Searched refs:kNumWarps (Results 1 – 8 of 8) sorted by relevance
29 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in blockSelect() local31 __shared__ K smemK[kNumWarps * NumWarpQ]; in blockSelect()32 __shared__ IndexType smemV[kNumWarps * NumWarpQ]; in blockSelect()86 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in blockSelectPair() local88 __shared__ K smemK[kNumWarps * NumWarpQ]; in blockSelectPair()89 __shared__ IndexType smemV[kNumWarps * NumWarpQ]; in blockSelectPair()
29 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in warpSelect() local42 int row = blockIdx.x * kNumWarps + warpId; in warpSelect()
148 static constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; member302 FinalBlockMerge<kNumWarps, ThreadsPerBlock, K, V, NumWarpQ, Dir, Comp>:: in reduce()346 static constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; member397 for (int i = 1; i < kNumWarps; ++i) { in reduce()597 static constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; member
32 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in pass1SelectLists() local34 __shared__ float smemK[kNumWarps * NumWarpQ]; in pass1SelectLists()35 __shared__ int smemV[kNumWarps * NumWarpQ]; in pass1SelectLists()
63 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in pass2SelectLists() local65 __shared__ float smemK[kNumWarps * NumWarpQ]; in pass2SelectLists()66 __shared__ int smemV[kNumWarps * NumWarpQ]; in pass2SelectLists()
31 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in ivfInterleavedScan2() local33 __shared__ float smemK[kNumWarps * NumWarpQ]; in ivfInterleavedScan2()34 __shared__ uint32_t smemV[kNumWarps * NumWarpQ]; in ivfInterleavedScan2()
137 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in l2SelectMinK() local139 __shared__ T smemK[kNumWarps * NumWarpQ]; in l2SelectMinK()140 __shared__ int smemV[kNumWarps * NumWarpQ]; in l2SelectMinK()
54 constexpr int kNumWarps = ThreadsPerBlock / kWarpSize; in ivfInterleavedScan() local81 __shared__ float smemK[kNumWarps * NumWarpQ]; in ivfInterleavedScan()82 __shared__ int smemV[kNumWarps * NumWarpQ]; in ivfInterleavedScan()110 for (int block = warpId; block < numBlocks; block += kNumWarps) { in ivfInterleavedScan()