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

Giovanni Mascellani : vkd3d: Handle D3D12_FEATURE_D3D12_OPTIONS17 in CheckFeatureSupport().
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 7c7503713a0e898c77ff217e26d3de70a3f69f4d URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/7c7503713a0e898c77ff217e26d3d… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Wed May 8 11:25:22 2024 +0200 vkd3d: Handle D3D12_FEATURE_D3D12_OPTIONS17 in CheckFeatureSupport(). --- include/vkd3d_d3d12.idl | 7 +++++++ libs/vkd3d/device.c | 17 +++++++++++++++++ 2 files changed, 24 insertions(+) diff --git a/include/vkd3d_d3d12.idl b/include/vkd3d_d3d12.idl index 23f1567e..30776092 100644 --- a/include/vkd3d_d3d12.idl +++ b/include/vkd3d_d3d12.idl @@ -2466,6 +2466,12 @@ typedef struct D3D12_FEATURE_DATA_D3D12_OPTIONS16 BOOL GPUUploadHeapSupported; } D3D12_FEATURE_DATA_D3D12_OPTIONS16; +typedef struct D3D12_FEATURE_DATA_D3D12_OPTIONS17 +{ + BOOL NonNormalizedCoordinateSamplersSupported; + BOOL ManualWriteTrackingResourceSupported; +} D3D12_FEATURE_DATA_D3D12_OPTIONS17; + typedef enum D3D12_FEATURE { D3D12_FEATURE_D3D12_OPTIONS = 0, @@ -2504,6 +2510,7 @@ typedef enum D3D12_FEATURE D3D12_FEATURE_D3D12_OPTIONS14 = 43, D3D12_FEATURE_D3D12_OPTIONS15 = 44, D3D12_FEATURE_D3D12_OPTIONS16 = 45, + D3D12_FEATURE_D3D12_OPTIONS17 = 46, } D3D12_FEATURE; typedef struct D3D12_MEMCPY_DEST diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index 2a11105d..4c2e9d4a 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -3972,6 +3972,23 @@ static HRESULT STDMETHODCALLTYPE d3d12_device_CheckFeatureSupport(ID3D12Device9 return S_OK; } + case D3D12_FEATURE_D3D12_OPTIONS17: + { + D3D12_FEATURE_DATA_D3D12_OPTIONS17 *data = feature_data; + + if (feature_data_size != sizeof(*data)) + { + WARN("Invalid size %u.\n", feature_data_size); + } + + data->NonNormalizedCoordinateSamplersSupported = FALSE; + data->ManualWriteTrackingResourceSupported = FALSE; + + TRACE("Non-normalized coordinate samplers %#x.\n", data->NonNormalizedCoordinateSamplersSupported); + TRACE("Manual write tracking resource %#x.\n", data->ManualWriteTrackingResourceSupported); + return S_OK; + } + default: FIXME("Unhandled feature %#x.\n", feature); return E_NOTIMPL;

1 0

Giovanni Mascellani : vkd3d: Handle D3D12_FEATURE_D3D12_OPTIONS18 in CheckFeatureSupport().
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 05394b624a7921b451a906bc68a95c25a8a92ad8 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/05394b624a7921b451a906bc68a95… Author: Giovanni Mascellani <gmascellani(a)codeweavers.com> Date: Wed May 8 11:26:44 2024 +0200 vkd3d: Handle D3D12_FEATURE_D3D12_OPTIONS18 in CheckFeatureSupport(). --- include/vkd3d_d3d12.idl | 6 ++++++ libs/vkd3d/device.c | 15 +++++++++++++++ 2 files changed, 21 insertions(+) diff --git a/include/vkd3d_d3d12.idl b/include/vkd3d_d3d12.idl index 30776092..a72732e4 100644 --- a/include/vkd3d_d3d12.idl +++ b/include/vkd3d_d3d12.idl @@ -2472,6 +2472,11 @@ typedef struct D3D12_FEATURE_DATA_D3D12_OPTIONS17 BOOL ManualWriteTrackingResourceSupported; } D3D12_FEATURE_DATA_D3D12_OPTIONS17; +typedef struct D3D12_FEATURE_DATA_D3D12_OPTIONS18 +{ + BOOL RenderPassesValid; +} D3D12_FEATURE_DATA_D3D12_OPTIONS18; + typedef enum D3D12_FEATURE { D3D12_FEATURE_D3D12_OPTIONS = 0, @@ -2511,6 +2516,7 @@ typedef enum D3D12_FEATURE D3D12_FEATURE_D3D12_OPTIONS15 = 44, D3D12_FEATURE_D3D12_OPTIONS16 = 45, D3D12_FEATURE_D3D12_OPTIONS17 = 46, + D3D12_FEATURE_D3D12_OPTIONS18 = 47, } D3D12_FEATURE; typedef struct D3D12_MEMCPY_DEST diff --git a/libs/vkd3d/device.c b/libs/vkd3d/device.c index 4c2e9d4a..8b817264 100644 --- a/libs/vkd3d/device.c +++ b/libs/vkd3d/device.c @@ -3989,6 +3989,21 @@ static HRESULT STDMETHODCALLTYPE d3d12_device_CheckFeatureSupport(ID3D12Device9 return S_OK; } + case D3D12_FEATURE_D3D12_OPTIONS18: + { + D3D12_FEATURE_DATA_D3D12_OPTIONS18 *data = feature_data; + + if (feature_data_size != sizeof(*data)) + { + WARN("Invalid size %u.\n", feature_data_size); + } + + data->RenderPassesValid = FALSE; + + TRACE("Render passes valid %#x.\n", data->RenderPassesValid); + return S_OK; + } + default: FIXME("Unhandled feature %#x.\n", feature); return E_NOTIMPL;

1 0

Francisco Casas : tests: Test SM1 constant register allocation.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 7debe25a8b70e01caa249a1139b3d1be96ad7275 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/7debe25a8b70e01caa249a1139b3d… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Thu May 2 19:51:36 2024 -0400 tests: Test SM1 constant register allocation. --- Makefile.am | 1 + tests/hlsl/sm1-const-allocation.shader_test | 227 ++++++++++++++++++++++++++++ 2 files changed, 228 insertions(+) diff --git a/Makefile.am b/Makefile.am index 558938c5..b4f863fb 100644 --- a/Makefile.am +++ b/Makefile.am @@ -193,6 +193,7 @@ vkd3d_shader_tests = \ tests/hlsl/sign.shader_test \ tests/hlsl/single-numeric-initializer.shader_test \ tests/hlsl/sm1-const-folding.shader_test \ + tests/hlsl/sm1-const-allocation.shader_test \ tests/hlsl/sm6-ternary.shader_test \ tests/hlsl/sm6-uav-rwtexture.shader_test \ tests/hlsl/smoothstep.shader_test \ diff --git a/tests/hlsl/sm1-const-allocation.shader_test b/tests/hlsl/sm1-const-allocation.shader_test new file mode 100644 index 00000000..29625fa5 --- /dev/null +++ b/tests/hlsl/sm1-const-allocation.shader_test @@ -0,0 +1,227 @@ +[require] +shader model < 4.0 + + +[pixel shader] +// Only the used size is reserved. +// Note that these matrices are column-major. +float4x2 a; +float4x2 b; + +float4 main() : sv_target +{ + float4x2 c = a; + float4x2 d = c + b; + + return float4(d[0].x, d[1].x, d[2].x, d[3].x); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// a c0 1 +// b c1 1 +// +[test] +uniform 0 float4 1 2 3 4 +uniform 4 float4 100 200 300 400 +draw quad +todo probe all rgba (101, 202, 303, 404) + + +[pixel shader] +// Size 2 always are allocated before size 1 +float4x2 a; +float4x2 b; + +float4 main() : sv_target +{ + return float4(a[0].x, a[1].x, a[2].x, b[0].y); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// b c0 2 +// a c2 1 +// +[test] +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) + + +[pixel shader] +// Size 3 always allocated before size 2, probably means that they are allocated in decreasing used size. +float4x3 a; +float4x3 b; + +float4 main() : sv_target +{ + return float4(a[0].y, b[0].z, 0, 0); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// b c0 3 +// a c3 2 +// +[test] +uniform 0 float4 1 2 3 4 +uniform 4 float4 11 12 13 14 +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) + + +[pixel shader] +// Test same used size, how a comes before b. +float4x3 a; +float4x3 b; + +float4 main() : sv_target +{ + return float4(a[0].z, b[0].z, 0, 0); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// a c0 3 +// b c3 3 +// +[test] +uniform 0 float4 1 2 3 4 +uniform 4 float4 11 12 13 14 +uniform 8 float4 21 22 23 24 +uniform 12 float4 31 32 33 34 +uniform 16 float4 41 42 43 44 +uniform 20 float4 51 52 53 54 +draw quad +probe all rgba (21, 51, 0, 0) + + +[pixel shader] +// Declaration order is more important than instruction order to break ties. +float4x3 zzz; +float4x3 aaa; + +float4 main() : sv_target +{ + return float4(aaa[0].z, zzz[0].z, 0, 0); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// zzz c0 3 +// aaa c3 3 +// +[test] +uniform 0 float4 1 2 3 4 +uniform 4 float4 11 12 13 14 +uniform 8 float4 21 22 23 24 +uniform 12 float4 31 32 33 34 +uniform 16 float4 41 42 43 44 +uniform 20 float4 51 52 53 54 +draw quad +probe all rgba (51, 21, 0, 0) + + +[pixel shader] +// Arrays follow the same rules +float2 zzz[2], aaa[4]; + +float4 main() : sv_target +{ + return float4(aaa[1], zzz[1]); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// zzz c0 2 +// aaa c2 2 +// +[test] +uniform 0 float4 1 2 3 4 +uniform 4 float4 11 12 13 14 +uniform 8 float4 21 22 23 24 +uniform 12 float4 31 32 33 34 +draw quad +probe all rgba (31, 32, 11, 12) + + +[pixel shader] +struct apple +{ + float2 a; + float3x2 b; + float4 c[2]; +}; + +struct apple zzz, hhh, aaa; + +float4 main() : sv_target +{ + return float4(zzz.c[0].x, hhh.b[1].x, aaa.a); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// zzz c0 4 +// hhh c4 2 +// aaa c6 1 +// +[test] +uniform 0 float4 1 2 3 4 +uniform 4 float4 11 12 13 14 +uniform 8 float4 21 22 23 24 +uniform 12 float4 31 32 33 34 +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) + + +[pixel shader] +// Constant buffer variables are treated indistinctively. +float4x4 size3; +float4x4 unused; + +cbuffer buff +{ + float4x4 size1; + float4x4 size4; +}; + +float4 main() : sv_target +{ + return float4(size1[3].x, size4[0].w, size3[2].z, 0); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// size4 c0 4 +// size3 c4 3 +// size1 c7 1 + +[test] +uniform 0 float4 1 2 3 4 +uniform 4 float4 11 12 13 14 +uniform 8 float4 21 22 23 24 +uniform 12 float4 31 32 33 34 +uniform 16 float4 41 42 43 44 +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)

1 0

Nikolay Sivov : tests: Test SM1 constant register allocation with a row_major matrix.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 9dcb74341f0cffb2dafcba0fd189c03c7b83ca7b URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/9dcb74341f0cffb2dafcba0fd189c… Author: Nikolay Sivov <nsivov(a)codeweavers.com> Date: Thu May 2 20:01:19 2024 -0400 tests: Test SM1 constant register allocation with a row_major matrix. Co-authored-by: Francisco Casas <fcasas(a)codeweavers.com> --- tests/hlsl/sm1-const-allocation.shader_test | 29 +++++++++++++++++++++++++++++ 1 file changed, 29 insertions(+) diff --git a/tests/hlsl/sm1-const-allocation.shader_test b/tests/hlsl/sm1-const-allocation.shader_test index 29625fa5..741545b1 100644 --- a/tests/hlsl/sm1-const-allocation.shader_test +++ b/tests/hlsl/sm1-const-allocation.shader_test @@ -225,3 +225,32 @@ uniform 24 float4 61 62 63 64 uniform 28 float4 71 72 73 74 draw quad todo probe all rgba (74, 31, 63, 0) + + +[pixel shader] +typedef float2x2 matrix_t; +typedef row_major matrix_t row_matrix_t; +typedef column_major matrix_t col_matrix_t; +uniform row_matrix_t m1; +uniform col_matrix_t m2; +float4 main() : COLOR +{ + float4 ret; + ret.xy = m1[0]; + ret.zw = m2[0]; + return ret; +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// m2 c0 2 +// m1 c2 1 +// + +[test] +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)

1 0

Francisco Casas : tests: Test SM1 constant register allocation with reservations.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: b56ae0f1c85bd9bb0ac2591f4122049fe53420c1 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/b56ae0f1c85bd9bb0ac2591f41220… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri May 3 04:11:58 2024 -0400 tests: Test SM1 constant register allocation with reservations. --- tests/hlsl/sm1-const-allocation.shader_test | 119 ++++++++++++++++++++++++++++ 1 file changed, 119 insertions(+) diff --git a/tests/hlsl/sm1-const-allocation.shader_test b/tests/hlsl/sm1-const-allocation.shader_test index 741545b1..3f4da40a 100644 --- a/tests/hlsl/sm1-const-allocation.shader_test +++ b/tests/hlsl/sm1-const-allocation.shader_test @@ -254,3 +254,122 @@ uniform 4 float4 11 12 13 14 uniform 8 float4 21 22 23 24 draw quad todo probe all rgba (21, 22, 1, 11) + + +[pixel shader] +// If there is a register reservation, the whole variable size is reserved. +float a[4] : register(c1); +float b[4]; + +float4 main() : sv_target +{ + return float4(a[0], b[1], 0, 0); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// a c1 1 +// b c5 2 +// + +[test] +uniform 0 float 0 +uniform 4 float 1 +uniform 8 float 2 +uniform 12 float 3 +uniform 16 float 4 +uniform 20 float 5 +uniform 24 float 6 +draw quad +probe all rgba (1, 6, 0, 0) + + +[pixel shader] +// If there is a register reservation, the whole variable size is reserved. +float4x4 a : register(c1); +float b[4]; + +float4 main() : sv_target +{ + return float4(a[0].x, b[1], 0, 0); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// a c1 1 +// b c5 2 +// + +[test] +uniform 0 float 0 +uniform 4 float 1 +uniform 8 float 2 +uniform 12 float 3 +uniform 16 float 4 +uniform 20 float 5 +uniform 24 float 6 +draw quad +probe all rgba (1, 6, 0, 0) + + +[pixel shader] +// If there is a register reservation, the whole variable size is reserved, even if unused. +float4x4 unused : register(c1); +float b[4]; + +float4 main() : sv_target +{ + return float4(0, b[1], 0, 0); +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// b c5 2 +// + +[test] +uniform 0 float 0 +uniform 4 float 1 +uniform 8 float 2 +uniform 12 float 3 +uniform 16 float 4 +uniform 20 float 5 +uniform 24 float 6 +draw quad +todo probe all rgba (0, 6, 0, 0) + + +[pixel shader] +float unused[2] : register(c2); // will create a gap in c0. +float size3[3]; // cannot be placed in c0. +float size2[2]; // can be placed in c0. + +float4 main() : sv_target +{ + float4 res; + + res.x = size3[2]; + res.y = size2[1]; + res.z = 0; + res.w = 0; + return res; +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// size2 c0 2 +// size3 c4 3 +[test] +uniform 0 float 0 +uniform 4 float 1 +uniform 8 float 2 +uniform 12 float 3 +uniform 16 float 4 +uniform 20 float 5 +uniform 24 float 6 +draw quad +todo probe all rgba (6, 1, 0, 0)

1 0

Francisco Casas : tests: Test SM1 const register allocation with relative addressing.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 7b3644d9f28a9e25016a4fa8deb5df4162630d56 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/7b3644d9f28a9e25016a4fa8deb5d… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri May 3 04:15:09 2024 -0400 tests: Test SM1 const register allocation with relative addressing. --- tests/hlsl/sm1-const-allocation.shader_test | 37 +++++++++++++++++++++++++++++ 1 file changed, 37 insertions(+) diff --git a/tests/hlsl/sm1-const-allocation.shader_test b/tests/hlsl/sm1-const-allocation.shader_test index 3f4da40a..ca641b14 100644 --- a/tests/hlsl/sm1-const-allocation.shader_test +++ b/tests/hlsl/sm1-const-allocation.shader_test @@ -373,3 +373,40 @@ uniform 20 float 5 uniform 24 float 6 draw quad todo probe all rgba (6, 1, 0, 0) + + +[pixel shader todo] +// Relative addressing extends the allocation size only up to the array's size. +float idx; + +struct +{ + float dyn[5]; + float unused; +} a; + + +float4 main() : sv_target +{ + return a.dyn[idx]; +} +// Registers: +// +// Name Reg Size +// ------------ ----- ---- +// a c0 5 +// idx c5 1 +// + +[test] +uniform 0 float 0 +uniform 4 float 1 +uniform 8 float 2 +uniform 12 float 3 +uniform 16 float 4 +uniform 20 float 3 +todo draw quad +probe all rgba (3, 3, 3, 3) +uniform 20 float 1 +todo draw quad +probe all rgba (1, 1, 1, 1)

1 0

Francisco Casas : vkd3d-shader/hlsl: Allocate unused variables with register reservations on SM1.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: 657e460d11a08fabce2b412de0f88a4c34a201f8 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/657e460d11a08fabce2b412de0f88… Author: Francisco Casas <fcasas(a)codeweavers.com> Date: Fri May 3 19:32:52 2024 -0400 vkd3d-shader/hlsl: Allocate unused variables with register reservations on SM1. --- libs/vkd3d-shader/d3dbc.c | 6 +++--- libs/vkd3d-shader/hlsl_codegen.c | 2 +- tests/hlsl/sm1-const-allocation.shader_test | 4 ++-- 3 files changed, 6 insertions(+), 6 deletions(-) diff --git a/libs/vkd3d-shader/d3dbc.c b/libs/vkd3d-shader/d3dbc.c index 24a95224..c7991407 100644 --- a/libs/vkd3d-shader/d3dbc.c +++ b/libs/vkd3d-shader/d3dbc.c @@ -1713,7 +1713,7 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffe for (r = 0; r <= HLSL_REGSET_LAST; ++r) { - if (var->semantic.name || !var->regs[r].allocated) + if (var->semantic.name || !var->regs[r].allocated || !var->last_read) continue; ++uniform_count; @@ -1751,7 +1751,7 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffe for (r = 0; r <= HLSL_REGSET_LAST; ++r) { - if (var->semantic.name || !var->regs[r].allocated) + if (var->semantic.name || !var->regs[r].allocated || !var->last_read) continue; put_u32(buffer, 0); /* name */ @@ -1780,7 +1780,7 @@ static void write_sm1_uniforms(struct hlsl_ctx *ctx, struct vkd3d_bytecode_buffe { size_t var_offset, name_offset; - if (var->semantic.name || !var->regs[r].allocated) + if (var->semantic.name || !var->regs[r].allocated || !var->last_read) continue; var_offset = vars_start + (uniform_count * 5 * sizeof(uint32_t)); diff --git a/libs/vkd3d-shader/hlsl_codegen.c b/libs/vkd3d-shader/hlsl_codegen.c index 8882deaf..4af55293 100644 --- a/libs/vkd3d-shader/hlsl_codegen.c +++ b/libs/vkd3d-shader/hlsl_codegen.c @@ -4529,7 +4529,7 @@ static void allocate_const_registers(struct hlsl_ctx *ctx, struct hlsl_ir_functi { unsigned int reg_size = var->data_type->reg_size[HLSL_REGSET_NUMERIC]; - if (!var->is_uniform || !var->last_read || reg_size == 0) + if (!var->is_uniform || reg_size == 0) continue; if (var->reg_reservation.reg_type == 'c') diff --git a/tests/hlsl/sm1-const-allocation.shader_test b/tests/hlsl/sm1-const-allocation.shader_test index ca641b14..0a4340ea 100644 --- a/tests/hlsl/sm1-const-allocation.shader_test +++ b/tests/hlsl/sm1-const-allocation.shader_test @@ -339,7 +339,7 @@ uniform 16 float 4 uniform 20 float 5 uniform 24 float 6 draw quad -todo probe all rgba (0, 6, 0, 0) +probe all rgba (0, 6, 0, 0) [pixel shader] @@ -372,7 +372,7 @@ uniform 16 float 4 uniform 20 float 5 uniform 24 float 6 draw quad -todo probe all rgba (6, 1, 0, 0) +probe all rgba (6, 1, 0, 0) [pixel shader todo]

1 0

Conor McCarthy : vkd3d-shader/dxil: Implement DX intrinsic WaveAllBitCount.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: cb5163f9a428f05e80ab9bcf75a0b7320dae51be URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/cb5163f9a428f05e80ab9bcf75a0b… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Tue Apr 23 21:40:12 2024 +1000 vkd3d-shader/dxil: Implement DX intrinsic WaveAllBitCount. --- libs/vkd3d-shader/d3d_asm.c | 1 + libs/vkd3d-shader/dxil.c | 4 ++++ libs/vkd3d-shader/vkd3d_shader_private.h | 1 + 3 files changed, 6 insertions(+) diff --git a/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d-shader/d3d_asm.c index f2ad39f2..fbd26623 100644 --- a/libs/vkd3d-shader/d3d_asm.c +++ b/libs/vkd3d-shader/d3d_asm.c @@ -332,6 +332,7 @@ static const char * const shader_opcode_names[] = [VKD3DSIH_WAVE_ACTIVE_BIT_AND ] = "wave_active_bit_and", [VKD3DSIH_WAVE_ACTIVE_BIT_OR ] = "wave_active_bit_or", [VKD3DSIH_WAVE_ACTIVE_BIT_XOR ] = "wave_active_bit_xor", + [VKD3DSIH_WAVE_ALL_BIT_COUNT ] = "wave_all_bit_count", [VKD3DSIH_WAVE_ALL_TRUE ] = "wave_all_true", [VKD3DSIH_WAVE_ANY_TRUE ] = "wave_any_true", [VKD3DSIH_WAVE_OP_ADD ] = "wave_op_add", diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 29f73636..f1530109 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -457,6 +457,7 @@ enum dx_intrinsic_opcode DX_WAVE_PREFIX_OP = 121, DX_LEGACY_F32TOF16 = 130, DX_LEGACY_F16TOF32 = 131, + DX_WAVE_ALL_BIT_COUNT = 135, DX_RAW_BUFFER_LOAD = 139, DX_RAW_BUFFER_STORE = 140, }; @@ -4555,6 +4556,8 @@ static enum vkd3d_shader_opcode map_dx_unary_op(enum dx_intrinsic_opcode op) return VKD3DSIH_F16TOF32; case DX_WAVE_ACTIVE_ALL_EQUAL: return VKD3DSIH_WAVE_ACTIVE_ALL_EQUAL; + case DX_WAVE_ALL_BIT_COUNT: + return VKD3DSIH_WAVE_ALL_BIT_COUNT; case DX_WAVE_ALL_TRUE: return VKD3DSIH_WAVE_ALL_TRUE; case DX_WAVE_ANY_TRUE: @@ -6233,6 +6236,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_WAVE_ACTIVE_BALLOT ] = {"V", "1", sm6_parser_emit_dx_wave_active_ballot}, [DX_WAVE_ACTIVE_BIT ] = {"m", "Rc", sm6_parser_emit_dx_wave_active_bit}, [DX_WAVE_ACTIVE_OP ] = {"n", "Rcc", sm6_parser_emit_dx_wave_op}, + [DX_WAVE_ALL_BIT_COUNT ] = {"i", "1", sm6_parser_emit_dx_unary}, [DX_WAVE_ALL_TRUE ] = {"1", "1", sm6_parser_emit_dx_unary}, [DX_WAVE_ANY_TRUE ] = {"1", "1", sm6_parser_emit_dx_unary}, [DX_WAVE_GET_LANE_COUNT ] = {"i", "", sm6_parser_emit_dx_wave_builtin}, diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 3951ca70..8e3aae52 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -533,6 +533,7 @@ enum vkd3d_shader_opcode VKD3DSIH_WAVE_ACTIVE_BIT_AND, VKD3DSIH_WAVE_ACTIVE_BIT_OR, VKD3DSIH_WAVE_ACTIVE_BIT_XOR, + VKD3DSIH_WAVE_ALL_BIT_COUNT, VKD3DSIH_WAVE_ALL_TRUE, VKD3DSIH_WAVE_ANY_TRUE, VKD3DSIH_WAVE_OP_ADD,

1 0

Conor McCarthy : vkd3d-shader/spirv: Implement the WAVE_ALL_BIT_COUNT instruction.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: feef2577eff1c9f06166f9f4ece549f53f7581a0 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/feef2577eff1c9f06166f9f4ece54… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Tue Apr 23 21:52:12 2024 +1000 vkd3d-shader/spirv: Implement the WAVE_ALL_BIT_COUNT instruction. --- libs/vkd3d-shader/spirv.c | 44 ++++++++++++++++++++++++++++++++---- tests/hlsl/wave-ops-uint.shader_test | 2 +- 2 files changed, 41 insertions(+), 5 deletions(-) diff --git a/libs/vkd3d-shader/spirv.c b/libs/vkd3d-shader/spirv.c index 7ac9f3cd..fd2babc3 100644 --- a/libs/vkd3d-shader/spirv.c +++ b/libs/vkd3d-shader/spirv.c @@ -1760,6 +1760,14 @@ static uint32_t vkd3d_spirv_build_op_group_nonuniform_ballot(struct vkd3d_spirv_ result_type, vkd3d_spirv_get_op_scope_subgroup(builder), val_id); } +static uint32_t vkd3d_spirv_build_op_group_nonuniform_ballot_bit_count(struct vkd3d_spirv_builder *builder, + uint32_t result_type, SpvGroupOperation group_op, uint32_t val_id) +{ + vkd3d_spirv_enable_capability(builder, SpvCapabilityGroupNonUniformBallot); + return vkd3d_spirv_build_op_tr3(builder, &builder->function_stream, SpvOpGroupNonUniformBallotBitCount, + result_type, vkd3d_spirv_get_op_scope_subgroup(builder), group_op, val_id); +} + static uint32_t vkd3d_spirv_build_op_glsl_std450_tr1(struct vkd3d_spirv_builder *builder, enum GLSLstd450 op, uint32_t result_type, uint32_t operand) { @@ -9801,18 +9809,26 @@ static void spirv_compiler_emit_wave_bool_op(struct spirv_compiler *compiler, spirv_compiler_emit_store_dst(compiler, dst, val_id); } -static void spirv_compiler_emit_wave_active_ballot(struct spirv_compiler *compiler, - const struct vkd3d_shader_instruction *instruction) +static uint32_t spirv_compiler_emit_group_nonuniform_ballot(struct spirv_compiler *compiler, + const struct vkd3d_shader_src_param *src) { struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; - const struct vkd3d_shader_dst_param *dst = instruction->dst; - const struct vkd3d_shader_src_param *src = instruction->src; uint32_t type_id, val_id; type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_UINT, VKD3D_VEC4_SIZE); val_id = spirv_compiler_emit_load_src(compiler, src, VKD3DSP_WRITEMASK_0); val_id = vkd3d_spirv_build_op_group_nonuniform_ballot(builder, type_id, val_id); + return val_id; +} + +static void spirv_compiler_emit_wave_active_ballot(struct spirv_compiler *compiler, + const struct vkd3d_shader_instruction *instruction) +{ + const struct vkd3d_shader_dst_param *dst = instruction->dst; + uint32_t val_id; + + val_id = spirv_compiler_emit_group_nonuniform_ballot(compiler, instruction->src); spirv_compiler_emit_store_dst(compiler, dst, val_id); } @@ -9871,6 +9887,23 @@ static void spirv_compiler_emit_wave_alu_op(struct spirv_compiler *compiler, spirv_compiler_emit_store_dst(compiler, dst, val_id); } +static void spirv_compiler_emit_wave_bit_count(struct spirv_compiler *compiler, + const struct vkd3d_shader_instruction *instruction) +{ + struct vkd3d_spirv_builder *builder = &compiler->spirv_builder; + const struct vkd3d_shader_dst_param *dst = instruction->dst; + SpvGroupOperation group_op; + uint32_t type_id, val_id; + + group_op = SpvGroupOperationReduce; + + val_id = spirv_compiler_emit_group_nonuniform_ballot(compiler, instruction->src); + type_id = vkd3d_spirv_get_type_id(builder, VKD3D_SHADER_COMPONENT_UINT, 1); + val_id = vkd3d_spirv_build_op_group_nonuniform_ballot_bit_count(builder, type_id, group_op, val_id); + + spirv_compiler_emit_store_dst(compiler, dst, val_id); +} + /* This function is called after declarations are processed. */ static void spirv_compiler_emit_main_prolog(struct spirv_compiler *compiler) { @@ -10236,6 +10269,9 @@ static int spirv_compiler_handle_instruction(struct spirv_compiler *compiler, case VKD3DSIH_WAVE_OP_UMIN: spirv_compiler_emit_wave_alu_op(compiler, instruction); break; + case VKD3DSIH_WAVE_ALL_BIT_COUNT: + spirv_compiler_emit_wave_bit_count(compiler, instruction); + break; case VKD3DSIH_DCL: case VKD3DSIH_DCL_HS_MAX_TESSFACTOR: case VKD3DSIH_DCL_INPUT_CONTROL_POINT_COUNT: diff --git a/tests/hlsl/wave-ops-uint.shader_test b/tests/hlsl/wave-ops-uint.shader_test index c16adab4..2c817366 100644 --- a/tests/hlsl/wave-ops-uint.shader_test +++ b/tests/hlsl/wave-ops-uint.shader_test @@ -177,7 +177,7 @@ void main(uint id : SV_GroupIndex) } [test] -todo dispatch 4 1 1 +dispatch 4 1 1 probe uav 1 (0) rui (2) probe uav 1 (1) rui (2) probe uav 1 (2) rui (2)

1 0

Conor McCarthy : vkd3d-shader/dxil: Implement DX intrinsic WaveIsFirstLane.
by Alexandre Julliard 13 May '24

13 May '24

Module: vkd3d Branch: master Commit: c30976e5c22a9824700153816b57af2d9de2a044 URL: https://gitlab.winehq.org/wine/vkd3d/-/commit/c30976e5c22a9824700153816b57a… Author: Conor McCarthy <cmccarthy(a)codeweavers.com> Date: Tue Apr 23 21:53:53 2024 +1000 vkd3d-shader/dxil: Implement DX intrinsic WaveIsFirstLane. --- libs/vkd3d-shader/d3d_asm.c | 1 + libs/vkd3d-shader/dxil.c | 21 +++++++++++++++++++++ libs/vkd3d-shader/vkd3d_shader_private.h | 1 + 3 files changed, 23 insertions(+) diff --git a/libs/vkd3d-shader/d3d_asm.c b/libs/vkd3d-shader/d3d_asm.c index fbd26623..2114b270 100644 --- a/libs/vkd3d-shader/d3d_asm.c +++ b/libs/vkd3d-shader/d3d_asm.c @@ -335,6 +335,7 @@ static const char * const shader_opcode_names[] = [VKD3DSIH_WAVE_ALL_BIT_COUNT ] = "wave_all_bit_count", [VKD3DSIH_WAVE_ALL_TRUE ] = "wave_all_true", [VKD3DSIH_WAVE_ANY_TRUE ] = "wave_any_true", + [VKD3DSIH_WAVE_IS_FIRST_LANE ] = "wave_is_first_lane", [VKD3DSIH_WAVE_OP_ADD ] = "wave_op_add", [VKD3DSIH_WAVE_OP_IMAX ] = "wave_op_imax", [VKD3DSIH_WAVE_OP_IMIN ] = "wave_op_imin", diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index f1530109..dd88bce8 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -446,6 +446,7 @@ enum dx_intrinsic_opcode DX_STORE_PATCH_CONSTANT = 106, DX_OUTPUT_CONTROL_POINT_ID = 107, DX_PRIMITIVE_ID = 108, + DX_WAVE_IS_FIRST_LANE = 110, DX_WAVE_GET_LANE_INDEX = 111, DX_WAVE_GET_LANE_COUNT = 112, DX_WAVE_ANY_TRUE = 113, @@ -4490,6 +4491,25 @@ static bool sm6_parser_emit_coordinate_construct(struct sm6_parser *sm6, const s return sm6_parser_emit_reg_composite_construct(sm6, operand_regs, component_count, state, reg); } +static enum vkd3d_shader_opcode sm6_dx_map_void_op(enum dx_intrinsic_opcode op) +{ + switch (op) + { + case DX_WAVE_IS_FIRST_LANE: + return VKD3DSIH_WAVE_IS_FIRST_LANE; + default: + vkd3d_unreachable(); + } +} + +static void sm6_parser_emit_dx_void(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; + vsir_instruction_init(ins, &sm6->p.location, sm6_dx_map_void_op(op)); + instruction_dst_param_init_ssa_scalar(ins, sm6); +} + static enum vkd3d_shader_opcode map_dx_unary_op(enum dx_intrinsic_opcode op) { switch (op) @@ -6241,6 +6261,7 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_WAVE_ANY_TRUE ] = {"1", "1", sm6_parser_emit_dx_unary}, [DX_WAVE_GET_LANE_COUNT ] = {"i", "", sm6_parser_emit_dx_wave_builtin}, [DX_WAVE_GET_LANE_INDEX ] = {"i", "", sm6_parser_emit_dx_wave_builtin}, + [DX_WAVE_IS_FIRST_LANE ] = {"1", "", sm6_parser_emit_dx_void}, [DX_WAVE_PREFIX_OP ] = {"n", "Rcc", sm6_parser_emit_dx_wave_op}, }; diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index 8e3aae52..1d4cbde8 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -536,6 +536,7 @@ enum vkd3d_shader_opcode VKD3DSIH_WAVE_ALL_BIT_COUNT, VKD3DSIH_WAVE_ALL_TRUE, VKD3DSIH_WAVE_ANY_TRUE, + VKD3DSIH_WAVE_IS_FIRST_LANE, VKD3DSIH_WAVE_OP_ADD, VKD3DSIH_WAVE_OP_IMAX, VKD3DSIH_WAVE_OP_IMIN,

1 0