superrob/IQ.Pilot
1
0
Fork 0
forked from IQ.Lvbs/IQ.Pilot
Code Pull Requests Activity

IQ.Pilot Release Commit @ 0798119

Browse Source
This commit is contained in:
IQ.Lvbs history cleanup
2026-08-22 23:42:42 -05:00
commit b42569dbca
4529 changed files with 1132125 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

24
tinygrad_repo/extra/thunder/metal/include/ops/group/group.metal Normal file
View File

@@ -0,0 +1,24 @@
/**
* @file
* @brief An aggregate header of all group (multi-warp) operations defined by Thundermittens
*/
#pragma once
#include "../../common/common.metal"
#include "../../types/types.metal"
#include "../warp/warp.metal" // several group memory ops rely on underlying warp-scope ops
namespace mittens {
template<int N_WARPS>
struct group {
constant static constexpr int GROUP_WARPS = N_WARPS; // This alias produces nice parallelism.
constant static constexpr int GROUP_THREADS = N_WARPS * mittens::SIMD_THREADS; // This alias produces nice parallelism.
static METAL_FUNC int simd_laneid(const unsigned threadIdx) { return threadIdx % mittens::SIMD_THREADS; }
static METAL_FUNC int laneid (const unsigned threadIdx) { return threadIdx % GROUP_THREADS; }
static METAL_FUNC int warpid (const unsigned threadIdx) { return laneid(threadIdx) / mittens::SIMD_THREADS; }
static METAL_FUNC int groupid (const unsigned threadIdx) { return threadIdx / GROUP_THREADS; }
#include "memory/memory.metal"
#include "shared/shared.metal"
};
}