mirror of
https://github.com/NVIDIA/TensorRT-LLM.git
synced 2026-01-14 06:27:45 +08:00
16d2467ea8
* Update TensorRT-LLM --------- Co-authored-by: Denis Kayshev <topenkoff@gmail.com> Co-authored-by: akhoroshev <arthoroshev@gmail.com> Co-authored-by: Patrick Reiter Horn <patrick.horn@gmail.com> Update
2291 lines
118 KiB
C++
2291 lines
118 KiB
C++
/*
|
|
* Copyright (c) 2022-2024, NVIDIA CORPORATION. All rights reserved.
|
|
*
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
* you may not use this file except in compliance with the License.
|
|
* You may obtain a copy of the License at
|
|
*
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
|
*
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
* See the License for the specific language governing permissions and
|
|
* limitations under the License.
|
|
*/
|
|
|
|
#pragma once
|
|
|
|
#include <cuda.h>
|
|
#include <cuda_fp8.h>
|
|
#include <mma.h>
|
|
|
|
#include "tensorrt_llm/common/cudaBf16Fallbacks.cuh"
|
|
#include "tensorrt_llm/common/cudaDriverWrapper.h"
|
|
|
|
#include "Common.h"
|
|
#include "CudaType.h"
|
|
#include "Poly.h"
|
|
|
|
namespace tensorrt_llm
|
|
{
|
|
namespace kernels
|
|
{
|
|
|
|
typedef void (*ChunkScanKernelFunc)(int B_, int L_, int H_, int P_, int G_, int N_,
|
|
void* g_mxY_, // Tp_ B*L*H*P
|
|
void const* g_mxOs_, // Tp_ B*C*H*N*P
|
|
// const void *g_mxFs_, // Tp_ B *H*N*P
|
|
// const void *g_mxSt_, // float B*C*H*N*P
|
|
void const* g_mxdc_, // float B*C*H*Q
|
|
void const* g_mxdA_, // float B*C*H*Q
|
|
// const void *g_mxdt_, // Tp_ B*L*((g_mxZ?2:1)*H*P+2*G+round_up(H,8))
|
|
// const void *g_mxdb_, // Wt_ H
|
|
// const void *g_mxA_, // Wt_ H
|
|
void const* g_mxCB_, // Tp_ B*C*G*Q*Q
|
|
void const* g_mxD_, // Wt_ H
|
|
void const* g_mxX_, // Tp_ B*L*(H*P+2*G*N)
|
|
void const* g_mxZ_, // g_mxdt_ or nullptr
|
|
bool removePadding_, int const* lastTokenIdsPtr_);
|
|
|
|
template <int Q_, int tileM_, int tileN_, int tileK_, // smem size, per sm
|
|
int warpM_, int warpN_, // warp number
|
|
int pipeS_, int pipeR_, int group_, int bnChk_, class Tp_, class Wt_>
|
|
__global__ std::enable_if_t<std::is_same_v<Tp_, half> || std::is_same_v<Tp_, __nv_bfloat16>> chunk_scan_kernel(int B_,
|
|
int L_, int H_, int P_, int G_, int N_,
|
|
void* g_mxY_, // Tp_ B*L*H*P
|
|
void const* g_mxOs_, // Tp_ B*C*H*N*P
|
|
// const void *g_mxFs_, // Tp_ B *H*N*P
|
|
// const void *g_mxSt_, // float B*C*H*N*P
|
|
void const* g_mxdc_, // float B*C*H*Q
|
|
void const* g_mxdA_, // float B*C*H*Q
|
|
// const void *g_mxdt_, // Tp_ B*L*((g_mxZ?2:1)*H*P+2*G+round_up(H,8))
|
|
// const void *g_mxdb_, // Wt_ H
|
|
// const void *g_mxA_, // Wt_ H
|
|
void const* g_mxCB_, // Tp_ B*C*G*Q*Q
|
|
void const* g_mxD_, // Wt_ H
|
|
void const* g_mxX_, // Tp_ B*L*(H*P+2*G*N)
|
|
void const* g_mxZ_, // g_mxdt_ or nullptr
|
|
bool removePadding_, int const* lastTokenIdsPtr_)
|
|
{
|
|
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 800
|
|
using namespace tensorrt_llm::common;
|
|
|
|
constexpr int wmmaM_ = 16; // wmma size, per instruction
|
|
constexpr int wmmaN_ = 8;
|
|
constexpr int wmmaK_ = 16;
|
|
|
|
auto blockIdx_x = Rn<ID>{int(blockIdx.x)};
|
|
auto blockIdx_y = Rn<ID>{int(blockIdx.y)};
|
|
auto blockIdx_z = Rn<ID>{int(blockIdx.z)};
|
|
|
|
auto threadIdx_x = Rn<ID, 32>{int(threadIdx.x)};
|
|
auto threadIdx_y = Rn<ID, warpN_>{int(threadIdx.y)};
|
|
auto threadIdx_z = Rn<ID, warpM_>{int(threadIdx.z)};
|
|
|
|
// auto B = Rn<ID>{B_};
|
|
auto L = Rn<ID>{L_};
|
|
auto H = Rn<ID>{H_};
|
|
auto P = Rn<ID>{P_};
|
|
auto G = Rn<ID>{G_};
|
|
auto N = Rn<ID>{N_};
|
|
auto Q = cn<Q_>;
|
|
auto C = Rn<ID>{div_up(L.var, Q_)};
|
|
|
|
auto X_stride = Rn<ID>{H_ * P_ + 2 * G_ * N_};
|
|
auto Z_stride = Rn<ID>{(g_mxZ_ ? 2 : 1) * H_ * P_ + 2 * G_ * N_ + round_up(H_, 8)};
|
|
|
|
auto aStart = blockIdx_z * L;
|
|
auto cStart = blockIdx_z * C;
|
|
|
|
if (removePadding_)
|
|
{
|
|
aStart = Rn<ID>{int(blockIdx.z ? lastTokenIdsPtr_[blockIdx.z - 1] : 0)};
|
|
cStart = Rn<ID>{int(blockIdx.z ? div_up(aStart.var, Q_) + blockIdx.z - 1 : 0)};
|
|
L = Rn<ID>{lastTokenIdsPtr_[blockIdx.z] - aStart.var};
|
|
C = Rn<ID>{div_up(L.var, Q_)};
|
|
}
|
|
else
|
|
{
|
|
L = Rn<ID>{lastTokenIdsPtr_[blockIdx.z]};
|
|
C = Rn<ID>{div_up(L.var, Q_)};
|
|
}
|
|
|
|
int numBlocks_m = H_ * (Q_ / tileM_);
|
|
int numBlocks_n = div_up(P_, tileN_);
|
|
|
|
int numBlocks_group = numBlocks_n * group_;
|
|
int blockIdx_0 = blockIdx.x / numBlocks_group * group_;
|
|
int group_m = std::min(group_, numBlocks_m - blockIdx_0);
|
|
|
|
int mStartInt = blockIdx.x % numBlocks_group % group_m + blockIdx_0;
|
|
int nStartInt = blockIdx.x % numBlocks_group / group_m;
|
|
|
|
auto hStart = Rn<ID>{mStartInt} / (Q / cn<tileM_>);
|
|
auto mStart = Rn<ID>{mStartInt} % (Q / cn<tileM_>);
|
|
auto nStart = Rn<ID>{nStartInt};
|
|
auto gStart = Rn<ID>{hStart.var / (H_ / G_)};
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> >= L)
|
|
return;
|
|
|
|
Tp_* g_mxY = (Tp_*) g_mxY_ + int64_t(get(aStart + blockIdx_y * Q)) * get(H * P);
|
|
Tp_ const* g_mxOs = (Tp_ const*) g_mxOs_ + int64_t(get(cStart + blockIdx_y)) * get(H * N * P);
|
|
// const Tp_ *g_mxFs = (const Tp_ *)g_mxFs_;
|
|
// const float *g_mxSt = (const float *)g_mxSt_;
|
|
float const* g_mxdc = (float const*) g_mxdc_ + int64_t(get(cStart + blockIdx_y)) * get(H * Q);
|
|
float const* g_mxdA = (float const*) g_mxdA_ + int64_t(get(cStart + blockIdx_y)) * get(H * Q);
|
|
// const Tp_ *g_mxdt = (const Tp_ *)g_mxdt_;
|
|
// const Wt_ *g_mxdb = (const Wt_ *)g_mxdb_;
|
|
// const Wt_ *g_mxA = (const Wt_ *)g_mxA_;
|
|
Tp_ const* g_mxCB = (Tp_ const*) g_mxCB_ + int64_t(get(cStart + blockIdx_y)) * get(G * Q * Q);
|
|
Wt_ const* g_mxD = (Wt_ const*) g_mxD_;
|
|
Tp_ const* g_mxX = (Tp_ const*) g_mxX_ + int64_t(get(aStart + blockIdx_y * Q)) * get(X_stride);
|
|
Tp_ const* g_mxZ = (Tp_ const*) g_mxZ_ + (g_mxZ_ ? int64_t(get(aStart + blockIdx_y * Q)) * get(Z_stride) : 0);
|
|
|
|
extern __shared__ float smem[];
|
|
|
|
Tp_* s_mxL = (Tp_*) smem;
|
|
Tp_* s_mxR = (Tp_*) smem + tileM_ * tileK_ * pipeS_;
|
|
Tp_* s_mxAcc = (Tp_*) smem;
|
|
|
|
float* s_mxdc = smem + (tileM_ + tileN_) * tileK_ * pipeS_ / 2;
|
|
float* s_mxdA = smem + (tileM_ + tileN_) * tileK_ * pipeS_ / 2 + Q_;
|
|
|
|
unsigned b_base = __nvvm_get_smem_pointer(smem);
|
|
|
|
unsigned b_mxL = b_base;
|
|
unsigned b_mxR = b_base + tileM_ * tileK_ * pipeS_ * sizeof(Tp_);
|
|
unsigned b_mxAcc = b_base;
|
|
|
|
using std::array;
|
|
|
|
array<array<array<float, wmmaM_ * wmmaN_ / 32>, tileN_ / wmmaN_ / warpN_>, tileM_ / wmmaM_ / warpM_> r_mxAcc
|
|
= array<array<array<float, wmmaM_ * wmmaN_ / 32>, tileN_ / wmmaN_ / warpN_>, tileM_ / wmmaM_ / warpM_>();
|
|
array<array<array<unsigned, wmmaM_ * wmmaK_ / 64>, tileM_ / wmmaM_ / warpM_>, pipeR_> r_mxL;
|
|
array<array<array<unsigned, wmmaK_ * wmmaN_ / 64>, tileN_ / wmmaN_ / warpN_>, pipeR_> r_mxR;
|
|
|
|
if (pipeR_ == 2)
|
|
{
|
|
r_mxL[1] = array<array<unsigned, wmmaM_ * wmmaK_ / 64>, tileM_ / wmmaM_ / warpM_>();
|
|
r_mxR[1] = array<array<unsigned, wmmaK_ * wmmaN_ / 64>, tileN_ / wmmaN_ / warpN_>();
|
|
}
|
|
|
|
constexpr int step = std::max(
|
|
1, tileM_ / wmmaM_ / warpM_ * tileN_ / wmmaN_ / warpN_ / (tileM_ / wmmaM_ / warpM_ + tileN_ / wmmaN_ / warpN_));
|
|
|
|
auto baseL = [](auto iK) { return iK % cn<pipeS_> * cn<tileM_> * cn<tileK_>; };
|
|
auto baseR = [](auto iK) { return iK % cn<pipeS_> * cn<tileN_> * cn<tileK_>; };
|
|
|
|
auto thread = [=](auto iStep)
|
|
{
|
|
return iStep * cn<warpM_ * warpN_ * 256> + threadIdx_z * cn<warpN_ * 256> + threadIdx_y * cn<256>
|
|
+ threadIdx_x * cn<8>;
|
|
};
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(Q_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<Q_>)
|
|
{
|
|
#pragma unroll
|
|
for (int i = 0; i < 8; i += 4)
|
|
{
|
|
// dc should be set to zero out of seq length, but this is done
|
|
// by chunkcumsum, so no need here.
|
|
*(int4*) (s_mxdc + get(thread(iStep)) + i) = *(int4*) (g_mxdc + get(hStart * Q + thread(iStep)) + i);
|
|
*(int4*) (s_mxdA + get(thread(iStep)) + i) = *(int4*) (g_mxdA + get(hStart * Q + thread(iStep)) + i);
|
|
}
|
|
}
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, pipeS_> iK; iK.var < iK.size; iK.var++)
|
|
{
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileM_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileM_ * tileK_>
|
|
&& thread(iStep) / cn<tileK_> < L - blockIdx_y * Q - mStart * cn<tileM_>)
|
|
cp_shared_global<16>(b_mxL + swizzle<tileK_ * 2, tileK_ * 2>(thread(iStep) * cn<2>, baseL(iK) * cn<2>),
|
|
g_mxX
|
|
+ get((mStart * cn<tileM_> + thread(iStep) / cn<tileK_>) *X_stride + H * P + cn<1> * G * N
|
|
+ gStart * N + iK * cn<tileK_> + thread(iStep) % cn<tileK_>));
|
|
else if (thread(iStep) < cn<tileM_ * tileK_>)
|
|
*(int4*) ((char*) s_mxL + swizzle<tileK_ * 2, tileK_ * 2>(thread(iStep) * cn<2>, baseL(iK) * cn<2>))
|
|
= int4{0, 0, 0, 0};
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileN_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileN_ * tileK_>
|
|
&& (!bnChk_ || nStart * cn<tileN_> + thread(iStep) % cn<tileN_> < P))
|
|
cp_shared_global<16>(b_mxR + swizzle<tileN_ * 2, tileN_ * 2>(thread(iStep) * cn<2>, baseR(iK) * cn<2>),
|
|
g_mxOs
|
|
+ get(hStart * N * P + (iK * cn<tileK_> + thread(iStep) / cn<tileN_>) *P + nStart * cn<tileN_>
|
|
+ thread(iStep) % cn<tileN_>));
|
|
|
|
cp_commit_group();
|
|
}
|
|
|
|
cp_wait_group<pipeS_ - 1>();
|
|
|
|
__syncthreads();
|
|
|
|
int Q1 = std::min(Q_, (get(mStart) + 1) * tileM_);
|
|
|
|
for (int iK = pipeS_; iK < (N_ + Q1) / tileK_ + pipeS_; iK++)
|
|
{
|
|
auto jK = Rn<>{iK};
|
|
if ((iK - pipeS_) * cn<tileK_> == N_)
|
|
{
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
if (pipeR_ == 2)
|
|
{
|
|
if (wmmaK_ == 16)
|
|
mma<Tp_>(r_mxAcc[y][x], r_mxL[1][y], r_mxR[1][x]);
|
|
}
|
|
|
|
float2 tmp2 = float2{expf(s_mxdA[get(mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>)]),
|
|
expf(s_mxdA[get(mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<8>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>)])};
|
|
|
|
r_mxAcc[y][x][0] *= tmp2.x;
|
|
r_mxAcc[y][x][1] *= tmp2.x;
|
|
r_mxAcc[y][x][2] *= tmp2.y;
|
|
r_mxAcc[y][x][3] *= tmp2.y;
|
|
}
|
|
|
|
if (pipeR_ == 2)
|
|
{
|
|
r_mxL[1] = array<array<unsigned, wmmaM_ * wmmaK_ / 64>, tileM_ / wmmaM_ / warpM_>();
|
|
r_mxR[1] = array<array<unsigned, wmmaK_ * wmmaN_ / 64>, tileN_ / wmmaN_ / warpN_>();
|
|
}
|
|
}
|
|
|
|
if ((iK - pipeS_) * cn<tileK_> >= N_)
|
|
{
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileM_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileM_ * tileK_>)
|
|
{
|
|
Tp_ tmpL[8];
|
|
|
|
*(int4*) &tmpL[0] = *(int4*) ((char*) s_mxL
|
|
+ swizzle<tileK_ * 2, tileK_ * 2>(thread(iStep) * cn<2>, baseL(jK) * cn<2>));
|
|
|
|
#pragma unroll
|
|
for (int i = 0; i < 8; i += 2)
|
|
{
|
|
float2 tmp2 = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmpL[i])
|
|
: bf1622float2(*(bf162*) &tmpL[i]);
|
|
|
|
int kStart = (iK - pipeS_) * cn<tileK_> - N_;
|
|
|
|
// dc should be set to zero out of seq length, but this is done
|
|
// by chunkcumsum, so no need here.
|
|
tmp2.x *= expf(s_mxdA[get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>)]
|
|
- s_mxdA[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i})])
|
|
* s_mxdc[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i})];
|
|
tmp2.y *= expf(s_mxdA[get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>)]
|
|
- s_mxdA[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i + 1})])
|
|
* s_mxdc[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i + 1})];
|
|
|
|
if (get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>)
|
|
< kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i}))
|
|
tmp2.x = 0;
|
|
if (get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>)
|
|
< kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i + 1}))
|
|
tmp2.y = 0;
|
|
|
|
if (std::is_same_v<Tp_, half>)
|
|
*(half2*) &tmpL[i] = __float22half2_rn(tmp2);
|
|
else
|
|
*(bf162*) &tmpL[i] = __float22bfloat162_rn(tmp2);
|
|
}
|
|
|
|
*(int4*) ((char*) s_mxL + swizzle<tileK_ * 2, tileK_ * 2>(thread(iStep) * cn<2>, baseL(jK) * cn<2>))
|
|
= *(int4*) &tmpL[0];
|
|
}
|
|
|
|
__syncthreads();
|
|
}
|
|
|
|
#pragma unroll
|
|
for (int k = 0; k < tileK_ / wmmaK_; k++)
|
|
{
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
if ((y * tileN_ / wmmaN_ / warpN_ + x) % step == 0)
|
|
{
|
|
int x1 = (y * tileN_ / wmmaN_ / warpN_ + x) / step;
|
|
int y1 = x1 - tileN_ / wmmaN_ / warpN_
|
|
+ (tileM_ / wmmaM_ / warpM_ == 1 || tileN_ / wmmaN_ / warpN_ == 1);
|
|
|
|
if (y1 >= 0 && y1 < tileM_ / wmmaM_ / warpM_)
|
|
{
|
|
if (wmmaK_ == 16)
|
|
ldmatrix_x4<>(r_mxL[k % pipeR_][y1][0], r_mxL[k % pipeR_][y1][1],
|
|
r_mxL[k % pipeR_][y1][2], r_mxL[k % pipeR_][y1][3],
|
|
b_mxL + iK % pipeS_ * (tileM_ * tileK_ * 2)
|
|
+ 2
|
|
* swz<tileK_ * 2, tileK_>(y1 * warpM_ * wmmaM_ * tileK_ + k * wmmaK_
|
|
+ threadIdx.z * wmmaM_ * tileK_ + threadIdx.x % 16 * tileK_
|
|
+ threadIdx.x / 16 * 8));
|
|
}
|
|
|
|
if (x1 >= 0 && x1 < tileN_ / wmmaN_ / warpN_)
|
|
{
|
|
if (wmmaK_ == 16 && x1 % 2 == 0)
|
|
ldmatrix_x4</* trans_ = */ true>(r_mxR[k % pipeR_][x1][0], r_mxR[k % pipeR_][x1][1],
|
|
r_mxR[k % pipeR_][x1 + 1][0], r_mxR[k % pipeR_][x1 + 1][1],
|
|
b_mxR + iK % pipeS_ * (tileK_ * tileN_ * 2)
|
|
+ 2
|
|
* swz<tileN_ * 2, tileN_>(x1 * warpN_ * wmmaN_ + k * wmmaK_ * tileN_
|
|
+ threadIdx.y * wmmaN_ + threadIdx.x % wmmaK_ * tileN_
|
|
+ threadIdx.x / wmmaK_ * warpN_ * wmmaN_));
|
|
}
|
|
}
|
|
|
|
if (pipeR_ == 2)
|
|
{
|
|
if (wmmaK_ == 16)
|
|
mma<Tp_>(r_mxAcc[y][x], r_mxL[(k + 1) % pipeR_][y], r_mxR[(k + 1) % pipeR_][x]);
|
|
}
|
|
}
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
if (pipeR_ == 1)
|
|
{
|
|
if (wmmaK_ == 16)
|
|
mma<Tp_>(r_mxAcc[y][x], r_mxL[(k + 1) % pipeR_][y], r_mxR[(k + 1) % pipeR_][x]);
|
|
}
|
|
}
|
|
}
|
|
|
|
__syncthreads();
|
|
|
|
if (iK * tileK_ < N_)
|
|
{
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileM_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileM_ * tileK_>
|
|
&& thread(iStep) / cn<tileK_> < L - blockIdx_y * Q - mStart * cn<tileM_>)
|
|
cp_shared_global<16>(
|
|
b_mxL + swizzle<tileK_ * 2, tileK_ * 2>(thread(iStep) * cn<2>, baseL(jK) * cn<2>),
|
|
g_mxX
|
|
+ get((mStart * cn<tileM_> + thread(iStep) / cn<tileK_>) *X_stride + H * P + cn<1> * G * N
|
|
+ gStart * N + jK * cn<tileK_> + thread(iStep) % cn<tileK_>));
|
|
else if (thread(iStep) < cn<tileM_ * tileK_>)
|
|
*(int4*) ((char*) s_mxL + swizzle<tileK_ * 2, tileK_ * 2>(thread(iStep) * cn<2>, baseL(jK) * cn<2>))
|
|
= int4{0, 0, 0, 0};
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileN_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileN_ * tileK_>
|
|
&& (!bnChk_ || nStart * cn<tileN_> + thread(iStep) % cn<tileN_> < P))
|
|
cp_shared_global<16>(
|
|
b_mxR + swizzle<tileN_ * 2, tileN_ * 2>(thread(iStep) * cn<2>, baseR(jK) * cn<2>),
|
|
g_mxOs
|
|
+ get(hStart * N * P + (jK * cn<tileK_> + thread(iStep) / cn<tileN_>) *P
|
|
+ nStart * cn<tileN_> + thread(iStep) % cn<tileN_>));
|
|
}
|
|
else if (iK * tileK_ < N_ + Q1)
|
|
{
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileM_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileM_ * tileK_>)
|
|
cp_shared_global<16>(
|
|
b_mxL + swizzle<tileK_ * 2, tileK_ * 2>(thread(iStep) * cn<2>, baseL(jK) * cn<2>),
|
|
g_mxCB
|
|
+ get(gStart * Q * Q + (mStart * cn<tileM_> + thread(iStep) / cn<tileK_>) *Q
|
|
+ jK * cn<tileK_> - N + thread(iStep) % cn<tileK_>));
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileN_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileN_ * tileK_>
|
|
&& thread(iStep) / cn<tileN_> < L - blockIdx_y * Q - jK * cn<tileK_> + N
|
|
&& (!bnChk_ || nStart * cn<tileN_> + thread(iStep) % cn<tileN_> < P))
|
|
cp_shared_global<16>(
|
|
b_mxR + swizzle<tileN_ * 2, tileN_ * 2>(thread(iStep) * cn<2>, baseR(jK) * cn<2>),
|
|
g_mxX
|
|
+ get((jK * cn<tileK_> - N + thread(iStep) / cn<tileN_>) *X_stride + hStart * P
|
|
+ nStart * cn<tileN_> + thread(iStep) % cn<tileN_>));
|
|
else if (thread(iStep) < cn<tileN_ * tileK_>)
|
|
*(int4*) ((char*) s_mxR + swizzle<tileN_ * 2, tileN_ * 2>(thread(iStep) * cn<2>, baseR(jK) * cn<2>))
|
|
= int4{0, 0, 0, 0};
|
|
}
|
|
|
|
cp_commit_group();
|
|
cp_wait_group<pipeS_ - 1>();
|
|
|
|
__syncthreads();
|
|
}
|
|
|
|
cp_wait_group<0>();
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
if (pipeR_ == 2)
|
|
{
|
|
if (wmmaK_ == 16)
|
|
mma<Tp_>(r_mxAcc[y][x], r_mxL[1][y], r_mxR[1][x]);
|
|
}
|
|
}
|
|
|
|
if (g_mxD)
|
|
{
|
|
float r_D = g_mxD[hStart.var];
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
Tp_ tmp16[4] = {0};
|
|
float tmp32[4] = {0};
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_> + threadIdx_y * cn<wmmaN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[0] = *(int*) (g_mxX
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + threadIdx_z * cn<wmmaM_>
|
|
+ threadIdx_x / cn<4>) *X_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_>
|
|
+ threadIdx_y * cn<wmmaN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[0] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[0])
|
|
: bf1622float2(*(bf162*) &tmp16[0]);
|
|
|
|
r_mxAcc[y][x][0] += r_D * tmp32[0];
|
|
r_mxAcc[y][x][1] += r_D * tmp32[1];
|
|
}
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<8>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_> + threadIdx_y * cn<wmmaN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[2] = *(int*) (g_mxX
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<8>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>) *X_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_>
|
|
+ threadIdx_y * cn<wmmaN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[2] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[2])
|
|
: bf1622float2(*(bf162*) &tmp16[2]);
|
|
|
|
r_mxAcc[y][x][2] += r_D * tmp32[2];
|
|
r_mxAcc[y][x][3] += r_D * tmp32[3];
|
|
}
|
|
}
|
|
}
|
|
|
|
if (g_mxZ)
|
|
{
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
Tp_ tmp16[4] = {0};
|
|
float tmp32[4] = {0};
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_> + threadIdx_y * cn<wmmaN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[0] = *(int*) (g_mxZ
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + threadIdx_z * cn<wmmaM_>
|
|
+ threadIdx_x / cn<4>) *Z_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_>
|
|
+ threadIdx_y * cn<wmmaN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[0] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[0])
|
|
: bf1622float2(*(bf162*) &tmp16[0]);
|
|
|
|
float cut0 = std::max(tmp32[0], -256.f);
|
|
float cut1 = std::max(tmp32[1], -256.f);
|
|
r_mxAcc[y][x][0] *= cut0 / (1.f + expf(-cut0));
|
|
r_mxAcc[y][x][1] *= cut1 / (1.f + expf(-cut1));
|
|
}
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<8>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_> + threadIdx_y * cn<wmmaN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[2] = *(int*) (g_mxZ
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<8>
|
|
+ threadIdx_z * cn<wmmaM_> + threadIdx_x / cn<4>) *Z_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<warpN_ * wmmaN_>
|
|
+ threadIdx_y * cn<wmmaN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[2] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[2])
|
|
: bf1622float2(*(bf162*) &tmp16[2]);
|
|
|
|
float cut2 = std::max(tmp32[2], -256.f);
|
|
float cut3 = std::max(tmp32[3], -256.f);
|
|
r_mxAcc[y][x][2] *= cut2 / (1.f + expf(-cut2));
|
|
r_mxAcc[y][x][3] *= cut3 / (1.f + expf(-cut3));
|
|
}
|
|
}
|
|
}
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
if (std::is_same_v<Tp_, half>)
|
|
{
|
|
*(half2*) &r_mxAcc[y][x][0] = __floats2half2_rn(r_mxAcc[y][x][0], r_mxAcc[y][x][1]);
|
|
*(half2*) &r_mxAcc[y][x][2] = __floats2half2_rn(r_mxAcc[y][x][2], r_mxAcc[y][x][3]);
|
|
}
|
|
else
|
|
{
|
|
*(bf162*) &r_mxAcc[y][x][0] = __floats2bfloat162_rn(r_mxAcc[y][x][0], r_mxAcc[y][x][1]);
|
|
*(bf162*) &r_mxAcc[y][x][2] = __floats2bfloat162_rn(r_mxAcc[y][x][2], r_mxAcc[y][x][3]);
|
|
}
|
|
}
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
asm volatile("st.shared.b32 [%0], %1;\n" ::"r"(b_mxAcc
|
|
+ 2
|
|
* swz<tileN_ * 2, tileN_>(y * warpM_ * wmmaM_ * tileN_ + x * warpN_ * wmmaN_
|
|
+ (threadIdx.z * wmmaM_ + threadIdx.x / 4) * tileN_
|
|
+ (threadIdx.y * wmmaN_ + threadIdx.x % 4 * 2))),
|
|
"r"(*(unsigned*) &r_mxAcc[y][x][0]));
|
|
asm volatile("st.shared.b32 [%0], %1;\n" ::"r"(b_mxAcc
|
|
+ 2
|
|
* swz<tileN_ * 2, tileN_>(y * warpM_ * wmmaM_ * tileN_ + 8 * tileN_
|
|
+ x * warpN_ * wmmaN_ + (threadIdx.z * wmmaM_ + threadIdx.x / 4) * tileN_
|
|
+ (threadIdx.y * wmmaN_ + threadIdx.x % 4 * 2))),
|
|
"r"(*(unsigned*) &r_mxAcc[y][x][2]));
|
|
}
|
|
|
|
__syncthreads();
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileM_ * tileN_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileM_ * tileN_> && thread(iStep) / cn<tileN_> < L - blockIdx_y * Q - mStart * cn<tileM_>
|
|
&& (!bnChk_ || nStart * cn<tileN_> + thread(iStep) % cn<tileN_> < P))
|
|
*(int4*) (g_mxY
|
|
+ get((mStart * cn<tileM_> + thread(iStep) / cn<tileN_>) *H * P + hStart * P + nStart * cn<tileN_>
|
|
+ thread(iStep) % cn<tileN_>))
|
|
= *(int4*) ((char*) s_mxAcc + swizzle<tileN_ * 2, tileN_ * 2>(thread(iStep) * cn<2>));
|
|
#endif
|
|
}
|
|
|
|
template <int Q_, int tileM_, int tileN_, int tileK_, // smem size, per sm
|
|
int warpM_, int warpN_, // warp number
|
|
int pipeS_, int pipeR_, int group_, int bnChk_, class Tp_, class Wt_>
|
|
__global__ std::enable_if_t<std::is_same_v<Tp_, half> || std::is_same_v<Tp_, __nv_bfloat16>> chunk_scan_hopper(int B_,
|
|
int L_, int H_, int P_, int G_, int N_,
|
|
void* g_mxY_, // Tp_ B*L*H*P
|
|
void const* g_mxOs_, // Tp_ B*C*H*N*P
|
|
// const void *g_mxFs_, // Tp_ B *H*N*P
|
|
// const void *g_mxSt_, // float B*C*H*N*P
|
|
void const* g_mxdc_, // float B*C*H*Q
|
|
void const* g_mxdA_, // float B*C*H*Q
|
|
// const void *g_mxdt_, // Tp_ B*L*((g_mxZ?2:1)*H*P+2*G+round_up(H,8))
|
|
// const void *g_mxdb_, // Wt_ H
|
|
// const void *g_mxA_, // Wt_ H
|
|
void const* g_mxCB_, // Tp_ B*C*G*Q*Q
|
|
void const* g_mxD_, // Wt_ H
|
|
void const* g_mxX_, // Tp_ B*L*(H*P+2*G*N)
|
|
void const* g_mxZ_, // g_mxdt_ or nullptr
|
|
bool removePadding_, int const* lastTokenIdsPtr_)
|
|
{
|
|
#if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 900 && defined(__CUDA_ARCH_FEAT_SM90_ALL)
|
|
using namespace tensorrt_llm::common;
|
|
|
|
constexpr int wmmaM_ = 16; // wmma size, per instruction
|
|
constexpr int wmmaN_ = 8;
|
|
constexpr int wmmaK_ = 16;
|
|
|
|
auto blockIdx_x = Rn<ID>{int(blockIdx.x)};
|
|
auto blockIdx_y = Rn<ID>{int(blockIdx.y)};
|
|
auto blockIdx_z = Rn<ID>{int(blockIdx.z)};
|
|
|
|
auto threadIdx_x = Rn<ID, 32>{int(threadIdx.x)};
|
|
auto threadIdx_y = Rn<ID, warpM_>{int(threadIdx.y)};
|
|
auto threadIdx_z = Rn<ID, warpN_>{int(threadIdx.z)};
|
|
|
|
// auto B = Rn<ID>{B_};
|
|
auto L = Rn<ID>{L_};
|
|
auto H = Rn<ID>{H_};
|
|
auto P = Rn<ID>{P_};
|
|
auto G = Rn<ID>{G_};
|
|
auto N = Rn<ID>{N_};
|
|
auto Q = cn<Q_>;
|
|
auto C = Rn<ID>{div_up(L.var, Q_)};
|
|
|
|
auto X_stride = Rn<ID>{H_ * P_ + 2 * G_ * N_};
|
|
auto Z_stride = Rn<ID>{(g_mxZ_ ? 2 : 1) * H_ * P_ + 2 * G_ * N_ + round_up(H_, 8)};
|
|
|
|
auto aStart = blockIdx_z * L;
|
|
auto cStart = blockIdx_z * C;
|
|
|
|
if (removePadding_)
|
|
{
|
|
aStart = Rn<ID>{int(blockIdx.z ? lastTokenIdsPtr_[blockIdx.z - 1] : 0)};
|
|
cStart = Rn<ID>{int(blockIdx.z ? div_up(aStart.var, Q_) + blockIdx.z - 1 : 0)};
|
|
L = Rn<ID>{lastTokenIdsPtr_[blockIdx.z] - aStart.var};
|
|
C = Rn<ID>{div_up(L.var, Q_)};
|
|
}
|
|
else
|
|
{
|
|
L = Rn<ID>{lastTokenIdsPtr_[blockIdx.z]};
|
|
C = Rn<ID>{div_up(L.var, Q_)};
|
|
}
|
|
|
|
int numBlocks_m = H_ * (Q_ / tileM_);
|
|
int numBlocks_n = div_up(P_, tileN_);
|
|
|
|
int numBlocks_group = numBlocks_n * group_;
|
|
int blockIdx_0 = blockIdx.x / numBlocks_group * group_;
|
|
int group_m = std::min(group_, numBlocks_m - blockIdx_0);
|
|
|
|
int mStartInt = blockIdx.x % numBlocks_group % group_m + blockIdx_0;
|
|
int nStartInt = blockIdx.x % numBlocks_group / group_m;
|
|
|
|
auto hStart = Rn<ID>{mStartInt} / (Q / cn<tileM_>);
|
|
auto mStart = Rn<ID>{mStartInt} % (Q / cn<tileM_>);
|
|
auto nStart = Rn<ID>{nStartInt};
|
|
auto gStart = Rn<ID>{hStart.var / (H_ / G_)};
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> >= L)
|
|
return;
|
|
|
|
Tp_* g_mxY = (Tp_*) g_mxY_ + int64_t(get(aStart + blockIdx_y * Q)) * get(H * P);
|
|
// const Tp_ *g_mxOs = (const Tp_ *)g_mxOs_;
|
|
// const Tp_ *g_mxFs = (const Tp_ *)g_mxFs_;
|
|
// const float *g_mxSt = (const float *)g_mxSt_;
|
|
float const* g_mxdc = (float const*) g_mxdc_ + int64_t(get(cStart + blockIdx_y)) * get(H * Q);
|
|
float const* g_mxdA = (float const*) g_mxdA_ + int64_t(get(cStart + blockIdx_y)) * get(H * Q);
|
|
// const Tp_ *g_mxdt = (const Tp_ *)g_mxdt_;
|
|
// const Wt_ *g_mxdb = (const Wt_ *)g_mxdb_;
|
|
// const Wt_ *g_mxA = (const Wt_ *)g_mxA_;
|
|
// const Tp_ *g_mxCB = (const Tp_ *)g_mxCB_;
|
|
Wt_ const* g_mxD = (Wt_ const*) g_mxD_;
|
|
Tp_ const* g_mxX = (Tp_ const*) (*(uint64_t const*) ((CUtensorMap const*) g_mxX_ + 3))
|
|
+ int64_t(get(aStart + blockIdx_y * Q)) * get(X_stride);
|
|
Tp_ const* g_mxZ = (Tp_ const*) g_mxZ_ + (g_mxZ_ ? int64_t(get(aStart + blockIdx_y * Q)) * get(Z_stride) : 0);
|
|
|
|
extern __shared__ float smem[];
|
|
|
|
Tp_* s_mxL = (Tp_*) smem + 512;
|
|
Tp_* s_mxR = (Tp_*) smem + 512 + tileM_ * tileK_ * pipeS_;
|
|
Tp_* s_mxAcc = (Tp_*) smem + 512;
|
|
|
|
float* s_mxdc = smem + 256 + (tileM_ + tileN_) * tileK_ * pipeS_ / 2;
|
|
float* s_mxdA = smem + 256 + (tileM_ + tileN_) * tileK_ * pipeS_ / 2 + Q_;
|
|
|
|
unsigned b_base = __nvvm_get_smem_pointer(smem);
|
|
|
|
unsigned b_mxL = b_base + 1024;
|
|
unsigned b_mxR = b_base + 1024 + tileM_ * tileK_ * pipeS_ * sizeof(Tp_);
|
|
unsigned b_mxAcc = b_base + 1024;
|
|
unsigned b_mbar = b_base;
|
|
|
|
uint32_t formatL = tileK_ >= 64 ? 1 : (tileK_ == 32 ? 2 : 3);
|
|
uint32_t formatR = tileN_ >= 64 ? 1 : (tileN_ == 32 ? 2 : 3);
|
|
uint32_t baseOffsetL = 0;
|
|
uint32_t baseOffsetR = 0;
|
|
uint32_t strideDimL = tileK_ >= 64 ? warpM_ / 4 * 64 : warpM_ / 4 * tileK_;
|
|
uint32_t strideDimR = tileN_ >= 64 ? 64 : tileN_;
|
|
uint32_t leadingDimL = 0;
|
|
uint32_t leadingDimR = tileN_ >= 64 ? tileK_ * 8 : 0;
|
|
uint32_t startAddrL = (b_mxL + threadIdx.y / 4 * 8 * tileK_ * 2) / 16;
|
|
uint32_t startAddrR = (b_mxR + threadIdx.z * tileN_ / warpN_ % strideDimR * 2
|
|
+ threadIdx.z * tileN_ / warpN_ / strideDimR * tileK_ * strideDimR * 2)
|
|
/ 16;
|
|
|
|
uint64_t d_mxL = ((uint64_t) formatL << 62) + ((uint64_t) baseOffsetL << 49) + ((uint64_t) strideDimL << 32)
|
|
+ ((uint64_t) leadingDimL << 16) + ((uint64_t) startAddrL);
|
|
uint64_t d_mxR = ((uint64_t) formatR << 62) + ((uint64_t) baseOffsetR << 49) + ((uint64_t) strideDimR << 32)
|
|
+ ((uint64_t) leadingDimR << 16) + ((uint64_t) startAddrR);
|
|
|
|
using std::array;
|
|
|
|
array<array<array<float, wmmaM_ * wmmaN_ / 32>, tileN_ / wmmaN_ / warpN_>, tileM_ / wmmaM_ / warpM_> r_mxAcc
|
|
= array<array<array<float, wmmaM_ * wmmaN_ / 32>, tileN_ / wmmaN_ / warpN_>, tileM_ / wmmaM_ / warpM_>();
|
|
|
|
auto baseL = [](auto iK) { return iK % cn<pipeS_> * cn<tileM_> * cn<tileK_>; };
|
|
auto baseR = [](auto iK) { return iK % cn<pipeS_> * cn<tileN_> * cn<tileK_>; };
|
|
|
|
auto thread = [=](auto iStep)
|
|
{
|
|
return iStep * cn<warpM_ * warpN_ * 256> + threadIdx_y * cn<warpN_ * 256> + threadIdx_z * cn<256>
|
|
+ threadIdx_x * cn<8>;
|
|
};
|
|
|
|
asm volatile(
|
|
".reg .pred elected_one;\n"
|
|
".reg .pred done;\n"
|
|
"elect.sync _|elected_one, 0xFFFFFFFF;\n"
|
|
"@!elected_one setp.eq.b32 done, 0, 0;\n" ::);
|
|
|
|
if (threadIdx.y == 0 && threadIdx.z == 0)
|
|
{
|
|
#pragma unroll
|
|
for (Rn<UNROLL, pipeS_> iK; iK.var < iK.size; iK.var++)
|
|
asm volatile("@elected_one mbarrier.init.shared.b64 [%0], %1;\n" ::"r"(b_mbar + iK.var * 8), "r"(1));
|
|
}
|
|
|
|
__syncthreads();
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(Q_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<Q_>)
|
|
{
|
|
#pragma unroll
|
|
for (int i = 0; i < 8; i += 4)
|
|
{
|
|
// dc should be set to zero out of seq length, but this is done
|
|
// by chunkcumsum, so no need here.
|
|
*(int4*) (s_mxdc + get(thread(iStep)) + i) = *(int4*) (g_mxdc + get(hStart * Q + thread(iStep)) + i);
|
|
*(int4*) (s_mxdA + get(thread(iStep)) + i) = *(int4*) (g_mxdA + get(hStart * Q + thread(iStep)) + i);
|
|
}
|
|
}
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, pipeS_> iK; iK.var < iK.size; iK.var++)
|
|
{
|
|
if (threadIdx.y == 0 && threadIdx.z == 0)
|
|
{
|
|
asm volatile("@elected_one mbarrier.arrive.expect_tx.shared.b64 _, [%0], %1;\n" ::"r"(b_mbar + iK.var * 8),
|
|
"r"((tileM_ + tileN_) * tileK_ * 2));
|
|
asm volatile(
|
|
"@elected_one cp.async.bulk.tensor.3d.shared::cluster.global.mbarrier::complete_tx::bytes"
|
|
" [%0], [%1, {%2, %3, %4}], [%5];\n" ::"r"(b_mxL + iK.var % pipeS_ * (tileM_ * tileK_ * 2)),
|
|
"l"((CUtensorMap const*) g_mxX_), "r"(iK.var * tileK_), "r"(gStart.var),
|
|
"r"(get(aStart + blockIdx_y * Q + mStart * cn<tileM_>)), "r"(b_mbar + iK.var * 8));
|
|
#pragma unroll
|
|
for (int iN = 0; iN < tileN_; iN += 64)
|
|
asm volatile(
|
|
"@elected_one cp.async.bulk.tensor.4d.shared::cluster.global.mbarrier::complete_tx::bytes"
|
|
" [%0], [%1, {%2, %3, %4, %5}], [%6];\n" ::"r"(
|
|
b_mxR + iK.var % pipeS_ * (tileN_ * tileK_ * 2) + iN * tileK_ * 2),
|
|
"l"((CUtensorMap const*) g_mxX_ + 1), "r"(get(nStart * cn<tileN_>) + iN), "r"(iK.var * tileK_),
|
|
"r"(hStart.var), "r"(get(cStart + blockIdx_y)), "r"(b_mbar + iK.var * 8));
|
|
}
|
|
}
|
|
|
|
__syncthreads();
|
|
|
|
int Q1 = std::min(Q_, (get(mStart) + 1) * tileM_);
|
|
|
|
for (int iK = pipeS_; iK < (N_ + Q1) / tileK_ + pipeS_; iK++)
|
|
{
|
|
uint64_t d_mxL0 = d_mxL + iK % pipeS_ * (tileM_ * tileK_ * 2) / 16;
|
|
uint64_t d_mxR0 = d_mxR + iK % pipeS_ * (tileN_ * tileK_ * 2) / 16;
|
|
|
|
asm volatile(
|
|
"{\n"
|
|
"TRY_AGAIN:\n"
|
|
"mbarrier.try_wait.parity.shared.b64 done, [%0], %1;\n"
|
|
"@!done nanosleep.u32 %2;\n"
|
|
"@!done bra TRY_AGAIN;\n"
|
|
"}\n" ::"r"(b_mbar + iK % pipeS_ * 8),
|
|
"r"(1 - iK / pipeS_ % 2), "r"(threadIdx.x % 6));
|
|
|
|
auto jK = Rn<>{iK};
|
|
if ((iK - pipeS_) * cn<tileK_> == N_)
|
|
{
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
float2 tmp2 = float2{expf(s_mxdA[get(mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_>
|
|
+ threadIdx_y / cn<4> * cn<8>
|
|
+ threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_> + threadIdx_x / cn<4>)]),
|
|
expf(s_mxdA[get(mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<warpM_ / 4 * 8>
|
|
+ threadIdx_y / cn<4> * cn<8> + threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_>
|
|
+ threadIdx_x / cn<4>)])};
|
|
|
|
r_mxAcc[y][x][0] *= tmp2.x;
|
|
r_mxAcc[y][x][1] *= tmp2.x;
|
|
r_mxAcc[y][x][2] *= tmp2.y;
|
|
r_mxAcc[y][x][3] *= tmp2.y;
|
|
}
|
|
}
|
|
|
|
if ((iK - pipeS_) * cn<tileK_> >= N_)
|
|
{
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileM_ * tileK_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileM_ * tileK_>)
|
|
{
|
|
Tp_ tmpL[8];
|
|
|
|
*(int4*) &tmpL[0]
|
|
= *(int4*) ((char*) s_mxL + swizzle<tileK_ * 2, 128>(thread(iStep) * cn<2>, baseL(jK) * cn<2>));
|
|
|
|
#pragma unroll
|
|
for (int i = 0; i < 8; i += 2)
|
|
{
|
|
float2 tmp2 = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmpL[i])
|
|
: bf1622float2(*(bf162*) &tmpL[i]);
|
|
|
|
int kStart = (iK - pipeS_) * cn<tileK_> - N_;
|
|
|
|
// dc should be set to zero out of seq length, but this is done
|
|
// by chunkcumsum, so no need here.
|
|
tmp2.x *= expf(s_mxdA[get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>)]
|
|
- s_mxdA[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i})])
|
|
* s_mxdc[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i})];
|
|
tmp2.y *= expf(s_mxdA[get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>)]
|
|
- s_mxdA[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i + 1})])
|
|
* s_mxdc[kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i + 1})];
|
|
|
|
if (std::min(
|
|
get(L - blockIdx_y * Q - cn<1>), get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>))
|
|
< kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i}))
|
|
tmp2.x = 0;
|
|
if (std::min(
|
|
get(L - blockIdx_y * Q - cn<1>), get(mStart * cn<tileM_> + thread(iStep) / cn<tileK_>))
|
|
< kStart + get(thread(iStep) % cn<tileK_> + Rn<UNROLL>{i + 1}))
|
|
tmp2.y = 0;
|
|
|
|
if (std::is_same_v<Tp_, half>)
|
|
*(half2*) &tmpL[i] = __float22half2_rn(tmp2);
|
|
else
|
|
*(bf162*) &tmpL[i] = __float22bfloat162_rn(tmp2);
|
|
}
|
|
|
|
*(int4*) ((char*) s_mxL + swizzle<tileK_ * 2, 128>(thread(iStep) * cn<2>, baseL(jK) * cn<2>))
|
|
= *(int4*) &tmpL[0];
|
|
}
|
|
|
|
__syncthreads();
|
|
}
|
|
|
|
asm volatile("fence.proxy.async.shared::cta; \n");
|
|
asm volatile("wgmma.fence.sync.aligned; \n");
|
|
|
|
#pragma unroll
|
|
for (int k = 0; k < tileK_ / wmmaK_; k++)
|
|
{
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
{
|
|
wgmma<Tp_, 0, 1>(r_mxAcc[y], d_mxL0 + k * 2 + y * warpM_ * wmmaM_ * (tileK_ * 2 / 16),
|
|
d_mxR0 + k * (tileN_ >= 64 ? 128 : 64));
|
|
}
|
|
}
|
|
|
|
asm volatile("wgmma.commit_group.sync.aligned;");
|
|
asm volatile("wgmma.wait_group.sync.aligned %0;" ::"n"(0));
|
|
|
|
__syncthreads();
|
|
|
|
if (iK * tileK_ < N_)
|
|
{
|
|
|
|
if (threadIdx.y == 0 && threadIdx.z == 0)
|
|
{
|
|
asm volatile("@elected_one mbarrier.arrive.expect_tx.shared.b64 _, [%0], %1;\n" ::"r"(
|
|
b_mbar + jK.var % pipeS_ * 8),
|
|
"r"((tileM_ + tileN_) * tileK_ * 2));
|
|
asm volatile(
|
|
"@elected_one cp.async.bulk.tensor.3d.shared::cluster.global.mbarrier::complete_tx::bytes"
|
|
" [%0], [%1, {%2, %3, %4}], [%5];\n" ::"r"(b_mxL + jK.var % pipeS_ * (tileM_ * tileK_ * 2)),
|
|
"l"((CUtensorMap const*) g_mxX_), "r"(jK.var * tileK_), "r"(gStart.var),
|
|
"r"(get(aStart + blockIdx_y * Q + mStart * cn<tileM_>)), "r"(b_mbar + jK.var % pipeS_ * 8));
|
|
#pragma unroll
|
|
for (int iN = 0; iN < tileN_; iN += 64)
|
|
asm volatile(
|
|
"@elected_one cp.async.bulk.tensor.4d.shared::cluster.global.mbarrier::complete_tx::bytes"
|
|
" [%0], [%1, {%2, %3, %4, %5}], [%6];\n" ::"r"(
|
|
b_mxR + jK.var % pipeS_ * (tileN_ * tileK_ * 2) + iN * tileK_ * 2),
|
|
"l"((CUtensorMap const*) g_mxX_ + 1), "r"(get(nStart * cn<tileN_>) + iN), "r"(jK.var * tileK_),
|
|
"r"(hStart.var), "r"(get(cStart + blockIdx_y)), "r"(b_mbar + jK.var % pipeS_ * 8));
|
|
}
|
|
}
|
|
else if (iK * tileK_ < N_ + Q1)
|
|
{
|
|
|
|
if (threadIdx.y == 0 && threadIdx.z == 0)
|
|
{
|
|
asm volatile("@elected_one mbarrier.arrive.expect_tx.shared.b64 _, [%0], %1;\n" ::"r"(
|
|
b_mbar + jK.var % pipeS_ * 8),
|
|
"r"((tileM_ + tileN_) * tileK_ * 2));
|
|
asm volatile(
|
|
"@elected_one cp.async.bulk.tensor.4d.shared::cluster.global.mbarrier::complete_tx::bytes"
|
|
" [%0], [%1, {%2, %3, %4, %5}], [%6];\n" ::"r"(b_mxL + jK.var % pipeS_ * (tileM_ * tileK_ * 2)),
|
|
"l"((CUtensorMap const*) g_mxX_ + 2), "r"(jK.var * tileK_ - N_), "r"(get(mStart * cn<tileM_>)),
|
|
"r"(gStart.var), "r"(get(cStart + blockIdx_y)), "r"(b_mbar + jK.var % pipeS_ * 8));
|
|
#pragma unroll
|
|
for (int iN = 0; iN < tileN_; iN += 64)
|
|
asm volatile(
|
|
"@elected_one cp.async.bulk.tensor.3d.shared::cluster.global.mbarrier::complete_tx::bytes"
|
|
" [%0], [%1, {%2, %3, %4}], [%5];\n" ::"r"(
|
|
b_mxR + jK.var % pipeS_ * (tileN_ * tileK_ * 2) + iN * tileK_ * 2),
|
|
"l"((CUtensorMap const*) g_mxX_ + 3), "r"(get(nStart * cn<tileN_>) + iN), "r"(hStart.var),
|
|
"r"(get(aStart + blockIdx_y * Q + jK * cn<tileK_> - N)), "r"(b_mbar + jK.var % pipeS_ * 8));
|
|
}
|
|
}
|
|
}
|
|
|
|
if (threadIdx.y == 0 && threadIdx.z == 0)
|
|
{
|
|
#pragma unroll
|
|
for (Rn<UNROLL, pipeS_> iK; iK.var < iK.size; iK.var++)
|
|
asm volatile("@elected_one mbarrier.inval.shared.b64 [%0];\n" ::"r"(b_mbar + iK.var * 8));
|
|
}
|
|
|
|
if (g_mxD)
|
|
{
|
|
float r_D = g_mxD[hStart.var];
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
Tp_ tmp16[4] = {0};
|
|
float tmp32[4] = {0};
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_>
|
|
+ threadIdx_y / cn<4> * cn<8> + threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_>
|
|
+ threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_> + threadIdx_z * cn<tileN_ / warpN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[0] = *(int*) (g_mxX
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + threadIdx_y / cn<4> * cn<8>
|
|
+ threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_> + threadIdx_x / cn<4>) *X_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_>
|
|
+ threadIdx_z * cn<tileN_ / warpN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[0] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[0])
|
|
: bf1622float2(*(bf162*) &tmp16[0]);
|
|
|
|
r_mxAcc[y][x][0] += r_D * tmp32[0];
|
|
r_mxAcc[y][x][1] += r_D * tmp32[1];
|
|
}
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<warpM_ / 4 * 8>
|
|
+ threadIdx_y / cn<4> * cn<8> + threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_>
|
|
+ threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_> + threadIdx_z * cn<tileN_ / warpN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[2] = *(int*) (g_mxX
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<warpM_ / 4 * 8>
|
|
+ threadIdx_y / cn<4> * cn<8> + threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_>
|
|
+ threadIdx_x / cn<4>) *X_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_>
|
|
+ threadIdx_z * cn<tileN_ / warpN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[2] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[2])
|
|
: bf1622float2(*(bf162*) &tmp16[2]);
|
|
|
|
r_mxAcc[y][x][2] += r_D * tmp32[2];
|
|
r_mxAcc[y][x][3] += r_D * tmp32[3];
|
|
}
|
|
}
|
|
}
|
|
|
|
if (g_mxZ)
|
|
{
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
Tp_ tmp16[4] = {0};
|
|
float tmp32[4] = {0};
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_>
|
|
+ threadIdx_y / cn<4> * cn<8> + threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_>
|
|
+ threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_> + threadIdx_z * cn<tileN_ / warpN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[0] = *(int*) (g_mxZ
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + threadIdx_y / cn<4> * cn<8>
|
|
+ threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_> + threadIdx_x / cn<4>) *Z_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_>
|
|
+ threadIdx_z * cn<tileN_ / warpN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[0] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[0])
|
|
: bf1622float2(*(bf162*) &tmp16[0]);
|
|
|
|
float cut0 = std::max(tmp32[0], -256.f);
|
|
float cut1 = std::max(tmp32[1], -256.f);
|
|
r_mxAcc[y][x][0] *= cut0 / (1.f + expf(-cut0));
|
|
r_mxAcc[y][x][1] *= cut1 / (1.f + expf(-cut1));
|
|
}
|
|
|
|
if (blockIdx_y * Q + mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<warpM_ / 4 * 8>
|
|
+ threadIdx_y / cn<4> * cn<8> + threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_>
|
|
+ threadIdx_x / cn<4>
|
|
< L
|
|
&& (!bnChk_
|
|
|| nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_> + threadIdx_z * cn<tileN_ / warpN_>
|
|
+ threadIdx_x % cn<4> * cn<2>
|
|
< P))
|
|
{
|
|
*(int*) &tmp16[2] = *(int*) (g_mxZ
|
|
+ get((mStart * cn<tileM_> + Rn<UNROLL>{y} * cn<warpM_ * wmmaM_> + cn<warpM_ / 4 * 8>
|
|
+ threadIdx_y / cn<4> * cn<8> + threadIdx_y % cn<4> * cn<warpM_ / 4 * wmmaM_>
|
|
+ threadIdx_x / cn<4>) *Z_stride
|
|
+ hStart * P + nStart * cn<tileN_> + Rn<UNROLL>{x} * cn<wmmaN_>
|
|
+ threadIdx_z * cn<tileN_ / warpN_> + threadIdx_x % cn<4> * cn<2>));
|
|
|
|
*(float2*) &tmp32[2] = std::is_same_v<Tp_, half> ? __half22float2(*(half2*) &tmp16[2])
|
|
: bf1622float2(*(bf162*) &tmp16[2]);
|
|
|
|
float cut2 = std::max(tmp32[2], -256.f);
|
|
float cut3 = std::max(tmp32[3], -256.f);
|
|
r_mxAcc[y][x][2] *= cut2 / (1.f + expf(-cut2));
|
|
r_mxAcc[y][x][3] *= cut3 / (1.f + expf(-cut3));
|
|
}
|
|
}
|
|
}
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
if (std::is_same_v<Tp_, half>)
|
|
{
|
|
*(half2*) &r_mxAcc[y][x][0] = __floats2half2_rn(r_mxAcc[y][x][0], r_mxAcc[y][x][1]);
|
|
*(half2*) &r_mxAcc[y][x][2] = __floats2half2_rn(r_mxAcc[y][x][2], r_mxAcc[y][x][3]);
|
|
}
|
|
else
|
|
{
|
|
*(bf162*) &r_mxAcc[y][x][0] = __floats2bfloat162_rn(r_mxAcc[y][x][0], r_mxAcc[y][x][1]);
|
|
*(bf162*) &r_mxAcc[y][x][2] = __floats2bfloat162_rn(r_mxAcc[y][x][2], r_mxAcc[y][x][3]);
|
|
}
|
|
}
|
|
|
|
#pragma unroll
|
|
for (int y = 0; y < tileM_ / wmmaM_ / warpM_; y++)
|
|
#pragma unroll
|
|
for (int x = 0; x < tileN_ / wmmaN_ / warpN_; x++)
|
|
{
|
|
asm volatile(
|
|
"st.shared.b32 [%0], %1;\n" ::"r"(b_mxAcc
|
|
+ 2
|
|
* swz<tileN_ * 2, tileN_>(y * warpM_ * wmmaM_ * tileN_ + x * wmmaN_
|
|
+ (threadIdx.y / 4 * 8 + threadIdx.y % 4 * (warpM_ / 4 * wmmaM_) + +threadIdx.x / 4)
|
|
* tileN_
|
|
+ (threadIdx.z * tileN_ / warpN_ + threadIdx.x % 4 * 2))),
|
|
"r"(*(unsigned*) &r_mxAcc[y][x][0]));
|
|
asm volatile(
|
|
"st.shared.b32 [%0], %1;\n" ::"r"(b_mxAcc
|
|
+ 2
|
|
* swz<tileN_ * 2, tileN_>(y * warpM_ * wmmaM_ * tileN_ + warpM_ / 4 * 8 * tileN_ + x * wmmaN_
|
|
+ (threadIdx.y / 4 * 8 + threadIdx.y % 4 * (warpM_ / 4 * wmmaM_) + +threadIdx.x / 4)
|
|
* tileN_
|
|
+ (threadIdx.z * tileN_ / warpN_ + threadIdx.x % 4 * 2))),
|
|
"r"(*(unsigned*) &r_mxAcc[y][x][2]));
|
|
}
|
|
|
|
__syncthreads();
|
|
|
|
#pragma unroll
|
|
for (Rn<UNROLL, div_up(tileM_ * tileN_, warpM_ * warpN_ * 256)> iStep; iStep.var < iStep.size; iStep.var++)
|
|
if (thread(iStep) < cn<tileM_ * tileN_> && thread(iStep) / cn<tileN_> < L - blockIdx_y * Q - mStart * cn<tileM_>
|
|
&& (!bnChk_ || nStart * cn<tileN_> + thread(iStep) % cn<tileN_> < P))
|
|
*(int4*) (g_mxY
|
|
+ get((mStart * cn<tileM_> + thread(iStep) / cn<tileN_>) *H * P + hStart * P + nStart * cn<tileN_>
|
|
+ thread(iStep) % cn<tileN_>))
|
|
= *(int4*) ((char*) s_mxAcc + swizzle<tileN_ * 2, tileN_ * 2>(thread(iStep) * cn<2>));
|
|
#endif
|
|
}
|
|
|
|
typedef ChunkScanKernelFunc (*GetChunkScanKernelFunc)(int B_, int L_, int H_, int P_, int G_, int N_, int Q_,
|
|
int numTokens_, int isHopper_, tensorrt_llm::common::CUDADriverWrapper* cudaDriver_, dim3* blockDims_,
|
|
dim3* threadDims_, int* sharedMem_, int* useTma_, CUtensorMap* descs_);
|
|
|
|
template <class Tp_, class Wt_>
|
|
ChunkScanKernelFunc getChunkScanKernel(int B_, int L_, int H_, int P_, int G_, int N_, int Q_, int numTokens_,
|
|
int isHopper_, tensorrt_llm::common::CUDADriverWrapper* cudaDriver_, dim3* blockDims_, dim3* threadDims_,
|
|
int* sharedMem_, int* useTma_, CUtensorMap* descs_)
|
|
{
|
|
int B = B_;
|
|
int L = L_;
|
|
int H = H_;
|
|
int P = P_;
|
|
int G = G_;
|
|
int N = N_;
|
|
int Q = Q_;
|
|
int C = div_up(L, Q);
|
|
|
|
int64_t compute = int64_t(numTokens_) * H * P_ * (N + Q);
|
|
|
|
auto set
|
|
= [&](int tileM, int tileN, int tileK, int warpM, int warpN, int pipeS, int useTma, ChunkScanKernelFunc func)
|
|
{
|
|
auto sharedMem
|
|
= useTma * 1024 + std::max((tileM * tileK + tileK * tileN) * pipeS * 2 + Q * 8, (tileM * tileN) * 2);
|
|
|
|
*blockDims_ = dim3(H * div_up(P_, tileN) * Q / tileM, C, B);
|
|
*threadDims_ = dim3(32, useTma ? warpM : warpN, useTma ? warpN : warpM);
|
|
*sharedMem_ = sharedMem;
|
|
*useTma_ = useTma;
|
|
|
|
if (useTma)
|
|
{
|
|
{
|
|
std::array<uint64_t, 2> tensorStrides{2uL * N, 2uL * (H * P_ + 2 * G * N)};
|
|
std::array<uint64_t, 3> tensorSizes{1uL * N, 1uL * G, 1uL * numTokens_};
|
|
std::array<uint32_t, 3> traveralStrides{1u, 1u, 1u};
|
|
std::array<uint32_t, 3> boxSizes{(uint32_t) tileK, 1u, (uint32_t) tileM};
|
|
|
|
cudaDriver_->cuTensorMapEncodeTiled(&descs_[0], CU_TENSOR_MAP_DATA_TYPE_FLOAT16, 3, nullptr,
|
|
&tensorSizes[0], &tensorStrides[0], &boxSizes[0], &traveralStrides[0],
|
|
CU_TENSOR_MAP_INTERLEAVE_NONE, tileK == 64 ? CU_TENSOR_MAP_SWIZZLE_128B : CU_TENSOR_MAP_SWIZZLE_64B,
|
|
CU_TENSOR_MAP_L2_PROMOTION_L2_128B, CU_TENSOR_MAP_FLOAT_OOB_FILL_NONE);
|
|
}
|
|
{
|
|
std::array<uint64_t, 3> tensorStrides{2uL * P_, 2uL * N * P_, 2uL * H * N * P_};
|
|
std::array<uint64_t, 4> tensorSizes{
|
|
1uL * P_, 1uL * N, 1uL * H, 1uL * div_up(numTokens_ + B * Q - Q, Q)};
|
|
std::array<uint32_t, 4> traveralStrides{1u, 1u, 1u, 1u};
|
|
std::array<uint32_t, 4> boxSizes{tileN >= 64 ? 64u : (uint32_t) tileN, (uint32_t) tileK, 1u, 1u};
|
|
|
|
cudaDriver_->cuTensorMapEncodeTiled(&descs_[1], CU_TENSOR_MAP_DATA_TYPE_FLOAT16, 4, nullptr,
|
|
&tensorSizes[0], &tensorStrides[0], &boxSizes[0], &traveralStrides[0],
|
|
CU_TENSOR_MAP_INTERLEAVE_NONE, tileN >= 64 ? CU_TENSOR_MAP_SWIZZLE_128B : CU_TENSOR_MAP_SWIZZLE_64B,
|
|
CU_TENSOR_MAP_L2_PROMOTION_L2_128B, CU_TENSOR_MAP_FLOAT_OOB_FILL_NONE);
|
|
}
|
|
{
|
|
std::array<uint64_t, 3> tensorStrides{2uL * Q, 2uL * Q * Q, 2uL * G * Q * Q};
|
|
std::array<uint64_t, 4> tensorSizes{1uL * Q, 1uL * Q, 1uL * G, 1uL * div_up(numTokens_ + B * Q - Q, Q)};
|
|
std::array<uint32_t, 4> traveralStrides{1u, 1u, 1u, 1u};
|
|
std::array<uint32_t, 4> boxSizes{(uint32_t) tileK, (uint32_t) tileM, 1u, 1u};
|
|
|
|
cudaDriver_->cuTensorMapEncodeTiled(&descs_[2], CU_TENSOR_MAP_DATA_TYPE_FLOAT16, 4, nullptr,
|
|
&tensorSizes[0], &tensorStrides[0], &boxSizes[0], &traveralStrides[0],
|
|
CU_TENSOR_MAP_INTERLEAVE_NONE, tileK == 64 ? CU_TENSOR_MAP_SWIZZLE_128B : CU_TENSOR_MAP_SWIZZLE_64B,
|
|
CU_TENSOR_MAP_L2_PROMOTION_L2_128B, CU_TENSOR_MAP_FLOAT_OOB_FILL_NONE);
|
|
}
|
|
{
|
|
std::array<uint64_t, 2> tensorStrides{2uL * P_, 2uL * (H * P_ + 2 * G * N)};
|
|
std::array<uint64_t, 3> tensorSizes{1uL * P_, 1uL * H, 1uL * numTokens_};
|
|
std::array<uint32_t, 3> traveralStrides{1u, 1u, 1u};
|
|
std::array<uint32_t, 3> boxSizes{tileN >= 64 ? 64u : (uint32_t) tileN, 1u, (uint32_t) tileK};
|
|
|
|
cudaDriver_->cuTensorMapEncodeTiled(&descs_[3], CU_TENSOR_MAP_DATA_TYPE_FLOAT16, 3, nullptr,
|
|
&tensorSizes[0], &tensorStrides[0], &boxSizes[0], &traveralStrides[0],
|
|
CU_TENSOR_MAP_INTERLEAVE_NONE, tileN >= 64 ? CU_TENSOR_MAP_SWIZZLE_128B : CU_TENSOR_MAP_SWIZZLE_64B,
|
|
CU_TENSOR_MAP_L2_PROMOTION_L2_128B, CU_TENSOR_MAP_FLOAT_OOB_FILL_NONE);
|
|
}
|
|
}
|
|
|
|
return func;
|
|
};
|
|
|
|
if (Q == 256 && isHopper_ == false)
|
|
{
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 34))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 35))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 34))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 35))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 37))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 34))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 34))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 34))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
}
|
|
if (round_up(P, 32) == 32)
|
|
P = 32;
|
|
else
|
|
P = round_up(P, 64);
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 28))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 37))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 37))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 37))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 28))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 24))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
}
|
|
else if (Q == 256 && isHopper_)
|
|
{
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 32, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 32, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 32, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 34))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 33))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 34))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 33))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 32))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 2, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 64, 32, 4, 1, 2, 0, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 64, 32, 4, 1, 2, 0, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 64, 32, 4, 1, 2, 0, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 2, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 2, 1, 2, 1, 1, 0, Tp_, Wt_>);
|
|
}
|
|
if (round_up(P, 32) == 32)
|
|
P = 32;
|
|
else
|
|
P = round_up(P, 64);
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 32, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 32, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 32, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 34))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 33))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 36))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 33))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 33))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 2, 3, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 2, 3, 0, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 128, 64, 1, 4, 2, 0, chunk_scan_kernel<256, 64, 128, 64, 1, 4, 2, 1, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 128, 32, 2, 2, 2, 1, 4, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 64, 32, 4, 1, 2, 0, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 24))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 32, 64, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 64, 32, 4, 1, 2, 0, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 24))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 64, 2, 2, 2, 0, chunk_scan_kernel<256, 64, 64, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<256, 64, 64, 32, 4, 1, 2, 0, 4, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 22))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<256, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
}
|
|
if (Q == 128 && isHopper_ == false)
|
|
{
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 33))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 32))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 37))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 35))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 35))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 36))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
if (round_up(P, 32) == 32)
|
|
P = 32;
|
|
else
|
|
P = round_up(P, 64);
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 33))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 32))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 38))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 37))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 29))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 27))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 2, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 2, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
}
|
|
else if (Q == 128 && isHopper_)
|
|
{
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 32, 32, 4, 1, 2, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 32, 32, 4, 1, 2, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 32, 32, 4, 1, 2, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 64, 32, 4, 1, 2, 0, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 64, 32, 4, 1, 2, 0, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 24))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 64, 32, 4, 1, 2, 0, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 24))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 4, 0, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (round_up(P, 32) == 32)
|
|
P = 32;
|
|
else
|
|
P = round_up(P, 64);
|
|
#ifndef FAST_BUILD
|
|
if (P == 32 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 32, 32, 4, 1, 2, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 32, 32, 4, 1, 2, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P == 32 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 32, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 32, 32, 4, 1, 2, 0, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 32))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 64, 32, 4, 1, 2, 0, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 256 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 29))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 28))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 27))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 128 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 30))
|
|
return set(64, 128, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 128, 32, 4, 1, 2, 0, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 26))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 128, 32, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 128, 32, 2, 2, 2, 1, 8, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 256)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 64, 32, 4, 1, 2, 0, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
if (P % 64 == 0 && N >= 128)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 64, 32, 4, 1, 2, 0, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 25))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 23))
|
|
return set(64, 32, 64, 2, 2, 2, 0, chunk_scan_kernel<128, 64, 32, 64, 2, 2, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
#endif
|
|
if (P % 64 == 0 && N >= 64)
|
|
{
|
|
if (compute >= (1LL << 31))
|
|
return set(64, 64, 32, 4, 1, 2, 1, chunk_scan_hopper<128, 64, 64, 32, 4, 1, 2, 0, 2, 1, Tp_, Wt_>);
|
|
else if (compute >= (1LL << 22))
|
|
return set(64, 64, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 64, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
else
|
|
return set(64, 32, 32, 4, 1, 2, 0, chunk_scan_kernel<128, 64, 32, 32, 4, 1, 2, 1, 1, 1, Tp_, Wt_>);
|
|
}
|
|
}
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
extern GetChunkScanKernelFunc getChunkScanKernel_fp16_fp16;
|
|
extern GetChunkScanKernelFunc getChunkScanKernel_fp16_fp32;
|
|
extern GetChunkScanKernelFunc getChunkScanKernel_bf16_bf16;
|
|
extern GetChunkScanKernelFunc getChunkScanKernel_bf16_fp32;
|
|
|
|
static inline ChunkScanKernelFunc getChunkScanKernel(int B_, int L_, int H_, int P_, int G_, int N_, int Q_,
|
|
int numTokens_, int isHopper_, tensorrt_llm::common::CUDADriverWrapper* cudaDriver_, dim3* blockDims_,
|
|
dim3* threadDims_, int* sharedMem_, int* useTma_, CUtensorMap* descs_, CudaType tp_ = CT_FP16,
|
|
CudaType wt_ = CT_FP32)
|
|
{
|
|
if (tp_ == CT_FP16 && wt_ == CT_FP16)
|
|
return getChunkScanKernel_fp16_fp16(B_, L_, H_, P_, G_, N_, Q_, numTokens_, isHopper_, cudaDriver_, blockDims_,
|
|
threadDims_, sharedMem_, useTma_, descs_);
|
|
else if (tp_ == CT_FP16 && wt_ == CT_FP32)
|
|
return getChunkScanKernel_fp16_fp32(B_, L_, H_, P_, G_, N_, Q_, numTokens_, isHopper_, cudaDriver_, blockDims_,
|
|
threadDims_, sharedMem_, useTma_, descs_);
|
|
else if (tp_ == CT_BF16 && wt_ == CT_BF16)
|
|
return getChunkScanKernel_bf16_bf16(B_, L_, H_, P_, G_, N_, Q_, numTokens_, isHopper_, cudaDriver_, blockDims_,
|
|
threadDims_, sharedMem_, useTma_, descs_);
|
|
else if (tp_ == CT_BF16 && wt_ == CT_FP32)
|
|
return getChunkScanKernel_bf16_fp32(B_, L_, H_, P_, G_, N_, Q_, numTokens_, isHopper_, cudaDriver_, blockDims_,
|
|
threadDims_, sharedMem_, useTma_, descs_);
|
|
|
|
return nullptr;
|
|
}
|
|
|
|
} // namespace kernels
|
|
} // namespace tensorrt_llm
|
|
|
|
// vim: ts=2 sw=2 sts=2 et sta
|