Pica/Shader: Add geometry shader definitions.

This commit is contained in:
Tony Wasserka 2015-03-22 00:31:40 +01:00 committed by Tony Wasserka
parent 7673cdfdae
commit 33568494af
6 changed files with 169 additions and 156 deletions

View file

@ -784,112 +784,119 @@ struct Regs {
INSERT_PADDING_WORDS(0x20);
enum class TriangleTopology : u32 {
List = 0,
Strip = 1,
Fan = 2,
ListIndexed = 3, // TODO: No idea if this is correct
List = 0,
Strip = 1,
Fan = 2,
Shader = 3, // Programmable setup unit implemented in a geometry shader
};
BitField<8, 2, TriangleTopology> triangle_topology;
INSERT_PADDING_WORDS(0x51);
INSERT_PADDING_WORDS(0x21);
BitField<0, 16, u32> vs_bool_uniforms;
union {
BitField< 0, 8, u32> x;
BitField< 8, 8, u32> y;
BitField<16, 8, u32> z;
BitField<24, 8, u32> w;
} vs_int_uniforms[4];
INSERT_PADDING_WORDS(0x5);
// Offset to shader program entry point (in words)
BitField<0, 16, u32> vs_main_offset;
union {
BitField< 0, 4, u64> attribute0_register;
BitField< 4, 4, u64> attribute1_register;
BitField< 8, 4, u64> attribute2_register;
BitField<12, 4, u64> attribute3_register;
BitField<16, 4, u64> attribute4_register;
BitField<20, 4, u64> attribute5_register;
BitField<24, 4, u64> attribute6_register;
BitField<28, 4, u64> attribute7_register;
BitField<32, 4, u64> attribute8_register;
BitField<36, 4, u64> attribute9_register;
BitField<40, 4, u64> attribute10_register;
BitField<44, 4, u64> attribute11_register;
BitField<48, 4, u64> attribute12_register;
BitField<52, 4, u64> attribute13_register;
BitField<56, 4, u64> attribute14_register;
BitField<60, 4, u64> attribute15_register;
int GetRegisterForAttribute(int attribute_index) const {
u64 fields[] = {
attribute0_register, attribute1_register, attribute2_register, attribute3_register,
attribute4_register, attribute5_register, attribute6_register, attribute7_register,
attribute8_register, attribute9_register, attribute10_register, attribute11_register,
attribute12_register, attribute13_register, attribute14_register, attribute15_register,
};
return (int)fields[attribute_index];
}
} vs_input_register_map;
INSERT_PADDING_WORDS(0x3);
struct {
enum Format : u32
{
FLOAT24 = 0,
FLOAT32 = 1
};
bool IsFloat32() const {
return format == FLOAT32;
}
struct ShaderConfig {
BitField<0, 16, u32> bool_uniforms;
union {
// Index of the next uniform to write to
// TODO: ctrulib uses 8 bits for this, however that seems to yield lots of invalid indices
BitField<0, 7, u32> index;
BitField< 0, 8, u32> x;
BitField< 8, 8, u32> y;
BitField<16, 8, u32> z;
BitField<24, 8, u32> w;
} int_uniforms[4];
BitField<31, 1, Format> format;
};
INSERT_PADDING_WORDS(0x5);
// Writing to these registers sets the "current" uniform.
// TODO: It's not clear how the hardware stores what the "current" uniform is.
u32 set_value[8];
// Offset to shader program entry point (in words)
BitField<0, 16, u32> main_offset;
} vs_uniform_setup;
union {
BitField< 0, 4, u64> attribute0_register;
BitField< 4, 4, u64> attribute1_register;
BitField< 8, 4, u64> attribute2_register;
BitField<12, 4, u64> attribute3_register;
BitField<16, 4, u64> attribute4_register;
BitField<20, 4, u64> attribute5_register;
BitField<24, 4, u64> attribute6_register;
BitField<28, 4, u64> attribute7_register;
BitField<32, 4, u64> attribute8_register;
BitField<36, 4, u64> attribute9_register;
BitField<40, 4, u64> attribute10_register;
BitField<44, 4, u64> attribute11_register;
BitField<48, 4, u64> attribute12_register;
BitField<52, 4, u64> attribute13_register;
BitField<56, 4, u64> attribute14_register;
BitField<60, 4, u64> attribute15_register;
INSERT_PADDING_WORDS(0x2);
int GetRegisterForAttribute(int attribute_index) const {
u64 fields[] = {
attribute0_register, attribute1_register, attribute2_register, attribute3_register,
attribute4_register, attribute5_register, attribute6_register, attribute7_register,
attribute8_register, attribute9_register, attribute10_register, attribute11_register,
attribute12_register, attribute13_register, attribute14_register, attribute15_register,
};
return (int)fields[attribute_index];
}
} input_register_map;
struct {
// Offset of the next instruction to write code to.
// Incremented with each instruction write.
u32 offset;
// OUTMAP_MASK, 0x28E, CODETRANSFER_END
INSERT_PADDING_WORDS(0x3);
// Writing to these registers sets the "current" word in the shader program.
// TODO: It's not clear how the hardware stores what the "current" word is.
u32 set_word[8];
} vs_program;
struct {
enum Format : u32
{
FLOAT24 = 0,
FLOAT32 = 1
};
INSERT_PADDING_WORDS(0x1);
bool IsFloat32() const {
return format == FLOAT32;
}
// This register group is used to load an internal table of swizzling patterns,
// which are indexed by each shader instruction to specify vector component swizzling.
struct {
// Offset of the next swizzle pattern to write code to.
// Incremented with each instruction write.
u32 offset;
union {
// Index of the next uniform to write to
// TODO: ctrulib uses 8 bits for this, however that seems to yield lots of invalid indices
// TODO: Maybe the uppermost index is for the geometry shader? Investigate!
BitField<0, 7, u32> index;
// Writing to these registers sets the "current" swizzle pattern in the table.
// TODO: It's not clear how the hardware stores what the "current" swizzle pattern is.
u32 set_word[8];
} vs_swizzle_patterns;
BitField<31, 1, Format> format;
};
INSERT_PADDING_WORDS(0x22);
// Writing to these registers sets the current uniform.
u32 set_value[8];
} uniform_setup;
INSERT_PADDING_WORDS(0x2);
struct {
// Offset of the next instruction to write code to.
// Incremented with each instruction write.
u32 offset;
// Writing to these registers sets the "current" word in the shader program.
u32 set_word[8];
} program;
INSERT_PADDING_WORDS(0x1);
// This register group is used to load an internal table of swizzling patterns,
// which are indexed by each shader instruction to specify vector component swizzling.
struct {
// Offset of the next swizzle pattern to write code to.
// Incremented with each instruction write.
u32 offset;
// Writing to these registers sets the current swizzle pattern in the table.
u32 set_word[8];
} swizzle_patterns;
INSERT_PADDING_WORDS(0x2);
};
ShaderConfig gs;
ShaderConfig vs;
INSERT_PADDING_WORDS(0x20);
// Map register indices to names readable by humans
// Used for debugging purposes, so performance is not an issue here
@ -936,13 +943,20 @@ struct Regs {
ADD_FIELD(vs_default_attributes_setup);
ADD_FIELD(command_buffer);
ADD_FIELD(triangle_topology);
ADD_FIELD(vs_bool_uniforms);
ADD_FIELD(vs_int_uniforms);
ADD_FIELD(vs_main_offset);
ADD_FIELD(vs_input_register_map);
ADD_FIELD(vs_uniform_setup);
ADD_FIELD(vs_program);
ADD_FIELD(vs_swizzle_patterns);
ADD_FIELD(gs.bool_uniforms);
ADD_FIELD(gs.int_uniforms);
ADD_FIELD(gs.main_offset);
ADD_FIELD(gs.input_register_map);
ADD_FIELD(gs.uniform_setup);
ADD_FIELD(gs.program);
ADD_FIELD(gs.swizzle_patterns);
ADD_FIELD(vs.bool_uniforms);
ADD_FIELD(vs.int_uniforms);
ADD_FIELD(vs.main_offset);
ADD_FIELD(vs.input_register_map);
ADD_FIELD(vs.uniform_setup);
ADD_FIELD(vs.program);
ADD_FIELD(vs.swizzle_patterns);
#undef ADD_FIELD
@ -1014,17 +1028,14 @@ ASSERT_REG_POSITION(trigger_draw_indexed, 0x22f);
ASSERT_REG_POSITION(vs_default_attributes_setup, 0x232);
ASSERT_REG_POSITION(command_buffer, 0x238);
ASSERT_REG_POSITION(triangle_topology, 0x25e);
ASSERT_REG_POSITION(vs_bool_uniforms, 0x2b0);
ASSERT_REG_POSITION(vs_int_uniforms, 0x2b1);
ASSERT_REG_POSITION(vs_main_offset, 0x2ba);
ASSERT_REG_POSITION(vs_input_register_map, 0x2bb);
ASSERT_REG_POSITION(vs_uniform_setup, 0x2c0);
ASSERT_REG_POSITION(vs_program, 0x2cb);
ASSERT_REG_POSITION(vs_swizzle_patterns, 0x2d5);
ASSERT_REG_POSITION(gs, 0x280);
ASSERT_REG_POSITION(vs, 0x2b0);
#undef ASSERT_REG_POSITION
#endif // !defined(_MSC_VER)
static_assert(sizeof(Regs::ShaderConfig) == 0x30 * sizeof(u32), "ShaderConfig structure has incorrect size");
// The total number of registers is chosen arbitrarily, but let's make sure it's not some odd value anyway.
static_assert(sizeof(Regs) <= 0x300 * sizeof(u32), "Register set structure larger than it should be");
static_assert(sizeof(Regs) >= 0x300 * sizeof(u32), "Register set structure smaller than it should be");
@ -1134,7 +1145,7 @@ struct State {
Regs regs;
/// Vertex shader memory
struct {
struct ShaderSetup {
struct {
Math::Vec4<float24> f[96];
std::array<bool, 16> b;
@ -1145,7 +1156,10 @@ struct State {
std::array<u32, 1024> program_code;
std::array<u32, 1024> swizzle_data;
} vs;
};
ShaderSetup vs;
ShaderSetup gs;
/// Current Pica command list
struct {