May 2024 - Wine-commits - list.winehq.org

Alexandre Julliard : wineboot: Report the correct model number on ARM platforms.
by Alexandre Julliard 14 May '24

14 May '24

Module: wine Branch: master Commit: 73c798021be9b492b35071d662b182bf8f5c310c URL: https://gitlab.winehq.org/wine/wine/-/commit/73c798021be9b492b35071d662b182… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Tue May 14 12:44:46 2024 +0200 wineboot: Report the correct model number on ARM platforms. --- programs/wineboot/wineboot.c | 9 ++++----- 1 file changed, 4 insertions(+), 5 deletions(-) diff --git a/programs/wineboot/wineboot.c b/programs/wineboot/wineboot.c index 4de60a439d2..b14754ce638 100644 --- a/programs/wineboot/wineboot.c +++ b/programs/wineboot/wineboot.c @@ -461,7 +461,6 @@ static void create_user_shared_data(void) case PROCESSOR_ARCHITECTURE_ARM: features[PF_ARM_VFP_32_REGISTERS_AVAILABLE] = !!(sci.ProcessorFeatureBits & CPU_FEATURE_ARM_VFP_32); features[PF_ARM_NEON_INSTRUCTIONS_AVAILABLE] = !!(sci.ProcessorFeatureBits & CPU_FEATURE_ARM_NEON); - features[PF_ARM_V8_INSTRUCTIONS_AVAILABLE] = (sci.ProcessorLevel >= 8); break; case PROCESSOR_ARCHITECTURE_ARM64: @@ -801,14 +800,14 @@ static void create_hardware_registry_keys(void) { case PROCESSOR_ARCHITECTURE_ARM: arch = L"ARM"; - swprintf( id, ARRAY_SIZE(id), L"ARM Family %u Model %u Revision %u", - sci.ProcessorLevel, HIBYTE(sci.ProcessorRevision), LOBYTE(sci.ProcessorRevision) ); + swprintf( id, ARRAY_SIZE(id), L"ARM Family 7 Model %X Revision %X", + sci.ProcessorLevel, sci.ProcessorRevision ); break; case PROCESSOR_ARCHITECTURE_ARM64: arch = L"ARM64"; - swprintf( id, ARRAY_SIZE(id), L"ARMv8 (64-bit) Family %u Model %u Revision %u", - sci.ProcessorLevel, HIBYTE(sci.ProcessorRevision), LOBYTE(sci.ProcessorRevision) ); + swprintf( id, ARRAY_SIZE(id), L"ARMv8 (64-bit) Family 8 Model %X Revision %X", + sci.ProcessorLevel, sci.ProcessorRevision ); break; case PROCESSOR_ARCHITECTURE_AMD64:

1 0

Alexandre Julliard : wineboot: Support multiple SMBIOS entries of the same type.
by Alexandre Julliard 14 May '24

14 May '24

Module: wine Branch: master Commit: 7097c614b282d35904b8bb7d96e24124b11c40f4 URL: https://gitlab.winehq.org/wine/wine/-/commit/7097c614b282d35904b8bb7d96e241… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Tue May 14 12:56:58 2024 +0200 wineboot: Support multiple SMBIOS entries of the same type. --- programs/wineboot/wineboot.c | 25 +++++++++++-------------- 1 file changed, 11 insertions(+), 14 deletions(-) diff --git a/programs/wineboot/wineboot.c b/programs/wineboot/wineboot.c index b14754ce638..645b7173196 100644 --- a/programs/wineboot/wineboot.c +++ b/programs/wineboot/wineboot.c @@ -604,7 +604,8 @@ struct smbios_system #define RSMB (('R' << 24) | ('S' << 16) | ('M' << 8) | 'B') -static const struct smbios_header *find_smbios_entry( enum smbios_type type, const char *buf, UINT len ) +static const struct smbios_header *find_smbios_entry( enum smbios_type type, unsigned int index, + const char *buf, UINT len ) { const char *ptr, *start; const struct smbios_prologue *prologue; @@ -630,20 +631,16 @@ static const struct smbios_header *find_smbios_entry( enum smbios_type type, con if (hdr->type == type) { if ((const char *)hdr - start + hdr->length > prologue->length) return NULL; - break; + if (!index--) return hdr; } - else /* skip other entries and their strings */ + /* skip other entries and their strings */ + for (ptr = (const char *)hdr + hdr->length; ptr - buf < len && *ptr; ptr++) { - for (ptr = (const char *)hdr + hdr->length; ptr - buf < len && *ptr; ptr++) - { - for (; ptr - buf < len; ptr++) if (!*ptr) break; - } - if (ptr == (const char *)hdr + hdr->length) ptr++; - hdr = (const struct smbios_header *)(ptr + 1); + for (; ptr - buf < len; ptr++) if (!*ptr) break; } + if (ptr == (const char *)hdr + hdr->length) ptr++; + hdr = (const struct smbios_header *)(ptr + 1); } - - return hdr; } static inline WCHAR *heap_strdupAW( const char *src ) @@ -684,7 +681,7 @@ static void create_bios_baseboard_values( HKEY bios_key, const char *buf, UINT l const struct smbios_baseboard *baseboard; UINT offset; - if (!(hdr = find_smbios_entry( SMBIOS_TYPE_BASEBOARD, buf, len ))) return; + if (!(hdr = find_smbios_entry( SMBIOS_TYPE_BASEBOARD, 0, buf, len ))) return; baseboard = (const struct smbios_baseboard *)hdr; offset = (const char *)baseboard - buf + baseboard->hdr.length; @@ -699,7 +696,7 @@ static void create_bios_bios_values( HKEY bios_key, const char *buf, UINT len ) const struct smbios_bios *bios; UINT offset; - if (!(hdr = find_smbios_entry( SMBIOS_TYPE_BIOS, buf, len ))) return; + if (!(hdr = find_smbios_entry( SMBIOS_TYPE_BIOS, 0, buf, len ))) return; bios = (const struct smbios_bios *)hdr; offset = (const char *)bios - buf + bios->hdr.length; @@ -729,7 +726,7 @@ static void create_bios_system_values( HKEY bios_key, const char *buf, UINT len const struct smbios_system *system; UINT offset; - if (!(hdr = find_smbios_entry( SMBIOS_TYPE_SYSTEM, buf, len ))) return; + if (!(hdr = find_smbios_entry( SMBIOS_TYPE_SYSTEM, 0, buf, len ))) return; system = (const struct smbios_system *)hdr; offset = (const char *)system - buf + system->hdr.length;

1 0

Alexandre Julliard : wineboot: Retrieve CPU details through the SMBIOS table.
by Alexandre Julliard 14 May '24

14 May '24

Module: wine Branch: master Commit: ecbc4cf06d6c9253ba92065f320ddd6eeff3177f URL: https://gitlab.winehq.org/wine/wine/-/commit/ecbc4cf06d6c9253ba92065f320ddd… Author: Alexandre Julliard <julliard(a)winehq.org> Date: Tue May 14 13:12:53 2024 +0200 wineboot: Retrieve CPU details through the SMBIOS table. --- programs/wineboot/wineboot.c | 284 ++++++++++++++++++++++--------------------- 1 file changed, 144 insertions(+), 140 deletions(-)

1 0

Francisco Casas : vkd3d-shader/hlsl: Improve tracking of used components running DCE before.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: e0a801e796c8c524890795dbebf804df3318e382 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/e0a801e796c8c524890795dbebf80… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Mon May 6 23:37:31 2024 -0400 vkd3d-shader/hlsl: Improve tracking of used components running DCE before. track_object_components_usage() had to be improved to also register derefs on resource stores. It was not doing it because it assumed that for every resource store there was a resource load already, which was true, before calling DCE. --- libs/vkd3d-shader/hlsl_codegen.c | 51 +++++++++++++++++++++++----------------- tests/hlsl/ternary.shader_test | 2 +- 2 files changed, 31 insertions(+), 22 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 4af55293..a05b8e3f 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -4248,34 +4248,38 @@ static bool track_object_components_sampler_dim(struct hlsl_ctx *ctx, struct hls return false; } -static bool track_object_components_usage(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) +static void register_deref_usage(struct hlsl_ctx *ctx, struct hlsl_deref *deref) { - struct hlsl_ir_resource_load *load; - struct hlsl_ir_var *var; - enum hlsl_regset regset; + struct hlsl_ir_var *var = deref->var; + enum hlsl_regset regset = hlsl_deref_get_regset(ctx, deref); unsigned int index; - if (instr->type != HLSL_IR_RESOURCE_LOAD) - return false; - - load = hlsl_ir_resource_load(instr); - var = load->resource.var; - - regset = hlsl_deref_get_regset(ctx, &load->resource); + if (!hlsl_regset_index_from_deref(ctx, deref, regset, &index)) + return; - if (!hlsl_regset_index_from_deref(ctx, &load->resource, regset, &index)) - return false; + if (regset <= HLSL_REGSET_LAST_OBJECT) + { + var->objects_usage[regset][index].used = true; + var->bind_count[regset] = max(var->bind_count[regset], index + 1); + } +} - var->objects_usage[regset][index].used = true; - var->bind_count[regset] = max(var->bind_count[regset], index + 1); - if (load->sampler.var) +static bool track_object_components_usage(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) +{ + switch (instr->type) { - var = load->sampler.var; - if (!hlsl_regset_index_from_deref(ctx, &load->sampler, HLSL_REGSET_SAMPLERS, &index)) - return false; + case HLSL_IR_RESOURCE_LOAD: + register_deref_usage(ctx, &hlsl_ir_resource_load(instr)->resource); + if (hlsl_ir_resource_load(instr)->sampler.var) + register_deref_usage(ctx, &hlsl_ir_resource_load(instr)->sampler); + break; + + case HLSL_IR_RESOURCE_STORE: + register_deref_usage(ctx, &hlsl_ir_resource_store(instr)->resource); + break; - var->objects_usage[HLSL_REGSET_SAMPLERS][index].used = true; - var->bind_count[HLSL_REGSET_SAMPLERS] = max(var->bind_count[HLSL_REGSET_SAMPLERS], index + 1); + default: + break; } return false; @@ -5440,6 +5444,11 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry hlsl_transform_ir(ctx, track_object_components_sampler_dim, body, NULL); if (profile->major_version >= 4) hlsl_transform_ir(ctx, lower_combined_samples, body, NULL); + + do + compute_liveness(ctx, entry_func); + while (hlsl_transform_ir(ctx, dce, body, NULL)); + hlsl_transform_ir(ctx, track_object_components_usage, body, NULL); sort_synthetic_separated_samplers_first(ctx); diff --git a/tests/hlsl/ternary.shader_test b/tests/hlsl/ternary.shader_test index 0ac78914..bd68dbc3 100644 --- a/tests/hlsl/ternary.shader_test +++ b/tests/hlsl/ternary.shader_test @@ -318,7 +318,7 @@ probe all rgba (2.0, 3.0, 3.0, 2.0) % Objects can be used, but their types have to be identical. -[pixel shader todo(sm<4)] +[pixel shader] Texture2D t; float4 main() : sv_target

1 0

Francisco Casas : vkd3d-shader/hlsl: Track bind count according to usage for uniforms.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 573d5113440d992d4838e70bcffb75a62a6ddc4a URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/573d5113440d992d4838e70bcffb7… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Tue May 7 02:45:14 2024 -0400 vkd3d-shader/hlsl: Track bind count according to usage for uniforms. Where bind count means the number of registers required to include all dereferences to the variable within the shader as in hlsl_ir_var.bind_count[]. --- libs/vkd3d-shader/hlsl.h | 7 +-- libs/vkd3d-shader/hlsl_codegen.c | 101 ++++++++++++++++++++++++++++++--------- 2 files changed, 82 insertions(+), 26 deletions(-) diff --git a/libs/vkd3d-shader/hlsl.h b/libs/vkd3d-shader/hlsl.h index 209d59e1..0210cf9d 100644 --- a/libs/vkd3d-shader/hlsl.h +++ b/libs/vkd3d-shader/hlsl.h @@ -447,9 +447,10 @@ struct hlsl_ir_var enum hlsl_sampler_dim sampler_dim; struct vkd3d_shader_location first_sampler_dim_loc; } *objects_usage[HLSL_REGSET_LAST_OBJECT + 1]; - /* Minimum number of binds required to include all object components actually used in the shader. - * It may be less than the allocation size, e.g. for texture arrays. */ - unsigned int bind_count[HLSL_REGSET_LAST_OBJECT + 1]; + /* Minimum number of binds required to include all components actually used in the shader. + * It may be less than the allocation size, e.g. for texture arrays. + * The bind_count for HLSL_REGSET_NUMERIC is only used in uniforms for now. */ + unsigned int bind_count[HLSL_REGSET_LAST + 1]; /* Whether the shader performs dereferences with non-constant offsets in the variable. */ bool indexable; diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index a05b8e3f..25bd5d9c 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -4252,6 +4252,8 @@ static void register_deref_usage(struct hlsl_ctx *ctx, struct hlsl_deref *deref) { struct hlsl_ir_var *var = deref->var; enum hlsl_regset regset = hlsl_deref_get_regset(ctx, deref); + uint32_t required_bind_count; + struct hlsl_type *type; unsigned int index; if (!hlsl_regset_index_from_deref(ctx, deref, regset, &index)) @@ -4262,12 +4264,39 @@ static void register_deref_usage(struct hlsl_ctx *ctx, struct hlsl_deref *deref) var->objects_usage[regset][index].used = true; var->bind_count[regset] = max(var->bind_count[regset], index + 1); } + else if (regset == HLSL_REGSET_NUMERIC) + { + type = hlsl_deref_get_type(ctx, deref); + + hlsl_regset_index_from_deref(ctx, deref, regset, &index); + required_bind_count = align(index + type->reg_size[regset], 4) / 4; + var->bind_count[regset] = max(var->bind_count[regset], required_bind_count); + } + else + { + vkd3d_unreachable(); + } } -static bool track_object_components_usage(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) +static bool track_components_usage(struct hlsl_ctx *ctx, struct hlsl_ir_node *instr, void *context) { switch (instr->type) { + case HLSL_IR_LOAD: + { + struct hlsl_ir_load *load = hlsl_ir_load(instr); + + if (!load->src.var->is_uniform) + return false; + + /* These will are handled by validate_static_object_references(). */ + if (hlsl_deref_get_regset(ctx, &load->src) != HLSL_REGSET_NUMERIC) + return false; + + register_deref_usage(ctx, &load->src); + break; + } + case HLSL_IR_RESOURCE_LOAD: register_deref_usage(ctx, &hlsl_ir_resource_load(instr)->resource); if (hlsl_ir_resource_load(instr)->sampler.var) @@ -5106,14 +5135,15 @@ bool hlsl_component_index_range_from_deref(struct hlsl_ctx *ctx, const struct hl return true; } +/* Retrieves true if the index is constant, and false otherwise. In the latter case, the maximum + * possible index is retrieved, assuming there is not out-of-bounds access. */ bool hlsl_regset_index_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, enum hlsl_regset regset, unsigned int *index) { struct hlsl_type *type = deref->var->data_type; + bool index_is_constant = true; unsigned int i; - assert(regset <= HLSL_REGSET_LAST_OBJECT); - *index = 0; for (i = 0; i < deref->path_len; ++i) @@ -5122,37 +5152,62 @@ bool hlsl_regset_index_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref unsigned int idx = 0; assert(path_node); - if (path_node->type != HLSL_IR_CONSTANT) - return false; + if (path_node->type == HLSL_IR_CONSTANT) + { + /* We should always have generated a cast to UINT. */ + assert(path_node->data_type->class == HLSL_CLASS_SCALAR + && path_node->data_type->e.numeric.type == HLSL_TYPE_UINT); - /* We should always have generated a cast to UINT. */ - assert(path_node->data_type->class == HLSL_CLASS_SCALAR - && path_node->data_type->e.numeric.type == HLSL_TYPE_UINT); + idx = hlsl_ir_constant(path_node)->value.u[0].u; - idx = hlsl_ir_constant(path_node)->value.u[0].u; + switch (type->class) + { + case HLSL_CLASS_ARRAY: + if (idx >= type->e.array.elements_count) + return false; - switch (type->class) + *index += idx * type->e.array.type->reg_size[regset]; + break; + + case HLSL_CLASS_STRUCT: + *index += type->e.record.fields[idx].reg_offset[regset]; + break; + + case HLSL_CLASS_MATRIX: + *index += 4 * idx; + break; + + default: + vkd3d_unreachable(); + } + } + else { - case HLSL_CLASS_ARRAY: - if (idx >= type->e.array.elements_count) - return false; + index_is_constant = false; - *index += idx * type->e.array.type->reg_size[regset]; - break; + switch (type->class) + { + case HLSL_CLASS_ARRAY: + idx = type->e.array.elements_count - 1; + *index += idx * type->e.array.type->reg_size[regset]; + break; - case HLSL_CLASS_STRUCT: - *index += type->e.record.fields[idx].reg_offset[regset]; - break; + case HLSL_CLASS_MATRIX: + idx = hlsl_type_major_size(type) - 1; + *index += idx * 4; + break; - default: - vkd3d_unreachable(); + default: + vkd3d_unreachable(); + } } type = hlsl_get_element_type_from_path_index(ctx, type, path_node); } - assert(type->reg_size[regset] == 1); - return true; + assert(!(regset <= HLSL_REGSET_LAST_OBJECT) || (type->reg_size[regset] == 1)); + assert(!(regset == HLSL_REGSET_NUMERIC) || type->reg_size[regset] <= 4); + return index_is_constant; } bool hlsl_offset_from_deref(struct hlsl_ctx *ctx, const struct hlsl_deref *deref, unsigned int *offset) @@ -5449,7 +5504,7 @@ int hlsl_emit_bytecode(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry compute_liveness(ctx, entry_func); while (hlsl_transform_ir(ctx, dce, body, NULL)); - hlsl_transform_ir(ctx, track_object_components_usage, body, NULL); + hlsl_transform_ir(ctx, track_components_usage, body, NULL); sort_synthetic_separated_samplers_first(ctx); if (profile->major_version < 4)

1 0

Francisco Casas : vkd3d-shader/hlsl: Only allocate numeric bind count for SM1 numeric uniforms.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: f5486442227c9e2d6ec45fcfd515660f56a1b720 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/f5486442227c9e2d6ec45fcfd5156… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri May 3 02:06:37 2024 -0400 vkd3d-shader/hlsl: Only allocate numeric bind count for SM1 numeric uniforms. Unless they have register reservations, in which case the whole variable is still reserved. --- libs/vkd3d-shader/hlsl_codegen.c | 7 +++---- tests/hlsl/sm1-const-allocation.shader_test | 4 ++-- 2 files changed, 5 insertions(+), 6 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 25bd5d9c..de778000 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -4593,15 +4593,14 @@ static void allocate_const_registers(struct hlsl_ctx *ctx, struct hlsl_ir_functi LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) { - unsigned int reg_size = var->data_type->reg_size[HLSL_REGSET_NUMERIC]; + unsigned int alloc_size = 4 * var->bind_count[HLSL_REGSET_NUMERIC]; - if (!var->is_uniform || !var->last_read || reg_size == 0) + if (!var->is_uniform || alloc_size == 0) continue; if (!var->regs[HLSL_REGSET_NUMERIC].allocated) { - var->regs[HLSL_REGSET_NUMERIC] = allocate_numeric_registers_for_type(ctx, &allocator, - 1, UINT_MAX, var->data_type); + var->regs[HLSL_REGSET_NUMERIC] = allocate_range(ctx, &allocator, 1, UINT_MAX, alloc_size); TRACE("Allocated %s to %s.\n", var->name, debug_register('c', var->regs[HLSL_REGSET_NUMERIC], var->data_type)); } diff --git a/tests/hlsl/sm1-const-allocation.shader_test b/tests/hlsl/sm1-const-allocation.shader_test index 0a4340ea..e8a60255 100644 --- a/tests/hlsl/sm1-const-allocation.shader_test +++ b/tests/hlsl/sm1-const-allocation.shader_test @@ -26,7 +26,7 @@ float4 main() : sv_target uniform 0 float4 1 2 3 4 uniform 4 float4 100 200 300 400 draw quad -todo probe all rgba (101, 202, 303, 404) +probe all rgba (101, 202, 303, 404) [pixel shader] @@ -188,7 +188,7 @@ uniform 16 float4 41 42 43 44 uniform 20 float4 51 52 53 54 uniform 24 float4 61 62 63 64 draw quad -todo probe all rgba (31, 52, 61, 62) +probe all rgba (31, 52, 61, 62) [pixel shader]

1 0

Francisco Casas : vkd3d-shader/d3dbc: Write used bind count for numeric uniforms.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: e7450ce539f95cce39d6148f867564e665363191 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/e7450ce539f95cce39d6148f86756… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri May 3 02:51:17 2024 -0400 vkd3d-shader/d3dbc: Write used bind count for numeric uniforms. --- libs/vkd3d-shader/d3dbc.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/d3dbc.c b/libs/vkd3d-shader/d3dbc.c index c7991407..7c7c71e3 100644 --- a/libs/vkd3d-shader/d3dbc.c +++ b/libs/vkd3d-shader/d3dbc.c @@ -1758,7 +1758,7 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffe if (r == HLSL_REGSET_NUMERIC) { put_u32(buffer, vkd3d_make_u32(D3DXRS_FLOAT4, var->regs[r].id)); - put_u32(buffer, var->data_type->reg_size[r] / 4); + put_u32(buffer, var->bind_count[r]); } else {

1 0

Francisco Casas : vkd3d-shader/hlsl: Allocate SM1 numeric uniforms in decreasing bind count.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 28d267b7c0238e4430a60859406fbdb5f9cf960e URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/28d267b7c0238e4430a60859406fb… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri May 3 02:51:44 2024 -0400 vkd3d-shader/hlsl: Allocate SM1 numeric uniforms in decreasing bind count. --- libs/vkd3d-shader/hlsl_codegen.c | 36 +++++++++++++++++++++++++++++ tests/hlsl/sm1-const-allocation.shader_test | 8 +++---- 2 files changed, 40 insertions(+), 4 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index de778000..26179042 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -4553,11 +4553,47 @@ static void allocate_const_registers_recurse(struct hlsl_ctx *ctx, } } +static void sort_uniform_by_numeric_bind_count(struct list *sorted, struct hlsl_ir_var *to_sort) +{ + struct hlsl_ir_var *var; + + list_remove(&to_sort->extern_entry); + + LIST_FOR_EACH_ENTRY(var, sorted, struct hlsl_ir_var, extern_entry) + { + uint32_t to_sort_size = to_sort->bind_count[HLSL_REGSET_NUMERIC]; + uint32_t var_size = var->bind_count[HLSL_REGSET_NUMERIC]; + + if (to_sort_size > var_size) + { + list_add_before(&var->extern_entry, &to_sort->extern_entry); + return; + } + } + + list_add_tail(sorted, &to_sort->extern_entry); +} + +static void sort_uniforms_by_numeric_bind_count(struct hlsl_ctx *ctx) +{ + struct list sorted = LIST_INIT(sorted); + struct hlsl_ir_var *var, *next; + + LIST_FOR_EACH_ENTRY_SAFE(var, next, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) + { + if (var->is_uniform) + sort_uniform_by_numeric_bind_count(&sorted, var); + } + list_move_tail(&ctx->extern_vars, &sorted); +} + static void allocate_const_registers(struct hlsl_ctx *ctx, struct hlsl_ir_function_decl *entry_func) { struct register_allocator allocator = {0}; struct hlsl_ir_var *var; + sort_uniforms_by_numeric_bind_count(ctx); + LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) { unsigned int reg_size = var->data_type->reg_size[HLSL_REGSET_NUMERIC]; diff --git a/tests/hlsl/sm1-const-allocation.shader_test b/tests/hlsl/sm1-const-allocation.shader_test index e8a60255..3b14c42a 100644 --- a/tests/hlsl/sm1-const-allocation.shader_test +++ b/tests/hlsl/sm1-const-allocation.shader_test @@ -50,7 +50,7 @@ uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 draw quad -todo probe all rgba (21, 22, 23, 11) +probe all rgba (21, 22, 23, 11) [pixel shader] @@ -76,7 +76,7 @@ uniform 8 float4 21 22 23 24 uniform 12 float4 31 32 33 34 uniform 16 float4 41 42 43 44 draw quad -todo probe all rgba (41, 21, 0, 0) +probe all rgba (41, 21, 0, 0) [pixel shader] @@ -224,7 +224,7 @@ uniform 20 float4 51 52 53 54 uniform 24 float4 61 62 63 64 uniform 28 float4 71 72 73 74 draw quad -todo probe all rgba (74, 31, 63, 0) +probe all rgba (74, 31, 63, 0) [pixel shader] @@ -253,7 +253,7 @@ uniform 0 float4 1 2 3 4 uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 draw quad -todo probe all rgba (21, 22, 1, 11) +probe all rgba (21, 22, 1, 11) [pixel shader]

1 0

Francisco Casas : vkd3d-shader: Return a valid pointer when count=0 in param allocator (ubsan).
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 5b7191280b7709d51b585c4f517e7ede763e01b5 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/5b7191280b7709d51b585c4f517e7… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Mon May 6 16:51:36 2024 -0400 vkd3d-shader: Return a valid pointer when count=0 in param allocator (ubsan). After compiling and linking with '-fsanitize=undefined' the following error pops up in many tests: vkd3d_shader_main.c:2024:12: runtime error: member access within null pointer of type 'struct vkd3d_shader_param_node' This happens in the scenario where shader_param_allocator_get() gets called with 'count = 0' but no allocation has been made yet, so allocator->current is NULL. In this case the result of the function, given by: params = &allocator->current->param[allocator->index * allocator->stride]; is an invalid non-NULL pointer. Functions like shader_sm4_read_instruction() may call vsir_program_get_src_params() or vsir_program_get_dst_params() with 0 counts for various DCL_ instructions, as well as things like NOP, ELSE, and SYNC. We could avoid calling the functions in question with 0 counts, but it doesn't seem worth the effort. Alternatively, we could just return NULL on 'count == 0', but this is also complicated because NULL is interpreted as a memory allocation failure on the callers. So we force allocation of the next node even if 'count = 0' when allocator->current is NULL. --- libs/vkd3d-shader/vkd3d_shader_main.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/vkd3d_shader_main.c b/libs/vkd3d-shader/vkd3d_shader_main.c index a7b42217..18456480 100644 --- a/libs/vkd3d-shader/vkd3d_shader_main.c +++ b/libs/vkd3d-shader/vkd3d_shader_main.c @@ -2004,7 +2004,7 @@ void *shader_param_allocator_get(struct vkd3d_shader_param_allocator *allocator, { void *params; - if (count > allocator->count - allocator->index) + if (!allocator->current || count > allocator->count - allocator->index) { struct vkd3d_shader_param_node *next;

1 0

Conor McCarthy : vkd3d-shader/dxil: Emit bool inputs and outputs as uint.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: ea5624e5e866caac25bdf79c3c5b6c7925d0b481 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/ea5624e5e866caac25bdf79c3c5b6… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Fri Jul 28 16:14:13 2023 +1000 vkd3d-shader/dxil: Emit bool inputs and outputs as uint. DXIL reads/writes uint for these. --- libs/vkd3d-shader/dxil.c | 4 ++++ tests/hlsl/bool-semantics.shader_test | 2 +- 2 files changed, 5 insertions(+), 1 deletion(-) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 811a7c97..ab165910 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -9387,6 +9387,10 @@ static enum vkd3d_result sm6_parser_read_signature(struct sm6_parser *sm6, const e->used_mask <<= index; } + /* DXIL reads/writes uint for bool I/O. */ + if (e->component_type == VKD3D_SHADER_COMPONENT_BOOL) + e->component_type = VKD3D_SHADER_COMPONENT_UINT; + m = element_node->operands[4]; if (!sm6_metadata_value_is_node(m)) { diff --git a/tests/hlsl/bool-semantics.shader_test b/tests/hlsl/bool-semantics.shader_test index af945397..0214e957 100644 --- a/tests/hlsl/bool-semantics.shader_test +++ b/tests/hlsl/bool-semantics.shader_test @@ -49,5 +49,5 @@ float4 main(struct input i) : sv_target } [test] -todo(sm>=6 | glsl) draw triangle strip 4 +todo(glsl) draw triangle strip 4 probe all rgba (0.0, 2.0, 2.0, 2.0)

1 0