January 2024 - Wine-Commits

Giovanni Mascellani : vkd3d-shader: Convert write masks between 32 and 64 bit by cases.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 2b93aae069c9b6920db08d3fe4a7f68f18671946 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/2b93aae069c9b6920db08d3fe4a7f… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Sat Jan 20 23:30:23 2024 +0100 vkd3d-shader: Convert write masks between 32 and 64 bit by cases. There are only three cases, and while the code is longer it is also hopefully easier to read. Moreover, an error message is casted if we're doing something unexpected. --- libs/vkd3d-shader/vkd3d_shader_private.h | 35 +++++++++++++++++++++++++++----- 1 file changed, 30 insertions(+), 5 deletions(-) diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index f8c4d61e..cbac998a 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1605,15 +1605,40 @@ static inline unsigned int vkd3d_write_mask_from_component_count(unsigned int co static inline uint32_t vsir_write_mask_64_from_32(uint32_t write_mask32) { - uint32_t write_mask64 = write_mask32 | (write_mask32 >> 1); - return (write_mask64 & VKD3DSP_WRITEMASK_0) | ((write_mask64 & VKD3DSP_WRITEMASK_2) >> 1); + switch (write_mask32) + { + case VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_1: + return VKD3DSP_WRITEMASK_0; + + case VKD3DSP_WRITEMASK_2 | VKD3DSP_WRITEMASK_3: + return VKD3DSP_WRITEMASK_1; + + case VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_1 | VKD3DSP_WRITEMASK_2 | VKD3DSP_WRITEMASK_3: + return VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_1; + + default: + ERR("Invalid 32 bit writemask when converting to 64 bit: %#x.\n", write_mask32); + return VKD3DSP_WRITEMASK_0; + } } static inline uint32_t vsir_write_mask_32_from_64(uint32_t write_mask64) { - uint32_t write_mask32 = (write_mask64 | (write_mask64 << 1)) - & (VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_2); - return write_mask32 | (write_mask32 << 1); + switch (write_mask64) + { + case VKD3DSP_WRITEMASK_0: + return VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_1; + + case VKD3DSP_WRITEMASK_1: + return VKD3DSP_WRITEMASK_2 | VKD3DSP_WRITEMASK_3; + + case VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_1: + return VKD3DSP_WRITEMASK_0 | VKD3DSP_WRITEMASK_1 | VKD3DSP_WRITEMASK_2 | VKD3DSP_WRITEMASK_3; + + default: + ERR("Invalid 64 bit writemask: %#x.\n", write_mask64); + return VKD3DSP_WRITEMASK_0; + } } static inline unsigned int vsir_swizzle_get_component(uint32_t swizzle, unsigned int idx)

1 0

Francisco Casas : tests: Test register(cX) reservations.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 4791c7c23bda7c98c7e122c8818aeb2000c70d2d URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/4791c7c23bda7c98c7e122c8818ae… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Wed Nov 8 19:00:27 2023 -0300 tests: Test register(cX) reservations. --- Makefile.am | 1 + .../hlsl/register-reservations-numeric.shader_test | 279 +++++++++++++++++++++ 2 files changed, 280 insertions(+) diff --git a/Makefile.am b/Makefile.am index 0cb97b66..90e7dcfc 100644 --- a/Makefile.am +++ b/Makefile.am @@ -152,6 +152,7 @@ vkd3d_shader_tests = \ tests/hlsl/object-references.shader_test \ tests/hlsl/pow.shader_test \ tests/hlsl/reflect.shader_test \ + tests/hlsl/register-reservations-numeric.shader_test \ tests/hlsl/register-reservations-resources.shader_test \ tests/hlsl/return-implicit-conversion.shader_test \ tests/hlsl/return.shader_test \ diff --git a/tests/hlsl/register-reservations-numeric.shader_test b/tests/hlsl/register-reservations-numeric.shader_test new file mode 100644 index 00000000..f89cb139 --- /dev/null +++ b/tests/hlsl/register-reservations-numeric.shader_test @@ -0,0 +1,279 @@ +[pixel shader fail(sm<6) todo] +// Overlapping register(cX) reservations are not allowed except on SM6, where they are aliased. +// On SM1 this gives hr 0x88760b59. +float a : register(c0); +float b : register(c0); + +float4 main() : sv_target +{ + return a + b; +} + + +[pixel shader] +// It is not required to provide a register(cX) for all elements in the $Globals buffer. +float4 a; // will get register(c1) +float4 b : register(c0); + +float4 main() : sv_target +{ + return float4(a.xw, b.yz); +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +uniform 4 float4 1.1 1.2 1.3 1.4 +draw quad +todo(sm<6) probe all rgba (1.1, 1.4, 0.2, 0.3) + + +[pixel shader] +float4 a[3]; // will get register(c3) +float4 b[2] : register(c1); + +float4 main() : sv_target +{ + return float4(a[1].xy, b[0].zw); +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +uniform 4 float4 1.1 1.2 1.3 1.4 +uniform 8 float4 2.1 2.2 2.3 2.4 +uniform 12 float4 3.1 3.2 3.3 3.4 +uniform 16 float4 4.1 4.2 4.3 4.4 +draw quad +todo(sm<6) probe all rgba (4.1, 4.2, 1.3, 1.4) + + +[require] +shader model < 4.0 + +[pixel shader] +float a : register(c2); +float b; // will get register c0 in SM1 + +float4 main() : sv_target +{ + return float4(a, b, 0.0, 0.0); +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +uniform 4 float4 1.1 1.2 1.3 1.4 +uniform 8 float4 2.1 2.2 2.3 2.4 +uniform 12 float4 3.1 3.2 3.3 3.4 +draw quad +todo probe all rgba (2.1, 0.1, 0.0, 0.0) + + +[require] +shader model >= 4.0 + +[pixel shader] +float a : register(c2); +float b; // will get offset equivalent to c2.y in SM4 and SM6 + +float4 main() : sv_target +{ + return float4(a, b, 0.0, 0.0); +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +uniform 4 float4 1.1 1.2 1.3 1.4 +uniform 8 float4 2.1 2.2 2.3 2.4 +uniform 12 float4 3.1 3.2 3.3 3.4 +draw quad +todo(sm<6) probe all rgba (2.1, 2.2, 0.0, 0.0) + + +[require] +shader model >= 6.0 + +[pixel shader] +// Variables with overlapping register(cX) reservations are aliased in SM6. +float2 a : register(c2); +float3 b : register(c2); + +float4 main() : sv_target +{ + return float4(a, b.yz); +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +uniform 4 float4 1.1 1.2 1.3 1.4 +uniform 8 float4 2.1 2.2 2.3 2.4 +draw quad +probe all rgba (2.1, 2.2, 2.2, 2.3) + + +% Results differ between SM1 and SM4 because in the latter variables can share the same register, +% using different writemasks. +[require] +shader model < 4.0 + +[pixel shader] +struct +{ + float2 a; + float b; +} apple : register(c2); + +float4 main() : sv_target +{ + return float4(apple.a, apple.b, 0); +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +uniform 4 float4 1.1 1.2 1.3 1.4 +uniform 8 float4 2.1 2.2 2.3 2.4 +uniform 12 float4 3.1 3.2 3.3 3.4 +draw quad +todo probe all rgba (2.1, 2.2, 3.1, 0.0) + + +[require] +shader model >= 4.0 + +[pixel shader] +struct +{ + float2 a; + float b; +} apple : register(c2); + +float4 main() : sv_target +{ + return float4(apple.a, apple.b, 0); +} + +[test] +uniform 0 float4 0.1 0.2 0.3 0.4 +uniform 4 float4 1.1 1.2 1.3 1.4 +uniform 8 float4 2.1 2.2 2.3 2.4 +uniform 12 float4 3.1 3.2 3.3 3.4 +draw quad +todo(sm<6) probe all rgba (2.1, 2.2, 2.3, 0.0) + + +[pixel shader] +// On SM4, register(cX) has no effect unless in the $Globals buffer. +cbuffer extra +{ + float a : register(c1); +}; + +float4 main() : sv_target +{ + return a; +} + +[test] +uniform 0 float 100 +uniform 4 float 101 +draw quad +probe all rgba (100, 100, 100, 100) + + +[pixel shader fail(sm>=6)] +// On SM4 register(cX) has no effect unless in the $Globals buffer. +float4 main(uniform float a : register(c1)) : sv_target +{ + return a; +} + +[test] +uniform 0 float 100 +uniform 4 float 101 +draw quad +probe all rgba (100, 100, 100, 100) + +[pixel shader todo] +cbuffer c +{ + float a : packoffset(c1); + float b : packoffset(c2) : register(c1); + // ^ register(c1) ignored for cbuffer that is not $Globals. +} + +float4 main() : sv_target +{ + return float4(a, b, 0, 0); +} + +[test] +uniform 0 float 200 +uniform 4 float 201 +uniform 8 float 202 +todo(sm<6) draw quad +todo(sm<6) probe all rgba (201.0, 202.0, 0.0, 0.0) + + +[pixel shader fail(sm<4)] +int k : register(i0); // register(cX) is also required. + +float4 main() : sv_target +{ + return k; +} + + +[require] +% All shader models. + +% In SM1, most variables are needed in the "c" register group, for float operations. +% If a variable is needed in the "c" register group, register() reservations in other groups can be +% provided only if a register(cX) reservation is also provided. + +[pixel shader fail(sm<4) todo(sm<4)] +int k : register(i0); +// ^^ register(cX) is also required in SM1. + +float4 main() : sv_target +{ + return k; +} + +[pixel shader todo] +int k : register(i0) : register(c1); +// Shader compiles because a "c" register reservation is provided for "k". + +float4 main() : sv_target +{ + return k; +} + + +[require] +shader model >= 3.0 +% model 2.0 doesn't support unrollable loops. + +[pixel shader todo(sm<4)] +int k : register(i0); +// ^^ register(cX) is not required since "k" is just needed in the "i" register group. + +float4 main() : sv_target +{ + float f = 0; + + for (int i = 0; i < k; ++i) + f += i; + return f; +} + + +[pixel shader todo] +int k : register(c0) : register(b0); +// ^^ unlike the "c" register group, a reservation is not required for the "i" group, even though "k" is needed on it. + +float4 main() : sv_target +{ + float f = 0; + + for (int i = 0; i < k; ++i) + f += i; + return f; +}

1 0

Francisco Casas : vkd3d-shader/hlsl: Make register(cX) reservations work for SM1.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: c7f76365341b149578392233cf25024f21fcdf14 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/c7f76365341b149578392233cf250… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Tue Nov 7 21:13:52 2023 -0300 vkd3d-shader/hlsl: Make register(cX) reservations work for SM1. --- libs/vkd3d-shader/hlsl_codegen.c | 41 +++++++++++++++++++--- .../hlsl/register-reservations-numeric.shader_test | 10 +++--- 2 files changed, 42 insertions(+), 9 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 1fe141a3..2d2e904b 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -4123,13 +4123,46 @@ 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) { - if (var->is_uniform && var->last_read) + unsigned int reg_size = var->data_type->reg_size[HLSL_REGSET_NUMERIC]; + + if (!var->is_uniform || !var->last_read || reg_size == 0) + continue; + + if (var->reg_reservation.reg_type == 'c') { - unsigned int reg_size = var->data_type->reg_size[HLSL_REGSET_NUMERIC]; + unsigned int reg_idx = var->reg_reservation.reg_index; + unsigned int i; - if (reg_size == 0) - continue; + assert(reg_size % 4 == 0); + for (i = 0; i < reg_size / 4; ++i) + { + if (get_available_writemask(&allocator, 1, UINT_MAX, reg_idx + i) != VKD3DSP_WRITEMASK_ALL) + { + hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, + "Overlapping register() reservations on 'c%u'.", reg_idx + i); + } + + record_allocation(ctx, &allocator, reg_idx + i, VKD3DSP_WRITEMASK_ALL, 1, UINT_MAX); + } + var->regs[HLSL_REGSET_NUMERIC].id = reg_idx; + var->regs[HLSL_REGSET_NUMERIC].allocation_size = reg_size / 4; + var->regs[HLSL_REGSET_NUMERIC].writemask = VKD3DSP_WRITEMASK_ALL; + var->regs[HLSL_REGSET_NUMERIC].allocated = true; + TRACE("Allocated reserved %s to %s.\n", var->name, + debug_register('c', var->regs[HLSL_REGSET_NUMERIC], var->data_type)); + } + } + + 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]; + + if (!var->is_uniform || !var->last_read || reg_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); TRACE("Allocated %s to %s.\n", var->name, diff --git a/tests/hlsl/register-reservations-numeric.shader_test b/tests/hlsl/register-reservations-numeric.shader_test index f89cb139..9b2bae57 100644 --- a/tests/hlsl/register-reservations-numeric.shader_test +++ b/tests/hlsl/register-reservations-numeric.shader_test @@ -1,4 +1,4 @@ -[pixel shader fail(sm<6) todo] +[pixel shader fail(sm<6) todo(sm>=4)] // Overlapping register(cX) reservations are not allowed except on SM6, where they are aliased. // On SM1 this gives hr 0x88760b59. float a : register(c0); @@ -24,7 +24,7 @@ float4 main() : sv_target uniform 0 float4 0.1 0.2 0.3 0.4 uniform 4 float4 1.1 1.2 1.3 1.4 draw quad -todo(sm<6) probe all rgba (1.1, 1.4, 0.2, 0.3) +todo(sm>=4 & sm<6) probe all rgba (1.1, 1.4, 0.2, 0.3) [pixel shader] @@ -43,7 +43,7 @@ uniform 8 float4 2.1 2.2 2.3 2.4 uniform 12 float4 3.1 3.2 3.3 3.4 uniform 16 float4 4.1 4.2 4.3 4.4 draw quad -todo(sm<6) probe all rgba (4.1, 4.2, 1.3, 1.4) +todo(sm>=4 & sm<6) probe all rgba (4.1, 4.2, 1.3, 1.4) [require] @@ -64,7 +64,7 @@ uniform 4 float4 1.1 1.2 1.3 1.4 uniform 8 float4 2.1 2.2 2.3 2.4 uniform 12 float4 3.1 3.2 3.3 3.4 draw quad -todo probe all rgba (2.1, 0.1, 0.0, 0.0) +probe all rgba (2.1, 0.1, 0.0, 0.0) [require] @@ -132,7 +132,7 @@ uniform 4 float4 1.1 1.2 1.3 1.4 uniform 8 float4 2.1 2.2 2.3 2.4 uniform 12 float4 3.1 3.2 3.3 3.4 draw quad -todo probe all rgba (2.1, 2.2, 3.1, 0.0) +probe all rgba (2.1, 2.2, 3.1, 0.0) [require]

1 0

Francisco Casas : vkd3d-shader/hlsl: Turn register(cX) reservations into buffer offset for SM4.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 13f713f74bdd3ab3650cbe72a7db2e872ab41cdf URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/13f713f74bdd3ab3650cbe72a7db2… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Wed Nov 8 18:45:26 2023 -0300 vkd3d-shader/hlsl: Turn register(cX) reservations into buffer offset for SM4. --- libs/vkd3d-shader/hlsl_codegen.c | 96 ++++++++++++++-------- .../hlsl/register-reservations-numeric.shader_test | 10 +-- 2 files changed, 67 insertions(+), 39 deletions(-) diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 2d2e904b..6ad60e4c 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -4302,45 +4302,52 @@ static const struct hlsl_buffer *get_reserved_buffer(struct hlsl_ctx *ctx, uint3 return NULL; } -static void calculate_buffer_offset(struct hlsl_ctx *ctx, struct hlsl_ir_var *var) +static void calculate_buffer_offset(struct hlsl_ctx *ctx, struct hlsl_ir_var *var, bool register_reservation) { unsigned int var_reg_size = var->data_type->reg_size[HLSL_REGSET_NUMERIC]; enum hlsl_type_class var_class = var->data_type->class; struct hlsl_buffer *buffer = var->buffer; - if (var->reg_reservation.offset_type == 'c') + if (register_reservation) { - if (var->reg_reservation.offset_index % 4) + var->buffer_offset = 4 * var->reg_reservation.reg_index; + } + else + { + if (var->reg_reservation.offset_type == 'c') { - if (var_class == HLSL_CLASS_MATRIX) - { - hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, - "packoffset() reservations with matrix types must be aligned with the beginning of a register."); - } - else if (var_class == HLSL_CLASS_ARRAY) - { - hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, - "packoffset() reservations with array types must be aligned with the beginning of a register."); - } - else if (var_class == HLSL_CLASS_STRUCT) - { - hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, - "packoffset() reservations with struct types must be aligned with the beginning of a register."); - } - else if (var_class == HLSL_CLASS_VECTOR) + if (var->reg_reservation.offset_index % 4) { - unsigned int aligned_offset = hlsl_type_get_sm4_offset(var->data_type, var->reg_reservation.offset_index); - - if (var->reg_reservation.offset_index != aligned_offset) + if (var_class == HLSL_CLASS_MATRIX) + { hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, - "packoffset() reservations with vector types cannot span multiple registers."); + "packoffset() reservations with matrix types must be aligned with the beginning of a register."); + } + else if (var_class == HLSL_CLASS_ARRAY) + { + hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, + "packoffset() reservations with array types must be aligned with the beginning of a register."); + } + else if (var_class == HLSL_CLASS_STRUCT) + { + hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, + "packoffset() reservations with struct types must be aligned with the beginning of a register."); + } + else if (var_class == HLSL_CLASS_VECTOR) + { + unsigned int aligned_offset = hlsl_type_get_sm4_offset(var->data_type, var->reg_reservation.offset_index); + + if (var->reg_reservation.offset_index != aligned_offset) + hlsl_error(ctx, &var->loc, VKD3D_SHADER_ERROR_HLSL_INVALID_RESERVATION, + "packoffset() reservations with vector types cannot span multiple registers."); + } } + var->buffer_offset = var->reg_reservation.offset_index; + } + else + { + var->buffer_offset = hlsl_type_get_sm4_offset(var->data_type, buffer->size); } - var->buffer_offset = var->reg_reservation.offset_index; - } - else - { - var->buffer_offset = hlsl_type_get_sm4_offset(var->data_type, buffer->size); } TRACE("Allocated buffer offset %u to %s.\n", var->buffer_offset, var->name); @@ -4409,6 +4416,11 @@ static void validate_buffer_offsets(struct hlsl_ctx *ctx) } } +static bool var_has_buffer_offset_register_reservation(struct hlsl_ctx *ctx, const struct hlsl_ir_var *var) +{ + return var->reg_reservation.reg_type == 'c' && var->buffer == ctx->globals_buffer; +} + static void allocate_buffers(struct hlsl_ctx *ctx) { struct hlsl_buffer *buffer; @@ -4417,13 +4429,29 @@ static void allocate_buffers(struct hlsl_ctx *ctx) LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) { - if (var->is_uniform && !hlsl_type_is_resource(var->data_type)) - { - if (var->is_param) - var->buffer = ctx->params_buffer; + if (!var->is_uniform || hlsl_type_is_resource(var->data_type)) + continue; - calculate_buffer_offset(ctx, var); - } + if (var->is_param) + var->buffer = ctx->params_buffer; + } + + LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) + { + if (!var->is_uniform || hlsl_type_is_resource(var->data_type)) + continue; + + if (var_has_buffer_offset_register_reservation(ctx, var)) + calculate_buffer_offset(ctx, var, true); + } + + LIST_FOR_EACH_ENTRY(var, &ctx->extern_vars, struct hlsl_ir_var, extern_entry) + { + if (!var->is_uniform || hlsl_type_is_resource(var->data_type)) + continue; + + if (!var_has_buffer_offset_register_reservation(ctx, var)) + calculate_buffer_offset(ctx, var, false); } validate_buffer_offsets(ctx); diff --git a/tests/hlsl/register-reservations-numeric.shader_test b/tests/hlsl/register-reservations-numeric.shader_test index 9b2bae57..5067d54f 100644 --- a/tests/hlsl/register-reservations-numeric.shader_test +++ b/tests/hlsl/register-reservations-numeric.shader_test @@ -1,4 +1,4 @@ -[pixel shader fail(sm<6) todo(sm>=4)] +[pixel shader fail(sm<6)] // Overlapping register(cX) reservations are not allowed except on SM6, where they are aliased. // On SM1 this gives hr 0x88760b59. float a : register(c0); @@ -24,7 +24,7 @@ float4 main() : sv_target uniform 0 float4 0.1 0.2 0.3 0.4 uniform 4 float4 1.1 1.2 1.3 1.4 draw quad -todo(sm>=4 & sm<6) probe all rgba (1.1, 1.4, 0.2, 0.3) +probe all rgba (1.1, 1.4, 0.2, 0.3) [pixel shader] @@ -43,7 +43,7 @@ uniform 8 float4 2.1 2.2 2.3 2.4 uniform 12 float4 3.1 3.2 3.3 3.4 uniform 16 float4 4.1 4.2 4.3 4.4 draw quad -todo(sm>=4 & sm<6) probe all rgba (4.1, 4.2, 1.3, 1.4) +probe all rgba (4.1, 4.2, 1.3, 1.4) [require] @@ -85,7 +85,7 @@ uniform 4 float4 1.1 1.2 1.3 1.4 uniform 8 float4 2.1 2.2 2.3 2.4 uniform 12 float4 3.1 3.2 3.3 3.4 draw quad -todo(sm<6) probe all rgba (2.1, 2.2, 0.0, 0.0) +probe all rgba (2.1, 2.2, 0.0, 0.0) [require] @@ -156,7 +156,7 @@ uniform 4 float4 1.1 1.2 1.3 1.4 uniform 8 float4 2.1 2.2 2.3 2.4 uniform 12 float4 3.1 3.2 3.3 3.4 draw quad -todo(sm<6) probe all rgba (2.1, 2.2, 2.3, 0.0) +probe all rgba (2.1, 2.2, 2.3, 0.0) [pixel shader]

1 0

Giovanni Mascellani : include/vkd3d-shader: Generate swizzles without a function call.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 5ec1825eb31e68838f5763f17b920e7cd85003cd URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/5ec1825eb31e68838f5763f17b920… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Sat Jan 20 23:28:14 2024 +0100 include/vkd3d-shader: Generate swizzles without a function call. So they can be used when a constant expression is expected, for instance on case labels. --- include/vkd3d_shader.h | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/include/vkd3d_shader.h b/include/vkd3d_shader.h index a8cc3a33..2f4478a7 100644 --- a/include/vkd3d_shader.h +++ b/include/vkd3d_shader.h @@ -1777,10 +1777,10 @@ struct vkd3d_shader_dxbc_desc * \endcode */ #define VKD3D_SHADER_SWIZZLE(x, y, z, w) \ - vkd3d_shader_create_swizzle(VKD3D_SHADER_SWIZZLE_ ## x, \ - VKD3D_SHADER_SWIZZLE_ ## y, \ - VKD3D_SHADER_SWIZZLE_ ## z, \ - VKD3D_SHADER_SWIZZLE_ ## w) + (VKD3D_SHADER_SWIZZLE_ ## x << VKD3D_SHADER_SWIZZLE_SHIFT(0) \ + | VKD3D_SHADER_SWIZZLE_ ## y << VKD3D_SHADER_SWIZZLE_SHIFT(1) \ + | VKD3D_SHADER_SWIZZLE_ ## z << VKD3D_SHADER_SWIZZLE_SHIFT(2) \ + | VKD3D_SHADER_SWIZZLE_ ## w << VKD3D_SHADER_SWIZZLE_SHIFT(3)) /** The identity swizzle ".xyzw". */ #define VKD3D_SHADER_NO_SWIZZLE VKD3D_SHADER_SWIZZLE(X, Y, Z, W)

1 0

Francisco Casas : tests: Rename register-reservations.shader_test to register-reservations-resources.shader_test.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: d7d4c9bc909ab0cbe0e3f78357785071a741c147 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/d7d4c9bc909ab0cbe0e3f78357785… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri Nov 10 17:06:16 2023 -0300 tests: Rename register-reservations.shader_test to register-reservations-resources.shader_test. --- Makefile.am | 2 +- ...rvations.shader_test => register-reservations-resources.shader_test} | 0 2 files changed, 1 insertion(+), 1 deletion(-) diff --git a/Makefile.am b/Makefile.am index bfd11fdb..0cb97b66 100644 --- a/Makefile.am +++ b/Makefile.am @@ -152,7 +152,7 @@ vkd3d_shader_tests = \ tests/hlsl/object-references.shader_test \ tests/hlsl/pow.shader_test \ tests/hlsl/reflect.shader_test \ - tests/hlsl/register-reservations.shader_test \ + tests/hlsl/register-reservations-resources.shader_test \ tests/hlsl/return-implicit-conversion.shader_test \ tests/hlsl/return.shader_test \ tests/hlsl/round.shader_test \ diff --git a/tests/hlsl/register-reservations.shader_test b/tests/hlsl/register-reservations-resources.shader_test similarity index 100% rename from tests/hlsl/register-reservations.shader_test rename to tests/hlsl/register-reservations-resources.shader_test

1 0

Conor McCarthy : vkd3d-shader/spirv: Support 64-bit source value for bitfield instructions.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 92f0b371331cf1d0fa54ea2f1343f7bca4a00601 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/92f0b371331cf1d0fa54ea2f1343f… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Tue Jan 23 12:48:22 2024 +1000 vkd3d-shader/spirv: Support 64-bit source value for bitfield instructions. --- libs/vkd3d-shader/spirv.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 91411bf6..ef12d32b 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -7472,7 +7472,7 @@ static void spirv_compiler_emit_bitfield_instruction(struct spirv_compiler *comp const struct vkd3d_shader_dst_param *dst = instruction->dst; const struct vkd3d_shader_src_param *src = instruction->src; enum vkd3d_shader_component_type component_type; - unsigned int i, j, k, src_count; + unsigned int i, j, k, src_count, size; uint32_t write_mask; SpvOp op; @@ -7481,8 +7481,9 @@ static void spirv_compiler_emit_bitfield_instruction(struct spirv_compiler *comp component_type = vkd3d_component_type_from_data_type(dst->reg.data_type); type_id = vkd3d_spirv_get_type_id(builder, component_type, 1); - mask_id = spirv_compiler_get_constant_uint(compiler, 0x1f); - size_id = spirv_compiler_get_constant_uint(compiler, 0x20); + size = (src[src_count - 1].reg.data_type == VKD3D_DATA_UINT64) ? 0x40 : 0x20; + mask_id = spirv_compiler_get_constant_uint(compiler, size - 1); + size_id = spirv_compiler_get_constant_uint(compiler, size); switch (instruction->handler_idx) {

1 0

Conor McCarthy : vkd3d-shader/dxil: Implement DX intrinsic Tertiary.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 3bbe374ca8b9bb046a7e7cc06de257c0660d2605 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/3bbe374ca8b9bb046a7e7cc06de25… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Tue Jan 23 12:32:39 2024 +1000 vkd3d-shader/dxil: Implement DX intrinsic Tertiary. IBFE and UBFE are not emitted for HLSL sources which perform bitfield extractions, e.g. loading a value from a struct containing bitfields, or the equivalent done with bit shifts. These instructions are probably only emitted by the TPF -> DXIL converter, which makes them hard to test. --- libs/vkd3d-shader/dxil.c | 32 ++++++++++++++++++++++++++++++++ 1 file changed, 32 insertions(+) diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index ebe600e1..7a4e2ede 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -354,6 +354,8 @@ enum dx_intrinsic_opcode DX_IMIN = 38, DX_UMAX = 39, DX_UMIN = 40, + DX_IBFE = 51, + DX_UBFE = 52, DX_CREATE_HANDLE = 57, DX_CBUFFER_LOAD_LEGACY = 59, DX_BUFFER_LOAD = 68, @@ -3701,6 +3703,34 @@ static void sm6_parser_emit_dx_create_handle(struct sm6_parser *sm6, enum dx_int ins->handler_idx = VKD3DSIH_NOP; } +static enum vkd3d_shader_opcode sm6_dx_map_tertiary_op(enum dx_intrinsic_opcode op) +{ + switch (op) + { + case DX_IBFE: + return VKD3DSIH_IBFE; + case DX_UBFE: + return VKD3DSIH_UBFE; + default: + vkd3d_unreachable(); + } +} + +static void sm6_parser_emit_dx_tertiary(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, + const struct sm6_value **operands, struct function_emission_state *state) +{ + struct vkd3d_shader_instruction *ins = state->ins; + struct vkd3d_shader_src_param *src_params; + unsigned int i; + + vsir_instruction_init(ins, &sm6->p.location, sm6_dx_map_tertiary_op(op)); + src_params = instruction_src_params_alloc(ins, 3, sm6); + for (i = 0; i < 3; ++i) + src_param_init_from_value(&src_params[i], operands[i]); + + instruction_dst_param_init_ssa_scalar(ins, sm6); +} + static void sm6_parser_emit_dx_load_input(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, const struct sm6_value **operands, struct function_emission_state *state) { @@ -3936,6 +3966,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_FMAX ] = {"g", "RR", sm6_parser_emit_dx_binary}, [DX_FMIN ] = {"g", "RR", sm6_parser_emit_dx_binary}, [DX_FRC ] = {"g", "R", sm6_parser_emit_dx_unary}, + [DX_IBFE ] = {"m", "iiR", sm6_parser_emit_dx_tertiary}, [DX_IMAX ] = {"m", "RR", sm6_parser_emit_dx_binary}, [DX_IMIN ] = {"m", "RR", sm6_parser_emit_dx_binary}, [DX_ISFINITE ] = {"1", "g", sm6_parser_emit_dx_unary}, @@ -3955,6 +3986,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_SQRT ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_STORE_OUTPUT ] = {"v", "ii8o", sm6_parser_emit_dx_store_output}, [DX_TAN ] = {"g", "R", sm6_parser_emit_dx_unary}, + [DX_UBFE ] = {"m", "iiR", sm6_parser_emit_dx_tertiary}, [DX_UMAX ] = {"m", "RR", sm6_parser_emit_dx_binary}, [DX_UMIN ] = {"m", "RR", sm6_parser_emit_dx_binary}, };

1 0

Giovanni Mascellani : vkd3d-shader/ir: Fixup PHI nodes when lowering switches to selection ladders.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 026c502f31cfcdaf2679e1ece74766598916e9c1 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/026c502f31cfcdaf2679e1ece7476… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Mon Jan 15 17:02:07 2024 +0100 vkd3d-shader/ir: Fixup PHI nodes when lowering switches to selection ladders. A map between the blocks before and after the pass is built and then used to fix the PHI nodes. --- libs/vkd3d-shader/ir.c | 158 ++++++++++++++++++++++++++++++++++++++++++++++--- 1 file changed, 149 insertions(+), 9 deletions(-) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index c2568316..1bb35fd7 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -2388,15 +2388,47 @@ static bool reserve_instructions(struct vkd3d_shader_instruction **instructions, return true; } +/* A record represents replacing a jump from block `switch_label' to + * block `target_label' with a jump from block `if_label' to block + * `target_label'. */ +struct lower_switch_to_if_ladder_block_mapping +{ + unsigned int switch_label; + unsigned int if_label; + unsigned int target_label; +}; + +static bool lower_switch_to_if_ladder_add_block_mapping(struct lower_switch_to_if_ladder_block_mapping **block_map, + size_t *map_capacity, size_t *map_count, unsigned int switch_label, unsigned int if_label, unsigned int target_label) +{ + if (!vkd3d_array_reserve((void **)block_map, map_capacity, *map_count + 1, sizeof(**block_map))) + { + ERR("Failed to allocate block mapping.\n"); + return false; + } + + (*block_map)[*map_count].switch_label = switch_label; + (*block_map)[*map_count].if_label = if_label; + (*block_map)[*map_count].target_label = target_label; + + *map_count += 1; + + return true; +} + static enum vkd3d_result lower_switch_to_if_ladder(struct vkd3d_shader_parser *parser) { - unsigned int block_count = parser->program.block_count, ssa_count = parser->program.ssa_count; + unsigned int block_count = parser->program.block_count, ssa_count = parser->program.ssa_count, current_label = 0, if_label; + size_t ins_capacity = 0, ins_count = 0, i, map_capacity = 0, map_count = 0; struct vkd3d_shader_instruction *instructions = NULL; - size_t ins_capacity = 0, ins_count = 0, i; + struct lower_switch_to_if_ladder_block_mapping *block_map = NULL; if (!reserve_instructions(&instructions, &ins_capacity, parser->program.instructions.count)) goto fail; + /* First subpass: convert SWITCH_MONOLITHIC instructions to + * selection ladders, keeping a map between blocks before and + * after the subpass. */ for (i = 0; i < parser->program.instructions.count; ++i) { struct vkd3d_shader_instruction *ins = &parser->program.instructions.elements[i]; @@ -2404,15 +2436,16 @@ static enum vkd3d_result lower_switch_to_if_ladder(struct vkd3d_shader_parser *p switch (ins->handler_idx) { + case VKD3DSIH_LABEL: + current_label = label_from_src_param(&ins->src[0]); + if (!reserve_instructions(&instructions, &ins_capacity, ins_count + 1)) + goto fail; + instructions[ins_count++] = *ins; + continue; + case VKD3DSIH_SWITCH_MONOLITHIC: break; - case VKD3DSIH_PHI: - WARN("Unhandled PHI when lowering switch.\n"); - vkd3d_shader_parser_error(parser, VKD3D_SHADER_ERROR_VSIR_NOT_IMPLEMENTED, - "Unhandled PHI when lowering switch."); - return VKD3D_ERROR_NOT_IMPLEMENTED; - default: if (!reserve_instructions(&instructions, &ins_capacity, ins_count + 1)) goto fail; @@ -2439,6 +2472,8 @@ static enum vkd3d_result lower_switch_to_if_ladder(struct vkd3d_shader_parser *p if (!reserve_instructions(&instructions, &ins_capacity, ins_count + 3 * case_count - 1)) goto fail; + if_label = current_label; + for (j = 0; j < case_count; ++j) { unsigned int fallthrough_label, case_label = label_from_src_param(&ins->src[3 + 2 * j + 1]); @@ -2467,17 +2502,121 @@ static enum vkd3d_result lower_switch_to_if_ladder(struct vkd3d_shader_parser *p ++ssa_count; - if (j != case_count - 1) + if (!lower_switch_to_if_ladder_add_block_mapping(&block_map, &map_capacity, &map_count, + current_label, if_label, case_label)) + goto fail; + + if (j == case_count - 1) + { + if (!lower_switch_to_if_ladder_add_block_mapping(&block_map, &map_capacity, &map_count, + current_label, if_label, default_label)) + goto fail; + } + else { if (!vsir_instruction_init_with_params(parser, &instructions[ins_count], &ins->location, VKD3DSIH_LABEL, 0, 1)) goto fail; vsir_src_param_init_label(&instructions[ins_count].src[0], ++block_count); ++ins_count; + + if_label = block_count; } } } + /* Second subpass: creating new blocks might have broken + * references in PHI instructions, so we use the block map to fix + * them. */ + current_label = 0; + for (i = 0; i < ins_count; ++i) + { + struct vkd3d_shader_instruction *ins = &instructions[i]; + struct vkd3d_shader_src_param *new_src; + unsigned int j, l, new_src_count = 0; + + switch (ins->handler_idx) + { + case VKD3DSIH_LABEL: + current_label = label_from_src_param(&ins->src[0]); + continue; + + case VKD3DSIH_PHI: + break; + + default: + continue; + } + + /* First count how many source parameters we need. */ + for (j = 0; j < ins->src_count; j += 2) + { + unsigned int source_label = label_from_src_param(&ins->src[j + 1]); + size_t k, match_count = 0; + + for (k = 0; k < map_count; ++k) + { + struct lower_switch_to_if_ladder_block_mapping *mapping = &block_map[k]; + + if (mapping->switch_label == source_label && mapping->target_label == current_label) + match_count += 1; + } + + new_src_count += (match_count != 0) ? 2 * match_count : 2; + } + + assert(new_src_count >= ins->src_count); + + /* Allocate more source parameters if needed. */ + if (new_src_count == ins->src_count) + { + new_src = ins->src; + } + else + { + if (!(new_src = shader_parser_get_src_params(parser, new_src_count))) + { + ERR("Failed to allocate %u source parameters.\n", new_src_count); + goto fail; + } + } + + /* Then do the copy. */ + for (j = 0, l = 0; j < ins->src_count; j += 2) + { + unsigned int source_label = label_from_src_param(&ins->src[j + 1]); + size_t k, match_count = 0; + + for (k = 0; k < map_count; ++k) + { + struct lower_switch_to_if_ladder_block_mapping *mapping = &block_map[k]; + + if (mapping->switch_label == source_label && mapping->target_label == current_label) + { + match_count += 1; + + new_src[l] = ins->src[j]; + new_src[l + 1] = ins->src[j + 1]; + new_src[l + 1].reg.idx[0].offset = mapping->if_label; + l += 2; + } + } + + if (match_count == 0) + { + new_src[l] = ins->src[j]; + new_src[l + 1] = ins->src[j + 1]; + l += 2; + } + } + + assert(l == new_src_count); + + ins->src_count = new_src_count; + ins->src = new_src; + } + vkd3d_free(parser->program.instructions.elements); + vkd3d_free(block_map); parser->program.instructions.elements = instructions; parser->program.instructions.capacity = ins_capacity; parser->program.instructions.count = ins_count; @@ -2488,6 +2627,7 @@ static enum vkd3d_result lower_switch_to_if_ladder(struct vkd3d_shader_parser *p fail: vkd3d_free(instructions); + vkd3d_free(block_map); return VKD3D_ERROR_OUT_OF_MEMORY; }

1 0

Giovanni Mascellani : vkd3d-shader/ir: Lower monolithic switches to selection ladders.
by Alexandre Julliard Jan. 29, 2024

Jan. 29, 2024

Module: vkd3d Branch: master Commit: 378109051c601cebde705779169d0bfeb5bca051 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/378109051c601cebde705779169d0… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Wed Jan 3 16:18:50 2024 +0100 vkd3d-shader/ir: Lower monolithic switches to selection ladders. PHI nodes must be fixed up after this pass, because the block references might have become broken. For simplicitly this is not handled yet. The goal for this pass is to make the CFG structurizer simpler, because only conditional and unconditional branches must be supported. Eventually this limitation might be lifted if there is advantage in doing so. --- libs/vkd3d-shader/ir.c | 149 ++++++++++++++++++++++++++++++- libs/vkd3d-shader/spirv.c | 5 +- libs/vkd3d-shader/vkd3d_shader_private.h | 2 + 3 files changed, 153 insertions(+), 3 deletions(-) diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index 9079be48..c2568316 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -442,6 +442,15 @@ void vsir_src_param_init(struct vkd3d_shader_src_param *param, enum vkd3d_shader param->modifiers = VKD3DSPSM_NONE; } +void vsir_dst_param_init(struct vkd3d_shader_dst_param *param, enum vkd3d_shader_register_type reg_type, + enum vkd3d_data_type data_type, unsigned int idx_count) +{ + vsir_register_init(&param->reg, reg_type, data_type, idx_count); + param->write_mask = VKD3DSP_WRITEMASK_0; + param->modifiers = VKD3DSPDM_NONE; + param->shift = 0; +} + void vsir_src_param_init_label(struct vkd3d_shader_src_param *param, unsigned int label_id) { vsir_src_param_init(param, VKD3DSPR_LABEL, VKD3D_DATA_UINT, 1); @@ -449,6 +458,18 @@ void vsir_src_param_init_label(struct vkd3d_shader_src_param *param, unsigned in param->reg.idx[0].offset = label_id; } +static void src_param_init_ssa_bool(struct vkd3d_shader_src_param *src, unsigned int idx) +{ + vsir_src_param_init(src, VKD3DSPR_SSA, VKD3D_DATA_BOOL, 1); + src->reg.idx[0].offset = idx; +} + +static void dst_param_init_ssa_bool(struct vkd3d_shader_dst_param *dst, unsigned int idx) +{ + vsir_dst_param_init(dst, VKD3DSPR_SSA, VKD3D_DATA_BOOL, 1); + dst->reg.idx[0].offset = idx; +} + void vsir_instruction_init(struct vkd3d_shader_instruction *ins, const struct vkd3d_shader_location *location, enum vkd3d_shader_opcode handler_idx) { @@ -2350,6 +2371,127 @@ static enum vkd3d_result flatten_control_flow_constructs(struct vkd3d_shader_par return result; } +static unsigned int label_from_src_param(const struct vkd3d_shader_src_param *param) +{ + assert(param->reg.type == VKD3DSPR_LABEL); + return param->reg.idx[0].offset; +} + +static bool reserve_instructions(struct vkd3d_shader_instruction **instructions, size_t *capacity, size_t count) +{ + if (!vkd3d_array_reserve((void **)instructions, capacity, count, sizeof(**instructions))) + { + ERR("Failed to allocate instructions.\n"); + return false; + } + + return true; +} + +static enum vkd3d_result lower_switch_to_if_ladder(struct vkd3d_shader_parser *parser) +{ + unsigned int block_count = parser->program.block_count, ssa_count = parser->program.ssa_count; + struct vkd3d_shader_instruction *instructions = NULL; + size_t ins_capacity = 0, ins_count = 0, i; + + if (!reserve_instructions(&instructions, &ins_capacity, parser->program.instructions.count)) + goto fail; + + for (i = 0; i < parser->program.instructions.count; ++i) + { + struct vkd3d_shader_instruction *ins = &parser->program.instructions.elements[i]; + unsigned int case_count, j, default_label; + + switch (ins->handler_idx) + { + case VKD3DSIH_SWITCH_MONOLITHIC: + break; + + case VKD3DSIH_PHI: + WARN("Unhandled PHI when lowering switch.\n"); + vkd3d_shader_parser_error(parser, VKD3D_SHADER_ERROR_VSIR_NOT_IMPLEMENTED, + "Unhandled PHI when lowering switch."); + return VKD3D_ERROR_NOT_IMPLEMENTED; + + default: + if (!reserve_instructions(&instructions, &ins_capacity, ins_count + 1)) + goto fail; + instructions[ins_count++] = *ins; + continue; + } + + case_count = (ins->src_count - 3) / 2; + default_label = label_from_src_param(&ins->src[1]); + + /* In principle we can have a switch with no cases, and we + * just have to jump to the default label. */ + if (case_count == 0) + { + if (!reserve_instructions(&instructions, &ins_capacity, ins_count + 1)) + goto fail; + + if (!vsir_instruction_init_with_params(parser, &instructions[ins_count], &ins->location, VKD3DSIH_BRANCH, 0, 1)) + goto fail; + vsir_src_param_init_label(&instructions[ins_count].src[0], default_label); + ++ins_count; + } + + if (!reserve_instructions(&instructions, &ins_capacity, ins_count + 3 * case_count - 1)) + goto fail; + + for (j = 0; j < case_count; ++j) + { + unsigned int fallthrough_label, case_label = label_from_src_param(&ins->src[3 + 2 * j + 1]); + + if (!vsir_instruction_init_with_params(parser, &instructions[ins_count], &ins->location, VKD3DSIH_IEQ, 1, 2)) + goto fail; + dst_param_init_ssa_bool(&instructions[ins_count].dst[0], ssa_count); + instructions[ins_count].src[0] = ins->src[0]; + instructions[ins_count].src[1] = ins->src[3 + 2 * j]; + ++ins_count; + + /* For all cases except the last one we fall through to + * the following case; the last one has to jump to the + * default label. */ + if (j == case_count - 1) + fallthrough_label = default_label; + else + fallthrough_label = block_count + 1; + + if (!vsir_instruction_init_with_params(parser, &instructions[ins_count], &ins->location, VKD3DSIH_BRANCH, 0, 3)) + goto fail; + src_param_init_ssa_bool(&instructions[ins_count].src[0], ssa_count); + vsir_src_param_init_label(&instructions[ins_count].src[1], case_label); + vsir_src_param_init_label(&instructions[ins_count].src[2], fallthrough_label); + ++ins_count; + + ++ssa_count; + + if (j != case_count - 1) + { + if (!vsir_instruction_init_with_params(parser, &instructions[ins_count], &ins->location, VKD3DSIH_LABEL, 0, 1)) + goto fail; + vsir_src_param_init_label(&instructions[ins_count].src[0], ++block_count); + ++ins_count; + } + } + } + + vkd3d_free(parser->program.instructions.elements); + parser->program.instructions.elements = instructions; + parser->program.instructions.capacity = ins_capacity; + parser->program.instructions.count = ins_count; + parser->program.block_count = block_count; + parser->program.ssa_count = ssa_count; + + return VKD3D_OK; + +fail: + vkd3d_free(instructions); + + return VKD3D_ERROR_OUT_OF_MEMORY; +} + enum vkd3d_result vkd3d_shader_normalise(struct vkd3d_shader_parser *parser, const struct vkd3d_shader_compile_info *compile_info) { @@ -2361,7 +2503,12 @@ enum vkd3d_result vkd3d_shader_normalise(struct vkd3d_shader_parser *parser, if ((result = instruction_array_lower_texkills(parser)) < 0) return result; - if (!parser->shader_desc.is_dxil) + if (parser->shader_desc.is_dxil) + { + if ((result = lower_switch_to_if_ladder(parser)) < 0) + return result; + } + else { if (parser->program.shader_version.type != VKD3D_SHADER_TYPE_PIXEL) { diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 544ed9d7..91411bf6 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -7830,8 +7830,9 @@ static void spirv_compiler_emit_branch(struct spirv_compiler *compiler, } condition_id = spirv_compiler_emit_load_src(compiler, &src[0], VKD3DSP_WRITEMASK_0); - condition_id = spirv_compiler_emit_int_to_bool(compiler, - VKD3D_SHADER_CONDITIONAL_OP_NZ, src[0].reg.data_type, 1, condition_id); + if (src[0].reg.data_type != VKD3D_DATA_BOOL) + condition_id = spirv_compiler_emit_int_to_bool(compiler, + VKD3D_SHADER_CONDITIONAL_OP_NZ, src[0].reg.data_type, 1, condition_id); /* Emit the merge immediately before the branch instruction. */ if (instruction->src_count >= 4) spirv_compiler_emit_merge(compiler, src[3].reg.idx[0].offset, diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 28862884..f8c4d61e 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -895,6 +895,8 @@ struct vkd3d_shader_src_param void vsir_src_param_init(struct vkd3d_shader_src_param *param, enum vkd3d_shader_register_type reg_type, enum vkd3d_data_type data_type, unsigned int idx_count); +void vsir_dst_param_init(struct vkd3d_shader_dst_param *param, enum vkd3d_shader_register_type reg_type, + enum vkd3d_data_type data_type, unsigned int idx_count); void vsir_src_param_init_label(struct vkd3d_shader_src_param *param, unsigned int label_id); struct vkd3d_shader_index_range

1 0