From: Conor McCarthy cmccarthy@codeweavers.com
--- libs/vkd3d-shader/dxil.c | 69 +++++++++++++++++++++++++++++++ tests/hlsl/numthreads.shader_test | 10 +++-- 2 files changed, 75 insertions(+), 4 deletions(-)
diff --git a/libs/vkd3d-shader/dxil.c b/libs/vkd3d-shader/dxil.c index 4794a399f..3d540d269 100644 --- a/libs/vkd3d-shader/dxil.c +++ b/libs/vkd3d-shader/dxil.c @@ -398,6 +398,10 @@ enum dx_intrinsic_opcode DX_DERIV_COARSEY = 84, DX_DERIV_FINEX = 85, DX_DERIV_FINEY = 86, + DX_THREAD_ID = 93, + DX_GROUP_ID = 94, + DX_THREAD_ID_IN_GROUP = 95, + DX_FLATTENED_THREAD_ID_IN_GROUP = 96, DX_SPLIT_DOUBLE = 102, DX_LEGACY_F32TOF16 = 130, DX_LEGACY_F16TOF32 = 131, @@ -751,6 +755,7 @@ struct sm6_parser
struct vkd3d_shader_dst_param *output_params; struct vkd3d_shader_dst_param *input_params; + uint64_t input_reg_declarations;
struct sm6_function *functions; size_t function_count; @@ -4227,6 +4232,66 @@ static void sm6_parser_emit_dx_create_handle(struct sm6_parser *sm6, enum dx_int ins->handler_idx = VKD3DSIH_NOP; }
+static void sm6_parser_dcl_register_builtin(struct sm6_parser *sm6, + enum vkd3d_shader_register_type reg_type, enum vkd3d_data_type data_type, unsigned int component_count) +{ + struct vkd3d_shader_dst_param *dst_param; + struct vkd3d_shader_instruction *ins; + uint64_t reg_flag; + + reg_flag = 1ull << reg_type; + if (!(sm6->input_reg_declarations & reg_flag)) + { + sm6->input_reg_declarations |= reg_flag; + ins = sm6_parser_add_instruction(sm6, VKD3DSIH_DCL_INPUT); + dst_param = &ins->declaration.dst; + vsir_register_init(&dst_param->reg, reg_type, data_type, 0); + dst_param_init_vector(dst_param, component_count); + } +} + +static void sm6_parser_emit_dx_compute_builtin(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, + const struct sm6_value **operands, struct function_emission_state *state) +{ + unsigned int component_count = 3, component_idx = 0; + struct vkd3d_shader_instruction *ins = state->ins; + struct vkd3d_shader_src_param *src_param; + enum vkd3d_shader_register_type reg_type; + + switch (op) + { + case DX_THREAD_ID: + reg_type = VKD3DSPR_THREADID; + break; + case DX_GROUP_ID: + reg_type = VKD3DSPR_THREADGROUPID; + break; + case DX_THREAD_ID_IN_GROUP: + reg_type = VKD3DSPR_LOCALTHREADID; + break; + case DX_FLATTENED_THREAD_ID_IN_GROUP: + reg_type = VKD3DSPR_LOCALTHREADINDEX; + component_count = 1; + break; + default: + vkd3d_unreachable(); + } + + sm6_parser_dcl_register_builtin(sm6, reg_type, VKD3D_DATA_UINT, component_count); + vsir_instruction_init(ins, &sm6->p.location, VKD3DSIH_MOV); + if (!(src_param = instruction_src_params_alloc(ins, 1, sm6))) + return; + vsir_register_init(&src_param->reg, reg_type, VKD3D_DATA_UINT, 0); + if (component_count > 1) + { + src_param->reg.dimension = VSIR_DIMENSION_VEC4; + component_idx = sm6_value_get_constant_uint(operands[0]); + } + src_param_init_scalar(src_param, component_idx); + + instruction_dst_param_init_ssa_scalar(ins, sm6); +} + static void sm6_parser_emit_dx_get_dimensions(struct sm6_parser *sm6, enum dx_intrinsic_opcode op, const struct sm6_value **operands, struct function_emission_state *state) { @@ -4981,10 +5046,12 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_FIRST_BIT_HI ] = {"i", "m", sm6_parser_emit_dx_unary}, [DX_FIRST_BIT_LO ] = {"i", "m", sm6_parser_emit_dx_unary}, [DX_FIRST_BIT_SHI ] = {"i", "m", sm6_parser_emit_dx_unary}, + [DX_FLATTENED_THREAD_ID_IN_GROUP ] = {"i", "", sm6_parser_emit_dx_compute_builtin}, [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_GET_DIMENSIONS ] = {"D", "Hi", sm6_parser_emit_dx_get_dimensions}, + [DX_GROUP_ID ] = {"i", "c", sm6_parser_emit_dx_compute_builtin}, [DX_IBFE ] = {"m", "iiR", sm6_parser_emit_dx_tertiary}, [DX_HCOS ] = {"g", "R", sm6_parser_emit_dx_unary}, [DX_HSIN ] = {"g", "R", sm6_parser_emit_dx_unary}, @@ -5020,6 +5087,8 @@ static const struct sm6_dx_opcode_info sm6_dx_op_table[] = [DX_TEXTURE_GATHER_CMP ] = {"o", "HHffffiicf", sm6_parser_emit_dx_texture_gather}, [DX_TEXTURE_LOAD ] = {"o", "HiiiiCCC", sm6_parser_emit_dx_texture_load}, [DX_TEXTURE_STORE ] = {"v", "Hiiiooooc", sm6_parser_emit_dx_texture_store}, + [DX_THREAD_ID ] = {"i", "c", sm6_parser_emit_dx_compute_builtin}, + [DX_THREAD_ID_IN_GROUP ] = {"i", "c", sm6_parser_emit_dx_compute_builtin}, [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}, diff --git a/tests/hlsl/numthreads.shader_test b/tests/hlsl/numthreads.shader_test index e8f7900bf..ceeb40e3e 100644 --- a/tests/hlsl/numthreads.shader_test +++ b/tests/hlsl/numthreads.shader_test @@ -172,7 +172,7 @@ size (2d, 2, 2) 1.0 1.0
[compute shader] -/* Attributes are taken from the first function, and dropped from the second. */ +/* In SM < 6.0, attributes are taken from the first function, and dropped from the second. */ RWTexture2D<float> u;
[numthreads(2, 1, 1)] @@ -185,8 +185,10 @@ void main(uint2 id : sv_dispatchthreadid) }
[test] -todo(sm>=6) dispatch 1 1 1 +dispatch 1 1 1 probe uav 0 (0, 0) r (2.0) -probe uav 0 (0, 1) r (1.0) -probe uav 0 (1, 0) r (2.0) +if(sm<6) probe uav 0 (0, 1) r (1.0) +if(sm<6) probe uav 0 (1, 0) r (2.0) probe uav 0 (1, 1) r (1.0) +if(sm>=6) probe uav 0 (0, 1) r (2.0) +if(sm>=6) probe uav 0 (1, 0) r (1.0)