mirror of
https://github.com/yuzu-emu/yuzu-android
synced 2024-12-26 06:51:21 -08:00
reuse vectors memory
This commit is contained in:
parent
b8ca47e094
commit
0078e5a338
@ -154,19 +154,10 @@ int color_bitsread = 0;
|
|||||||
uint color_values[32];
|
uint color_values[32];
|
||||||
int colvals_index = 0;
|
int colvals_index = 0;
|
||||||
|
|
||||||
// Weight data globals
|
|
||||||
uvec4 texel_weight_data;
|
|
||||||
int texel_bitsread = 0;
|
|
||||||
|
|
||||||
bool texel_flag = false;
|
|
||||||
|
|
||||||
// Global "vectors" to be pushed into when decoding
|
// Global "vectors" to be pushed into when decoding
|
||||||
EncodingData result_vector[144];
|
EncodingData result_vector[144];
|
||||||
int result_index = 0;
|
int result_index = 0;
|
||||||
|
|
||||||
EncodingData texel_vector[144];
|
|
||||||
int texel_vector_index = 0;
|
|
||||||
|
|
||||||
// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
|
// Replicates low num_bits such that [(to_bit - 1):(to_bit - 1 - from_bit)]
|
||||||
// is the same as [(num_bits - 1):0] and repeats all the way down.
|
// is the same as [(num_bits - 1):0] and repeats all the way down.
|
||||||
uint Replicate(uint val, uint num_bits, uint to_bit) {
|
uint Replicate(uint val, uint num_bits, uint to_bit) {
|
||||||
@ -382,27 +373,16 @@ void SkipBits(uint num_bits) {
|
|||||||
}
|
}
|
||||||
|
|
||||||
uint StreamColorBits(uint num_bits) {
|
uint StreamColorBits(uint num_bits) {
|
||||||
uint ret = 0;
|
const int int_bits = int(num_bits);
|
||||||
int int_bits = int(num_bits);
|
const uint ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits);
|
||||||
if (texel_flag) {
|
|
||||||
ret = ExtractBits(texel_weight_data, texel_bitsread, int_bits);
|
|
||||||
texel_bitsread += int_bits;
|
|
||||||
} else {
|
|
||||||
ret = ExtractBits(color_endpoint_data, color_bitsread, int_bits);
|
|
||||||
color_bitsread += int_bits;
|
color_bitsread += int_bits;
|
||||||
}
|
|
||||||
return ret;
|
return ret;
|
||||||
}
|
}
|
||||||
|
|
||||||
void ResultEmplaceBack(EncodingData val) {
|
void ResultEmplaceBack(EncodingData val) {
|
||||||
if (texel_flag) {
|
|
||||||
texel_vector[texel_vector_index] = val;
|
|
||||||
++texel_vector_index;
|
|
||||||
} else {
|
|
||||||
result_vector[result_index] = val;
|
result_vector[result_index] = val;
|
||||||
++result_index;
|
++result_index;
|
||||||
}
|
}
|
||||||
}
|
|
||||||
|
|
||||||
// Returns the number of bits required to encode n_vals values.
|
// Returns the number of bits required to encode n_vals values.
|
||||||
uint GetBitLength(uint n_vals, uint encoding_index) {
|
uint GetBitLength(uint n_vals, uint encoding_index) {
|
||||||
@ -910,7 +890,7 @@ void UnquantizeTexelWeights(bool is_dual_plane, uvec2 size, out uint unquantized
|
|||||||
const uint loop_count = min(result_index, area * num_planes);
|
const uint loop_count = min(result_index, area * num_planes);
|
||||||
uint unquantized[2 * 144];
|
uint unquantized[2 * 144];
|
||||||
for (uint itr = 0; itr < loop_count; ++itr) {
|
for (uint itr = 0; itr < loop_count; ++itr) {
|
||||||
unquantized[itr] = UnquantizeTexelWeight(texel_vector[itr]);
|
unquantized[itr] = UnquantizeTexelWeight(result_vector[itr]);
|
||||||
}
|
}
|
||||||
for (uint plane = 0; plane < num_planes; ++plane) {
|
for (uint plane = 0; plane < num_planes; ++plane) {
|
||||||
for (uint t = 0; t < block_dims.y; t++) {
|
for (uint t = 0; t < block_dims.y; t++) {
|
||||||
@ -1215,22 +1195,26 @@ void DecompressBlock(ivec3 coord) {
|
|||||||
ComputeEndpoints(endpoints[i][0], endpoints[i][1], color_endpoint_mode[i]);
|
ComputeEndpoints(endpoints[i][0], endpoints[i][1], color_endpoint_mode[i]);
|
||||||
}
|
}
|
||||||
|
|
||||||
texel_weight_data = local_buff;
|
color_endpoint_data = local_buff;
|
||||||
texel_weight_data = bitfieldReverse(texel_weight_data).wzyx;
|
color_endpoint_data = bitfieldReverse(color_endpoint_data).wzyx;
|
||||||
uint clear_byte_start =
|
uint clear_byte_start =
|
||||||
(GetPackedBitSize(params.size, params.dual_plane, params.max_weight) >> 3) + 1;
|
(GetPackedBitSize(params.size, params.dual_plane, params.max_weight) >> 3) + 1;
|
||||||
|
|
||||||
uint byte_insert = ExtractBits(texel_weight_data, int(clear_byte_start - 1) * 8, 8) &
|
uint byte_insert = ExtractBits(color_endpoint_data, int(clear_byte_start - 1) * 8, 8) &
|
||||||
uint(
|
uint(
|
||||||
((1 << (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) % 8)) - 1));
|
((1 << (GetPackedBitSize(params.size, params.dual_plane, params.max_weight) % 8)) - 1));
|
||||||
uint vec_index = (clear_byte_start - 1) >> 2;
|
uint vec_index = (clear_byte_start - 1) >> 2;
|
||||||
texel_weight_data[vec_index] =
|
color_endpoint_data[vec_index] =
|
||||||
bitfieldInsert(texel_weight_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8);
|
bitfieldInsert(color_endpoint_data[vec_index], byte_insert, int((clear_byte_start - 1) % 4) * 8, 8);
|
||||||
for (uint i = clear_byte_start; i < 16; ++i) {
|
for (uint i = clear_byte_start; i < 16; ++i) {
|
||||||
uint idx = i >> 2;
|
uint idx = i >> 2;
|
||||||
texel_weight_data[idx] = bitfieldInsert(texel_weight_data[idx], 0, int(i % 4) * 8, 8);
|
color_endpoint_data[idx] = bitfieldInsert(color_endpoint_data[idx], 0, int(i % 4) * 8, 8);
|
||||||
}
|
}
|
||||||
texel_flag = true; // use texel "vector" and bit stream in integer decoding
|
|
||||||
|
// Re-init vector variables for next decode phase
|
||||||
|
result_index = 0;
|
||||||
|
color_bitsread = 0;
|
||||||
|
|
||||||
DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
|
DecodeIntegerSequence(params.max_weight, GetNumWeightValues(params.size, params.dual_plane));
|
||||||
|
|
||||||
uint unquantized_texel_weights[2 * 144];
|
uint unquantized_texel_weights[2 * 144];
|
||||||
|
Loading…
Reference in New Issue
Block a user