Catcrafts/Crafter.Math
Watch
2
0
Fork
You've already forked Crafter.Math
 0
Code Issues 1 Pull requests Projects Releases Packages Wiki Activity Actions

AGENTS.md

Browse source
This commit is contained in:
Jorijn van der Graaf 2026-10-31 06:52:22 +01:00
commit b701059c8c
Signed by: jorijnvdgraaf
GPG key ID: 2937E59CDCC1BCFB
7 changed files with 331 additions and 62 deletions
Download patch file Download diff file Expand all files Collapse all files

23
interfaces/Crafter.Math-Common.cppm
View file

@ -186,10 +186,17 @@ namespace Crafter {
template <std::array<std::uint8_t, Len> ShuffleValues>
static consteval bool CheckEpi32Shuffle() {
if constexpr (PerLane == 8) {
for(std::uint8_t i = 1; i < Len; i+=2) {
if(ShuffleValues[i-1] != ShuffleValues[i] - 1) {
return false;
}
// 16-bit lanes move as 32-bit words: every output word has to
// take both halves of one source word, in order. The check
// used to accept any two consecutive sources, (1, 2) included,
// so Shuffle<{{1,2,0}}> on a VectorF16 took the epi32 path and
// came out wrong. An odd Len puts the second slot half a word
// off, so only Packing 1 can qualify then, and its last lane
// has to be a word's low half.
if (Len % 2 != 0 && Packing > 1) return false;
for (std::uint8_t i = 0; i < Len; i += 2) {
if (ShuffleValues[i] % 2 != 0) return false;
if (i + 1 < Len && ShuffleValues[i + 1] != ShuffleValues[i] + 1) return false;
}
}
for(std::uint8_t i = 0; i < Len; i++) {
@ -300,9 +307,13 @@ namespace Crafter {
template <std::array<std::uint8_t, Len> ShuffleValues>
static consteval std::uint8_t GetShuffleMaskEpi32() {
// One 2-bit field per 32-bit word, four words to the immediate; a
// word holds perWord elements, so a 16-bit lane index is halved
// (the old shift of 4 * i for halves ran off the 8-bit immediate).
constexpr std::uint8_t perWord = 4 / sizeof(T);
std::uint8_t mask = 0;
for(std::uint8_t i = 0; i < std::min(Len, std::uint8_t(8)); i+=4/sizeof(T)) {
mask = mask | (ShuffleValues[i] & 0b11) << (8 / sizeof(T) * i);
for (std::uint8_t i = 0; i < std::min<std::uint8_t>(Len, 4 * perWord); i += perWord) {
mask = mask | (((ShuffleValues[i] / perWord) & 0b11) << (2 * (i / perWord)));
}
return mask;
}