daniel/yuzu-android
1
0
Fork 0
mirror of https://github.com/yuzu-emu/yuzu-android synced 2024-12-26 08:41:22 -08:00
Code Issues Actions Packages Projects Releases Wiki Activity

astc: Implement a fast precompiled alternative for Replicate

Browse Source
This commit is contained in:
ReinUsesLisp 2020-04-09 03:58:25 -03:00
parent 6b4d4473be
commit 5de130beea
1 changed files with 57 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

59
src/video_core/textures/astc.cpp
View File

@ -680,6 +680,61 @@ static constexpr u32 ReplicateBitTo9(std::size_t value) {
return REPLICATE_BIT_TO_9_TABLE[value]; return REPLICATE_BIT_TO_9_TABLE[value];
} }
static constexpr auto REPLICATE_1_BIT_TO_8_TABLE = MakeReplicateTable<u32, 1, 8>();
static constexpr auto REPLICATE_2_BIT_TO_8_TABLE = MakeReplicateTable<u32, 2, 8>();
static constexpr auto REPLICATE_3_BIT_TO_8_TABLE = MakeReplicateTable<u32, 3, 8>();
static constexpr auto REPLICATE_4_BIT_TO_8_TABLE = MakeReplicateTable<u32, 4, 8>();
static constexpr auto REPLICATE_5_BIT_TO_8_TABLE = MakeReplicateTable<u32, 5, 8>();
static constexpr auto REPLICATE_6_BIT_TO_8_TABLE = MakeReplicateTable<u32, 6, 8>();
static constexpr auto REPLICATE_7_BIT_TO_8_TABLE = MakeReplicateTable<u32, 7, 8>();
static constexpr auto REPLICATE_8_BIT_TO_8_TABLE = MakeReplicateTable<u32, 8, 8>();
/// Use a precompiled table with the most common usages, if it's not in the expected range, fallback
/// to the runtime implementation
static constexpr u32 FastReplicateTo8(u32 value, u32 num_bits) {
switch (num_bits) {
case 1:
return REPLICATE_1_BIT_TO_8_TABLE[value];
case 2:
return REPLICATE_2_BIT_TO_8_TABLE[value];
case 3:
return REPLICATE_3_BIT_TO_8_TABLE[value];
case 4:
return REPLICATE_4_BIT_TO_8_TABLE[value];
case 5:
return REPLICATE_5_BIT_TO_8_TABLE[value];
case 6:
return REPLICATE_6_BIT_TO_8_TABLE[value];
case 7:
return REPLICATE_7_BIT_TO_8_TABLE[value];
case 8:
return REPLICATE_8_BIT_TO_8_TABLE[value];
default:
return Replicate(value, num_bits, 8);
}
}
static constexpr auto REPLICATE_1_BIT_TO_6_TABLE = MakeReplicateTable<u32, 1, 6>();
static constexpr auto REPLICATE_2_BIT_TO_6_TABLE = MakeReplicateTable<u32, 2, 6>();
static constexpr auto REPLICATE_3_BIT_TO_6_TABLE = MakeReplicateTable<u32, 3, 6>();
static constexpr auto REPLICATE_4_BIT_TO_6_TABLE = MakeReplicateTable<u32, 4, 6>();
static constexpr auto REPLICATE_5_BIT_TO_6_TABLE = MakeReplicateTable<u32, 5, 6>();
static constexpr u32 FastReplicateTo6(u32 value, u32 num_bits) {
switch (num_bits) {
case 1:
return REPLICATE_1_BIT_TO_6_TABLE[value];
case 2:
return REPLICATE_2_BIT_TO_6_TABLE[value];
case 3:
return REPLICATE_3_BIT_TO_6_TABLE[value];
case 4:
return REPLICATE_4_BIT_TO_6_TABLE[value];
case 5:
return REPLICATE_5_BIT_TO_6_TABLE[value];
default:
return Replicate(value, num_bits, 6);
}
}
class Pixel { class Pixel {
protected: protected:
using ChannelType = s16; using ChannelType = s16;
@ -868,7 +923,7 @@ static void DecodeColorValues(u32* out, u8* data, const u32* modes, const u32 nP
switch (val.encoding) { switch (val.encoding) {
// Replicate bits // Replicate bits
case IntegerEncoding::JustBits: case IntegerEncoding::JustBits:
out[outIdx++] = Replicate(bitval, bitlen, 8); out[outIdx++] = FastReplicateTo8(bitval, bitlen);
break; break;
// Use algorithm in C.2.13 // Use algorithm in C.2.13
@ -992,7 +1047,7 @@ static u32 UnquantizeTexelWeight(const IntegerEncodedValue& val) {
u32 result = 0; u32 result = 0;
switch (val.encoding) { switch (val.encoding) {
case IntegerEncoding::JustBits: case IntegerEncoding::JustBits:
result = Replicate(bitval, bitlen, 6); result = FastReplicateTo6(bitval, bitlen);
break; break;
case IntegerEncoding::Trit: { case IntegerEncoding::Trit: {