; + + LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) + { + if (!var->last_read || var->data_type->type != HLSL_CLASS_OBJECT + || var->data_type->base_type != HLSL_TYPE_SAMPLER) + continue; + + if (var->reg_reservation.type == 's') + { + const struct hlsl_ir_var *reserved_sampler = get_reserved_sampler(ctx, var->reg_reservation.index); + + if (reserved_sampler && reserved_sampler != var) + { + hlsl_error(ctx, var->loc, VKD3D_SHADER_ERROR_HLSL_OVERLAPPING_RESERVATIONS, + "Multiple samplers bound to s%u.", var->reg_reservation.index); + hlsl_note(ctx, reserved_sampler->loc, VKD3D_SHADER_LOG_ERROR, + "Sampler '%s' is already bound to s%u.", reserved_sampler->name, + var->reg_reservation.index); + } + + var->reg.id = var->reg_reservation.index; + var->reg.allocated = true; + TRACE("Allocated reserved %s to s%u.\n", var->name, index); + } + else if (!var->reg_reservation.type) + { + while (get_reserved_sampler(ctx, index)) + ++index; + + var->reg.id = index; + var->reg.allocated = true; + TRACE("Allocated %s to s%u.\n", var->name, index); + ++index; + } + else + { + hlsl_error(ctx, var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, + "Samplers must be bound to register type 's'."); + } + } +} + static const struct hlsl_ir_var *get_reserved_texture(struct hlsl_ctx *ctx, uint32_t index) { const struct hlsl_ir_var *var; @@ -1379,6 +1437,7 @@ int hlsl_emit_dxbc(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_fun allocate_textures(ctx); } allocate_semantic_registers(ctx); + allocate_samplers(ctx); if (ctx->result) return ctx->result; -- 2.33.1

Nov. 9, 2021

[PATCH v4 vkd3d 1/5] vkd3d-shader/hlsl: Avoid hard-coding array sizes in hlsl_type_to_string().

by Zebediah Figura

From: Joshua Ashton <joshua(a)froggi.es> Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- libs/vkd3d-shader/hlsl.c | 18 +++++++++--------- 1 file changed, 9 insertions(+), 9 deletions(-) diff --git a/libs/vkd3d-shader/hlsl.c b/libs/vkd3d-shader/hlsl.c index c71eac3f1..1eee4278a 100644 --- a/libs/vkd3d-shader/hlsl.c +++ b/libs/vkd3d-shader/hlsl.c @@ -854,14 +854,14 @@ struct vkd3d_string_buffer *hlsl_type_to_string(struct hlsl_ctx *ctx, const stru { struct vkd3d_string_buffer *string; - static const char base_types[HLSL_TYPE_LAST_SCALAR + 1][7] = - { - "float", - "half", - "double", - "int", - "uint", - "bool", + static const char *const base_types[] = + { + [HLSL_TYPE_FLOAT] = "float", + [HLSL_TYPE_HALF] = "half", + [HLSL_TYPE_DOUBLE] = "double", + [HLSL_TYPE_INT] = "int", + [HLSL_TYPE_UINT] = "uint", + [HLSL_TYPE_BOOL] = "bool", }; if (!(string = hlsl_get_string_buffer(ctx))) @@ -915,7 +915,7 @@ struct vkd3d_string_buffer *hlsl_type_to_string(struct hlsl_ctx *ctx, const stru case HLSL_CLASS_OBJECT: { - static const char dimensions[5][HLSL_SAMPLER_DIM_MAX + 1] = + static const char *const dimensions[] = { [HLSL_SAMPLER_DIM_1D] = "1D", [HLSL_SAMPLER_DIM_2D] = "2D", -- 2.33.1

Nov. 9, 2021

[PATCH] kernelbase: Implement AppPolicyGetMediaFoundationCodecLoading.

by Mohamad Al-Jaf

Enables native Windows 10 media foundation libraries to run. Signed-off-by: Mohamad Al-Jaf <mohamadaljaf(a)gmail.com> --- .../api-ms-win-appmodel-runtime-l1-1-2.spec | 2 +- dlls/kernelbase/kernelbase.spec | 2 +- dlls/kernelbase/main.c | 14 ++++++++++++++ include/appmodel.h | 6 ++++++ 4 files changed, 22 insertions(+), 2 deletions(-) diff --git a/dlls/api-ms-win-appmodel-runtime-l1-1-2/api-ms-win-appmodel-runtime-l1-1-2.spec b/dlls/api-ms-win-appmodel-runtime-l1-1-2/api-ms-win-appmodel-runtime-l1-1-2.spec index 3c1b4058a9f..89fe702bbc2 100644 --- a/dlls/api-ms-win-appmodel-runtime-l1-1-2/api-ms-win-appmodel-runtime-l1-1-2.spec +++ b/dlls/api-ms-win-appmodel-runtime-l1-1-2/api-ms-win-appmodel-runtime-l1-1-2.spec @@ -1,7 +1,7 @@ @ stub AppPolicyGetClrCompat @ stub AppPolicyGetCreateFileAccess @ stub AppPolicyGetLifecycleManagement -@ stub AppPolicyGetMediaFoundationCodecLoading +@ stdcall AppPolicyGetMediaFoundationCodecLoading(ptr ptr) kernelbase.AppPolicyGetMediaFoundationCodecLoading @ stdcall AppPolicyGetProcessTerminationMethod(ptr ptr) kernelbase.AppPolicyGetProcessTerminationMethod @ stdcall AppPolicyGetShowDeveloperDiagnostic(ptr ptr) kernelbase.AppPolicyGetShowDeveloperDiagnostic @ stdcall AppPolicyGetThreadInitializationType(ptr ptr) kernelbase.AppPolicyGetThreadInitializationType diff --git a/dlls/kernelbase/kernelbase.spec b/dlls/kernelbase/kernelbase.spec index dd9f4a8ddce..55208889c35 100644 --- a/dlls/kernelbase/kernelbase.spec +++ b/dlls/kernelbase/kernelbase.spec @@ -43,7 +43,7 @@ # @ stub AppPolicyGetClrCompat # @ stub AppPolicyGetCreateFileAccess # @ stub AppPolicyGetLifecycleManagement -# @ stub AppPolicyGetMediaFoundationCodecLoading +@ stdcall AppPolicyGetMediaFoundationCodecLoading(ptr ptr) @ stdcall AppPolicyGetProcessTerminationMethod(ptr ptr) @ stdcall AppPolicyGetShowDeveloperDiagnostic(ptr ptr) @ stdcall AppPolicyGetThreadInitializationType(ptr ptr) diff --git a/dlls/kernelbase/main.c b/dlls/kernelbase/main.c index b90ee1cba2c..e6cf892a4dc 100644 --- a/dlls/kernelbase/main.c +++ b/dlls/kernelbase/main.c @@ -88,6 +88,20 @@ INT WINAPI MulDiv( INT a, INT b, INT c ) return ret; } +/*********************************************************************** + * AppPolicyGetMediaFoundationCodecLoading (KERNELBASE.@) + */ + +LONG WINAPI AppPolicyGetMediaFoundationCodecLoading(HANDLE token, AppPolicyMediaFoundationCodecLoading *policy) +{ + FIXME("%p, %p\n", token, policy); + + if(policy) + *policy = AppPolicyMediaFoundationCodecLoading_All; + + return ERROR_SUCCESS; +} + /*********************************************************************** * AppPolicyGetProcessTerminationMethod (KERNELBASE.@) */ diff --git a/include/appmodel.h b/include/appmodel.h index e4288bbfbb0..cf1b9659d14 100644 --- a/include/appmodel.h +++ b/include/appmodel.h @@ -22,6 +22,12 @@ extern "C" { #endif +typedef enum AppPolicyMediaFoundationCodecLoading +{ + AppPolicyMediaFoundationCodecLoading_All = 0, + AppPolicyMediaFoundationCodecLoading_InboxOnly = 1, +} AppPolicyMediaFoundationCodecLoading; + typedef enum AppPolicyProcessTerminationMethod { AppPolicyProcessTerminationMethod_ExitProcess = 0, -- 2.33.1

Nov. 9, 2021

[PATCH] wined3d: Parse ISG1, PSG1 and OSG1 signatures.

by Zebediah Figura

These signatures extend the OSG5 format by a minimum precision hint, which gets ignored for now but could be implemented using 16-bit floats in the future. Fixes some shader parsing errors in Resident Evil 2. Based on a vkd3d patch by Philip Rebohle. Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- dlls/wined3d/shader_sm4.c | 21 ++++++++++++++++++--- include/wine/wined3d.h | 1 + 2 files changed, 19 insertions(+), 3 deletions(-) diff --git a/dlls/wined3d/shader_sm4.c b/dlls/wined3d/shader_sm4.c index 4635910269f..005be96a51f 100644 --- a/dlls/wined3d/shader_sm4.c +++ b/dlls/wined3d/shader_sm4.c @@ -1820,10 +1820,13 @@ const struct wined3d_shader_frontend sm4_shader_frontend = #define TAG_AON9 WINEMAKEFOURCC('A', 'o', 'n', '9') #define TAG_DXBC WINEMAKEFOURCC('D', 'X', 'B', 'C') +#define TAG_ISG1 WINEMAKEFOURCC('I', 'S', 'G', '1') #define TAG_ISGN WINEMAKEFOURCC('I', 'S', 'G', 'N') +#define TAG_OSG1 WINEMAKEFOURCC('O', 'S', 'G', '1') #define TAG_OSG5 WINEMAKEFOURCC('O', 'S', 'G', '5') #define TAG_OSGN WINEMAKEFOURCC('O', 'S', 'G', 'N') #define TAG_PCSG WINEMAKEFOURCC('P', 'C', 'S', 'G') +#define TAG_PSG1 WINEMAKEFOURCC('P', 'S', 'G', '1') #define TAG_SHDR WINEMAKEFOURCC('S', 'H', 'D', 'R') #define TAG_SHEX WINEMAKEFOURCC('S', 'H', 'E', 'X') @@ -1955,6 +1958,7 @@ static HRESULT shader_parse_signature(DWORD tag, const char *data, DWORD data_si struct wined3d_shader_signature *s) { struct wined3d_shader_signature_element *e; + bool has_stream_index, has_min_precision; const char *ptr = data; unsigned int i; DWORD count; @@ -1982,11 +1986,14 @@ static HRESULT shader_parse_signature(DWORD tag, const char *data, DWORD data_si return E_OUTOFMEMORY; } + has_min_precision = tag == TAG_OSG1 || tag == TAG_PSG1 || tag == TAG_ISG1; + has_stream_index = tag == TAG_OSG5 || has_min_precision; + for (i = 0; i < count; ++i) { DWORD name_offset; - if (tag == TAG_OSG5) + if (has_stream_index) read_dword(&ptr, &e[i].stream_idx); else e[i].stream_idx = 0; @@ -2003,10 +2010,15 @@ static HRESULT shader_parse_signature(DWORD tag, const char *data, DWORD data_si read_dword(&ptr, &e[i].register_idx); read_dword(&ptr, &e[i].mask); + if (has_min_precision) + read_dword(&ptr, &e[i].min_precision); + else + e[i].min_precision = 0; + TRACE("Stream: %u, semantic: %s, semantic idx: %u, sysval_semantic %#x, " - "type %u, register idx: %u, use_mask %#x, input_mask %#x.\n", + "type %u, register idx: %u, use_mask %#x, input_mask %#x, min_precision %u.\n", e[i].stream_idx, debugstr_a(e[i].semantic_name), e[i].semantic_idx, e[i].sysval_semantic, - e[i].component_type, e[i].register_idx, (e[i].mask >> 8) & 0xff, e[i].mask & 0xff); + e[i].component_type, e[i].register_idx, (e[i].mask >> 8) & 0xff, e[i].mask & 0xff, e[i].min_precision); } s->elements = e; @@ -2024,6 +2036,7 @@ static HRESULT shader_dxbc_chunk_handler(const char *data, DWORD data_size, DWOR switch (tag) { case TAG_ISGN: + case TAG_ISG1: if (ctx->max_version < 4) { TRACE("Skipping shader input signature.\n"); @@ -2039,6 +2052,7 @@ static HRESULT shader_dxbc_chunk_handler(const char *data, DWORD data_size, DWOR break; case TAG_OSGN: + case TAG_OSG1: case TAG_OSG5: if (ctx->max_version < 4) { @@ -2055,6 +2069,7 @@ static HRESULT shader_dxbc_chunk_handler(const char *data, DWORD data_size, DWOR break; case TAG_PCSG: + case TAG_PSG1: if (shader->patch_constant_signature.elements) { FIXME("Multiple patch constant signatures.\n"); diff --git a/include/wine/wined3d.h b/include/wine/wined3d.h index 7a6588ea4f4..1fdbd105cc4 100644 --- a/include/wine/wined3d.h +++ b/include/wine/wined3d.h @@ -2107,6 +2107,7 @@ struct wined3d_shader_signature_element enum wined3d_component_type component_type; unsigned int register_idx; DWORD mask; + unsigned int min_precision; }; struct wined3d_shader_signature -- 2.33.0

Nov. 9, 2021

Re: [PATCH 1/3] wined3d: Test if an offset is needed to get the correct filling convention (v3).

by Francois Gouget

On Wed, 3 Nov 2021, Marvin wrote: [...] > === w1064v1809 (32 bit report) === > === w10pro64 (32 bit report) === > === debiant2 (32 bit report) === [...] Strange. I don't see any of these test failures in the WineTest results. However this patch does cause a todo to reliably succeed on Linux on the cw-gtx560 machine. But the same todo still fails on all the other configurations. It would be nice if there was an easy fix but that seems unlikely :-( https://test.winehq.org/data/patterns.html#d3drm:d3drm d3drm.c:7295: Test succeeded inside todo block: Got unexpected color 0x00bada55. -- Francois Gouget <fgouget(a)codeweavers.com>

Nov. 8, 2021

[PATCH] sechost: Fix hang when a device notification callback tries to register/unregister a notify.

by Brendan Shanks

Fixes 4 second hang in Yooka-Laylee and the Impossible Lair when a controller is connected. Signed-off-by: Brendan Shanks <bshanks(a)codeweavers.com> --- dlls/sechost/service.c | 25 +++++++++++++++++++++++-- 1 file changed, 23 insertions(+), 2 deletions(-) diff --git a/dlls/sechost/service.c b/dlls/sechost/service.c index c447263bd0f..e6f4eb75db0 100644 --- a/dlls/sechost/service.c +++ b/dlls/sechost/service.c @@ -1992,9 +1992,11 @@ static DWORD WINAPI device_notify_proc( void *arg ) RPC_WSTR binding_str; DWORD err = ERROR_SUCCESS; struct device_notify_registration *registration; + struct device_notification_details *details_copy; + unsigned int details_copy_nelems, details_copy_size; plugplay_rpc_handle handle = NULL; DWORD code = 0; - unsigned int size; + unsigned int i, size; BYTE *buf; if ((err = RpcStringBindingComposeW( NULL, protseq, NULL, endpoint, NULL, &binding_str ))) @@ -2026,6 +2028,9 @@ static DWORD WINAPI device_notify_proc( void *arg ) return 1; } + details_copy_size = 8; + details_copy = heap_alloc( details_copy_size * sizeof(*details_copy) ); + for (;;) { buf = NULL; @@ -2046,15 +2051,31 @@ static DWORD WINAPI device_notify_proc( void *arg ) break; } + /* Make a copy to avoid a hang if a callback tries to register or unregister for notifications. */ + i = 0; + details_copy_nelems = 0; EnterCriticalSection( &service_cs ); LIST_FOR_EACH_ENTRY(registration, &device_notify_list, struct device_notify_registration, entry) { - registration->details.cb( registration->details.handle, code, (DEV_BROADCAST_HDR *)buf ); + details_copy[i++] = registration->details; + details_copy_nelems++; + if (i == details_copy_size) + { + details_copy_size *= 2; + details_copy = heap_realloc( details_copy, details_copy_size * sizeof(*details_copy) ); + } } LeaveCriticalSection(&service_cs); + + for (i = 0; i < details_copy_nelems; i++) + { + details_copy[i].cb( details_copy[i].handle, code, (DEV_BROADCAST_HDR *)buf ); + } MIDL_user_free(buf); } + heap_free( details_copy ); + __TRY { plugplay_unregister_listener( handle ); -- 2.32.0

Nov. 8, 2021

[PATCH v4 6/6] wined3d: Try to allocate new Vulkan BOs from the client thread for DISCARD maps.

by Zebediah Figura

Signed-off-by: Zebediah Figura <zfigura(a)codeweavers.com> --- dlls/wined3d/adapter_gl.c | 13 +++++++++ dlls/wined3d/adapter_vk.c | 50 ++++++++++++++++++++++++++++++++++ dlls/wined3d/buffer.c | 35 ++++++++++++++++++++++-- dlls/wined3d/cs.c | 13 ++++++++- dlls/wined3d/directx.c | 12 ++++++++ dlls/wined3d/wined3d_private.h | 11 ++++++++ 6 files changed, 131 insertions(+), 3 deletions(-) diff --git a/dlls/wined3d/adapter_gl.c b/dlls/wined3d/adapter_gl.c index 7ec30cb4356..85b40b19896 100644 --- a/dlls/wined3d/adapter_gl.c +++ b/dlls/wined3d/adapter_gl.c @@ -4606,6 +4606,17 @@ static void adapter_gl_flush_bo_address(struct wined3d_context *context, wined3d_context_gl_flush_bo_address(wined3d_context_gl(context), data, size); } +static bool adapter_gl_alloc_bo(struct wined3d_device *device, struct wined3d_resource *resource, + unsigned int sub_resource_idx, struct wined3d_bo_address *addr) +{ + return false; +} + +static void adapter_gl_destroy_bo(struct wined3d_context *context, struct wined3d_bo *bo) +{ + wined3d_context_gl_destroy_bo(wined3d_context_gl(context), wined3d_bo_gl(bo)); +} + static HRESULT adapter_gl_create_swapchain(struct wined3d_device *device, struct wined3d_swapchain_desc *desc, struct wined3d_swapchain_state_parent *state_parent, void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_swapchain **swapchain) @@ -5059,6 +5070,8 @@ static const struct wined3d_adapter_ops wined3d_adapter_gl_ops = .adapter_unmap_bo_address = adapter_gl_unmap_bo_address, .adapter_copy_bo_address = adapter_gl_copy_bo_address, .adapter_flush_bo_address = adapter_gl_flush_bo_address, + .adapter_alloc_bo = adapter_gl_alloc_bo, + .adapter_destroy_bo = adapter_gl_destroy_bo, .adapter_create_swapchain = adapter_gl_create_swapchain, .adapter_destroy_swapchain = adapter_gl_destroy_swapchain, .adapter_create_buffer = adapter_gl_create_buffer, diff --git a/dlls/wined3d/adapter_vk.c b/dlls/wined3d/adapter_vk.c index 62a9a342e9b..e83f0a012d6 100644 --- a/dlls/wined3d/adapter_vk.c +++ b/dlls/wined3d/adapter_vk.c @@ -23,6 +23,7 @@ #include "wine/vulkan_driver.h" WINE_DEFAULT_DEBUG_CHANNEL(d3d); +WINE_DECLARE_DEBUG_CHANNEL(d3d_perf); static const struct wined3d_state_entry_template misc_state_template_vk[] = { @@ -1216,6 +1217,53 @@ static void adapter_vk_flush_bo_address(struct wined3d_context *context, flush_bo_range(context_vk, bo, (uintptr_t)data->addr, size); } +static bool adapter_vk_alloc_bo(struct wined3d_device *device, struct wined3d_resource *resource, + unsigned int sub_resource_idx, struct wined3d_bo_address *addr) +{ + struct wined3d_device_vk *device_vk = wined3d_device_vk(device); + struct wined3d_context_vk *context_vk = &device_vk->context_vk; + + wined3d_not_from_cs(device->cs); + assert(device->context_count); + + if (resource->type == WINED3D_RTYPE_BUFFER) + { + struct wined3d_bo_vk *bo_vk; + + if (!(bo_vk = heap_alloc(sizeof(*bo_vk)))) + return false; + + if (!(wined3d_context_vk_create_bo(context_vk, resource->size, + vk_buffer_usage_from_bind_flags(resource->bind_flags), + vk_memory_type_from_access_flags(resource->access, resource->usage), bo_vk))) + { + WARN("Failed to create Vulkan buffer.\n"); + heap_free(bo_vk); + return false; + } + + if (!bo_vk->b.map_ptr) + { + WARN_(d3d_perf)("BO %p (chunk %p, slab %p) is not persistently mapped.\n", + bo_vk, bo_vk->memory ? bo_vk->memory->chunk : NULL, bo_vk->slab); + + if (!wined3d_bo_vk_map(bo_vk, context_vk)) + ERR("Failed to map bo.\n"); + } + + addr->buffer_object = (uintptr_t)bo_vk; + addr->addr = NULL; + return true; + } + + return false; +} + +static void adapter_vk_destroy_bo(struct wined3d_context *context, struct wined3d_bo *bo) +{ + wined3d_context_vk_destroy_bo(wined3d_context_vk(context), wined3d_bo_vk(bo)); +} + static HRESULT adapter_vk_create_swapchain(struct wined3d_device *device, struct wined3d_swapchain_desc *desc, struct wined3d_swapchain_state_parent *state_parent, void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_swapchain **swapchain) @@ -1868,6 +1916,8 @@ static const struct wined3d_adapter_ops wined3d_adapter_vk_ops = .adapter_unmap_bo_address = adapter_vk_unmap_bo_address, .adapter_copy_bo_address = adapter_vk_copy_bo_address, .adapter_flush_bo_address = adapter_vk_flush_bo_address, + .adapter_alloc_bo = adapter_vk_alloc_bo, + .adapter_destroy_bo = adapter_vk_destroy_bo, .adapter_create_swapchain = adapter_vk_create_swapchain, .adapter_destroy_swapchain = adapter_vk_destroy_swapchain, .adapter_create_buffer = adapter_vk_create_buffer, diff --git a/dlls/wined3d/buffer.c b/dlls/wined3d/buffer.c index fca8ab707f2..82890cd8327 100644 --- a/dlls/wined3d/buffer.c +++ b/dlls/wined3d/buffer.c @@ -1069,6 +1069,34 @@ static HRESULT buffer_resource_sub_resource_unmap(struct wined3d_resource *resou return WINED3D_OK; } +static void wined3d_buffer_set_bo(struct wined3d_buffer *buffer, struct wined3d_context *context, struct wined3d_bo *bo) +{ + struct wined3d_bo *prev_bo = buffer->buffer_object; + + TRACE("buffer %p, context %p, bo %p.\n", buffer, context, bo); + + if (prev_bo) + { + struct wined3d_bo_user *bo_user; + + LIST_FOR_EACH_ENTRY(bo_user, &prev_bo->users, struct wined3d_bo_user, entry) + bo_user->valid = false; + assert(list_empty(&bo->users)); + list_move_head(&bo->users, &prev_bo->users); + + wined3d_context_destroy_bo(context, prev_bo); + heap_free(prev_bo); + } + else + { + list_add_head(&bo->users, &buffer->bo_user.entry); + } + + buffer->buffer_object = bo; + wined3d_buffer_validate_location(buffer, WINED3D_LOCATION_BUFFER); + wined3d_buffer_invalidate_location(buffer, ~WINED3D_LOCATION_BUFFER); +} + void wined3d_buffer_copy_bo_address(struct wined3d_buffer *dst_buffer, struct wined3d_context *context, unsigned int dst_offset, const struct wined3d_const_bo_address *src_addr, unsigned int size) { @@ -1104,6 +1132,9 @@ void wined3d_buffer_copy(struct wined3d_buffer *dst_buffer, unsigned int dst_off void wined3d_buffer_update_sub_resource(struct wined3d_buffer *buffer, struct wined3d_context *context, const struct upload_bo *upload_bo, unsigned int offset, unsigned int size) { + if (upload_bo->flags & UPLOAD_BO_RENAME_ON_UNMAP) + wined3d_buffer_set_bo(buffer, context, (struct wined3d_bo *)upload_bo->addr.buffer_object); + if (upload_bo->addr.buffer_object && upload_bo->addr.buffer_object == (uintptr_t)buffer->buffer_object) wined3d_context_flush_bo_address(context, &upload_bo->addr, size); else @@ -1418,7 +1449,7 @@ HRESULT wined3d_buffer_gl_init(struct wined3d_buffer_gl *buffer_gl, struct wined return wined3d_buffer_init(&buffer_gl->b, device, desc, data, parent, parent_ops, &wined3d_buffer_gl_ops); } -static VkBufferUsageFlags vk_buffer_usage_from_bind_flags(uint32_t bind_flags) +VkBufferUsageFlags vk_buffer_usage_from_bind_flags(uint32_t bind_flags) { VkBufferUsageFlags usage; @@ -1442,7 +1473,7 @@ static VkBufferUsageFlags vk_buffer_usage_from_bind_flags(uint32_t bind_flags) return usage; } -static VkMemoryPropertyFlags vk_memory_type_from_access_flags(uint32_t access, uint32_t usage) +VkMemoryPropertyFlags vk_memory_type_from_access_flags(uint32_t access, uint32_t usage) { VkMemoryPropertyFlags memory_type = 0; diff --git a/dlls/wined3d/cs.c b/dlls/wined3d/cs.c index 1d909e0c706..262123b5d90 100644 --- a/dlls/wined3d/cs.c +++ b/dlls/wined3d/cs.c @@ -3116,12 +3116,20 @@ static bool wined3d_cs_map_upload_bo(struct wined3d_device_context *context, str { /* Limit NOOVERWRITE maps to buffers for now; there are too many ways that * a texture can be invalidated to even count. */ - if (wined3d_map_persistent() && resource->type == WINED3D_RTYPE_BUFFER && (flags & WINED3D_MAP_NOOVERWRITE)) + if (wined3d_map_persistent() && resource->type == WINED3D_RTYPE_BUFFER + && (flags & (WINED3D_MAP_DISCARD | WINED3D_MAP_NOOVERWRITE))) { struct wined3d_client_resource *client = &resource->client; + struct wined3d_device *device = context->device; const struct wined3d_bo *bo; uint8_t *map_ptr; + if (flags & WINED3D_MAP_DISCARD) + { + if (!device->adapter->adapter_ops->adapter_alloc_bo(device, resource, sub_resource_idx, &client->addr)) + return false; + } + bo = (const struct wined3d_bo *)client->addr.buffer_object; map_ptr = bo ? bo->map_ptr : NULL; map_ptr += (uintptr_t)client->addr.addr; @@ -3145,6 +3153,9 @@ static bool wined3d_cs_map_upload_bo(struct wined3d_device_context *context, str } map_desc->data = resource_offset_map_pointer(resource, sub_resource_idx, map_ptr, box); + if (flags & WINED3D_MAP_DISCARD) + client->mapped_upload.flags |= UPLOAD_BO_UPLOAD_ON_UNMAP | UPLOAD_BO_RENAME_ON_UNMAP; + client->mapped_box = *box; TRACE("Returning bo %s, flags %#x.\n", debug_const_bo_address(&client->mapped_upload.addr), diff --git a/dlls/wined3d/directx.c b/dlls/wined3d/directx.c index c265bdc8c95..06387452b89 100644 --- a/dlls/wined3d/directx.c +++ b/dlls/wined3d/directx.c @@ -2805,6 +2805,16 @@ static void adapter_no3d_flush_bo_address(struct wined3d_context *context, { } +static bool adapter_no3d_alloc_bo(struct wined3d_device *device, struct wined3d_resource *resource, + unsigned int sub_resource_idx, struct wined3d_bo_address *addr) +{ + return false; +} + +static void adapter_no3d_destroy_bo(struct wined3d_context *context, struct wined3d_bo *bo) +{ +} + static HRESULT adapter_no3d_create_swapchain(struct wined3d_device *device, struct wined3d_swapchain_desc *desc, struct wined3d_swapchain_state_parent *state_parent, void *parent, const struct wined3d_parent_ops *parent_ops, struct wined3d_swapchain **swapchain) @@ -3075,6 +3085,8 @@ static const struct wined3d_adapter_ops wined3d_adapter_no3d_ops = .adapter_unmap_bo_address = adapter_no3d_unmap_bo_address, .adapter_copy_bo_address = adapter_no3d_copy_bo_address, .adapter_flush_bo_address = adapter_no3d_flush_bo_address, + .adapter_alloc_bo = adapter_no3d_alloc_bo, + .adapter_destroy_bo = adapter_no3d_destroy_bo, .adapter_create_swapchain = adapter_no3d_create_swapchain, .adapter_destroy_swapchain = adapter_no3d_destroy_swapchain, .adapter_create_buffer = adapter_no3d_create_buffer, diff --git a/dlls/wined3d/wined3d_private.h b/dlls/wined3d/wined3d_private.h index fc950cda3c3..0c5f7dda228 100644 --- a/dlls/wined3d/wined3d_private.h +++ b/dlls/wined3d/wined3d_private.h @@ -331,6 +331,8 @@ VkPipelineStageFlags vk_pipeline_stage_mask_from_bind_flags(uint32_t bind_flags) VkShaderStageFlagBits vk_shader_stage_from_wined3d(enum wined3d_shader_type shader_type) DECLSPEC_HIDDEN; VkAccessFlags vk_access_mask_from_buffer_usage(VkBufferUsageFlags usage) DECLSPEC_HIDDEN; VkPipelineStageFlags vk_pipeline_stage_mask_from_buffer_usage(VkBufferUsageFlags usage) DECLSPEC_HIDDEN; +VkBufferUsageFlags vk_buffer_usage_from_bind_flags(uint32_t bind_flags) DECLSPEC_HIDDEN; +VkMemoryPropertyFlags vk_memory_type_from_access_flags(uint32_t access, uint32_t usage) DECLSPEC_HIDDEN; static inline enum wined3d_cmp_func wined3d_sanitize_cmp_func(enum wined3d_cmp_func func) { @@ -3353,6 +3355,7 @@ bool wined3d_driver_info_init(struct wined3d_driver_info *driver_info, UINT64 vram_bytes, UINT64 sysmem_bytes) DECLSPEC_HIDDEN; #define UPLOAD_BO_UPLOAD_ON_UNMAP 0x1 +#define UPLOAD_BO_RENAME_ON_UNMAP 0x2 struct upload_bo { @@ -3385,6 +3388,9 @@ struct wined3d_adapter_ops const struct wined3d_bo_address *dst, const struct wined3d_bo_address *src, size_t size); void (*adapter_flush_bo_address)(struct wined3d_context *context, const struct wined3d_const_bo_address *data, size_t size); + bool (*adapter_alloc_bo)(struct wined3d_device *device, struct wined3d_resource *resource, + unsigned int sub_resource_idx, struct wined3d_bo_address *addr); + void (*adapter_destroy_bo)(struct wined3d_context *context, struct wined3d_bo *bo); HRESULT (*adapter_create_swapchain)(struct wined3d_device *device, struct wined3d_swapchain_desc *desc, struct wined3d_swapchain_state_parent *state_parent, void *parent, @@ -6307,6 +6313,11 @@ static inline void wined3d_context_flush_bo_address(struct wined3d_context *cont context->device->adapter->adapter_ops->adapter_flush_bo_address(context, data, size); } +static inline void wined3d_context_destroy_bo(struct wined3d_context *context, struct wined3d_bo *bo) +{ + context->device->adapter->adapter_ops->adapter_destroy_bo(context, bo); +} + static inline void wined3d_context_vk_reference_bo(const struct wined3d_context_vk *context_vk, struct wined3d_bo_vk *bo) { -- 2.33.0

Nov. 8, 2021