From: Conor McCarthy cmccarthy@codeweavers.com
With some changes by Giovanni Mascellani. --- libs/vkd3d-shader/ir.c | 56 +++++++++++++++++++ libs/vkd3d-shader/vkd3d_shader_private.h | 25 +++++++++ .../hlsl/arithmetic-float-uniform.shader_test | 2 +- 3 files changed, 82 insertions(+), 1 deletion(-)
diff --git a/libs/vkd3d-shader/ir.c b/libs/vkd3d-shader/ir.c index e4cc9dcbd..7c00d3c03 100644 --- a/libs/vkd3d-shader/ir.c +++ b/libs/vkd3d-shader/ir.c @@ -17,6 +17,7 @@ */
#include "vkd3d_shader_private.h" +#include "vkd3d_types.h"
bool vsir_program_init(struct vsir_program *program, const struct vkd3d_shader_version *version, unsigned int reserve) { @@ -159,6 +160,56 @@ static enum vkd3d_result vsir_program_lower_texkill(struct vsir_program *program return VKD3D_OK; }
+/* The Shader Model 5 Assembly documentation states: "If components of a mad + * instruction are tagged as precise, the hardware must execute a mad instruction + * or the exact equivalent, and it cannot split it into a multiply followed by an add." + * But DXIL.rst states the opposite: "Floating point multiply & add. This operation is + * not fused for "precise" operations." + * Windows drivers seem to conform with the latter, for SM 4-5 and SM 6. */ +static enum vkd3d_result vsir_program_lower_precise_mad(struct vsir_program *program, + struct vkd3d_shader_instruction *mad, unsigned int *tmp_idx) +{ + struct vkd3d_shader_instruction_array *instructions = &program->instructions; + struct vkd3d_shader_instruction *mul_ins, *add_ins; + size_t pos = mad - instructions->elements; + struct vkd3d_shader_dst_param *mul_dst; + + if (!(mad->flags & VKD3DSI_PRECISE_XYZW)) + return VKD3D_OK; + + if (!shader_instruction_array_insert_at(instructions, pos + 1, 1)) + return VKD3D_ERROR_OUT_OF_MEMORY; + + if (*tmp_idx == ~0u) + *tmp_idx = program->temp_count++; + + mul_ins = &instructions->elements[pos]; + add_ins = &instructions->elements[pos + 1]; + + mul_ins->handler_idx = VKD3DSIH_MUL; + mul_ins->src_count = 2; + + if (!(vsir_instruction_init_with_params(program, add_ins, &mul_ins->location, VKD3DSIH_ADD, 1, 2))) + return VKD3D_ERROR_OUT_OF_MEMORY; + + add_ins->flags = mul_ins->flags & VKD3DSI_PRECISE_XYZW; + + mul_dst = mul_ins->dst; + *add_ins->dst = *mul_dst; + + mul_dst->modifiers = 0; + vsir_register_init(&mul_dst->reg, VKD3DSPR_TEMP, mul_ins->src[0].reg.data_type, 1); + mul_dst->reg.dimension = add_ins->dst->reg.dimension; + mul_dst->reg.idx[0].offset = *tmp_idx; + + add_ins->src[0].reg = mul_dst->reg; + add_ins->src[0].swizzle = vsir_swizzle_from_writemask(mul_dst->write_mask); + add_ins->src[0].modifiers = 0; + add_ins->src[1] = mul_ins->src[2]; + + return VKD3D_OK; +} + static enum vkd3d_result vsir_program_lower_instructions(struct vsir_program *program) { struct vkd3d_shader_instruction_array *instructions = &program->instructions; @@ -176,6 +227,11 @@ static enum vkd3d_result vsir_program_lower_instructions(struct vsir_program *pr return ret; break;
+ case VKD3DSIH_MAD: + if ((ret = vsir_program_lower_precise_mad(program, ins, &tmp_idx)) < 0) + return ret; + break; + case VKD3DSIH_DCL_TEMPS: vkd3d_shader_instruction_make_nop(ins); break; diff --git a/libs/vkd3d-shader/vkd3d_shader_private.h b/libs/vkd3d-shader/vkd3d_shader_private.h index b07a7bff7..f401fd7a3 100644 --- a/libs/vkd3d-shader/vkd3d_shader_private.h +++ b/libs/vkd3d-shader/vkd3d_shader_private.h @@ -1761,6 +1761,31 @@ static inline unsigned int vkd3d_compact_swizzle(uint32_t swizzle, uint32_t writ return compacted_swizzle; }
+static inline uint32_t vsir_swizzle_from_writemask(unsigned int writemask) +{ + static const unsigned int swizzles[16] = + { + 0, + VKD3D_SHADER_SWIZZLE(X, X, X, X), + VKD3D_SHADER_SWIZZLE(Y, Y, Y, Y), + VKD3D_SHADER_SWIZZLE(X, Y, X, X), + VKD3D_SHADER_SWIZZLE(Z, Z, Z, Z), + VKD3D_SHADER_SWIZZLE(X, Z, X, X), + VKD3D_SHADER_SWIZZLE(Y, Z, X, X), + VKD3D_SHADER_SWIZZLE(X, Y, Z, X), + VKD3D_SHADER_SWIZZLE(W, W, W, W), + VKD3D_SHADER_SWIZZLE(X, W, X, X), + VKD3D_SHADER_SWIZZLE(Y, W, X, X), + VKD3D_SHADER_SWIZZLE(X, Y, W, X), + VKD3D_SHADER_SWIZZLE(Z, W, X, X), + VKD3D_SHADER_SWIZZLE(X, Z, W, X), + VKD3D_SHADER_SWIZZLE(Y, Z, W, X), + VKD3D_SHADER_SWIZZLE(X, Y, Z, W), + }; + + return swizzles[writemask & 0xf]; +} + struct vkd3d_struct { enum vkd3d_shader_structure_type type; diff --git a/tests/hlsl/arithmetic-float-uniform.shader_test b/tests/hlsl/arithmetic-float-uniform.shader_test index 61957f2bb..a49d3f00a 100644 --- a/tests/hlsl/arithmetic-float-uniform.shader_test +++ b/tests/hlsl/arithmetic-float-uniform.shader_test @@ -121,7 +121,7 @@ uniform 0 float4 1.00000007 -42.1 4.0 45.0 uniform 4 float4 1.625 -5.0 4.125 5.0 uniform 8 float4 1.00000007 -1.0 0.5 -0.5 todo(sm<6) draw quad -todo probe all rgba (2.62500048, 209.5, 17.0, 224.5) +probe all rgba (2.62500048, 209.5, 17.0, 224.5)
[require] shader model >= 5.0