zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-10-22 11:14:32 +08:00
Code Issues Actions Packages Projects Releases Wiki Activity

Update pre-commit hooks (#984)

Browse Source
This commit is contained in:
Nripesh Niketan
2024-04-11 18:27:53 +04:00
committed by GitHub
parent 12d4507ee3
commit ffff671273
10 changed files with 37 additions and 44 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_l.h
View File

@@ -394,7 +394,7 @@ struct Conv2DWeightBlockLoader {
const constant ImplicitGemmConv2DParams* gemm_params_,
uint simd_group_id [[simdgroup_index_in_threadgroup]],
uint simd_lane_id [[thread_index_in_simdgroup]])
: src_ld(params_->wt_strides[0]),
: src_ld(params_ -> wt_strides[0]),
thread_idx(simd_group_id * 32 + simd_lane_id),
bi(thread_idx / TCOLS),
bj(vec_size * (thread_idx % TCOLS)),

2
mlx/backend/metal/kernels/steel/conv/loaders/loader_channel_n.h
View File

@@ -244,7 +244,7 @@ struct Conv2DWeightBlockLoaderSmallChannels {
const constant ImplicitGemmConv2DParams* gemm_params_,
uint simd_group_id [[simdgroup_index_in_threadgroup]],
uint simd_lane_id [[thread_index_in_simdgroup]])
: src_ld(params_->wt_strides[0]),
: src_ld(params_ -> wt_strides[0]),
thread_idx(simd_group_id * 32 + simd_lane_id),
bi(thread_idx / TCOLS),
bj(vec_size * (thread_idx % TCOLS)),

2
mlx/backend/metal/kernels/steel/conv/loaders/loader_general.h
View File

@@ -220,7 +220,7 @@ struct Conv2DWeightBlockLoaderGeneral {
const short base_ww_,
uint simd_group_id [[simdgroup_index_in_threadgroup]],
uint simd_lane_id [[thread_index_in_simdgroup]])
: src_ld(params_->wt_strides[0]),
: src_ld(params_ -> wt_strides[0]),
thread_idx(simd_group_id * 32 + simd_lane_id),
bi(thread_idx / TCOLS),
bj(vec_size * (thread_idx % TCOLS)),

25
mlx/backend/metal/matmul.cpp
View File

@@ -197,8 +197,8 @@ inline auto collapse_batches(const array& a, const array& b) {
std::vector<int> B_bshape{b.shape().begin(), b.shape().end() - 2};
if (A_bshape != B_bshape) {
std::ostringstream msg;
msg << "[matmul] Got matrices with incorrectly broadcasted shapes: "
<< "A " << a.shape() << ", B " << b.shape() << ".";
msg << "[matmul] Got matrices with incorrectly broadcasted shapes: " << "A "
<< a.shape() << ", B " << b.shape() << ".";
throw std::runtime_error(msg.str());
}
@@ -227,9 +227,8 @@ inline auto collapse_batches(const array& a, const array& b, const array& c) {
std::vector<int> C_bshape{c.shape().begin(), c.shape().end() - 2};
if (A_bshape != B_bshape || A_bshape != C_bshape) {
std::ostringstream msg;
msg << "[addmm] Got matrices with incorrectly broadcasted shapes: "
<< "A " << a.shape() << ", B " << b.shape() << ", B " << c.shape()
<< ".";
msg << "[addmm] Got matrices with incorrectly broadcasted shapes: " << "A "
<< a.shape() << ", B " << b.shape() << ", B " << c.shape() << ".";
throw std::runtime_error(msg.str());
}
@@ -332,8 +331,8 @@ void steel_matmul(
<< (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
<< type_to_name(C_split) << "_bm" << bm << "_bn" << bn << "_bk" << bk
<< "_wm" << wm << "_wn" << wn << "_MN_"
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
<< "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned";
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
<< ((K % bk == 0) ? "t" : "n") << "aligned";
// Encode and dispatch gemm kernel
auto& compute_encoder = d.get_command_encoder(s.index);
@@ -422,8 +421,8 @@ void steel_matmul(
<< (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
<< type_to_name(out) << "_bm" << bm << "_bn" << bn << "_bk" << bk
<< "_wm" << wm << "_wn" << wn << "_MN_"
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
<< "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned";
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
<< ((K % bk == 0) ? "t" : "n") << "aligned";
// Encode and dispatch kernel
auto& compute_encoder = d.get_command_encoder(s.index);
@@ -903,8 +902,8 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
<< (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
<< type_to_name(C_split) << "_bm" << bm << "_bn" << bn << "_bk" << bk
<< "_wm" << wm << "_wn" << wn << "_MN_"
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
<< "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned";
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
<< ((K % bk == 0) ? "t" : "n") << "aligned";
// Encode and dispatch gemm kernel
auto& compute_encoder = d.get_command_encoder(s.index);
@@ -992,8 +991,8 @@ void AddMM::eval_gpu(const std::vector<array>& inputs, array& out) {
<< (transpose_b ? 't' : 'n') << "_" << type_to_name(a) << "_"
<< type_to_name(out) << "_bm" << bm << "_bn" << bn << "_bk" << bk
<< "_wm" << wm << "_wn" << wn << "_MN_"
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned"
<< "_K_" << ((K % bk == 0) ? "t" : "n") << "aligned"
<< ((M % bm == 0 && N % bn == 0) ? "t" : "n") << "aligned" << "_K_"
<< ((K % bk == 0) ? "t" : "n") << "aligned"
<< ((alpha_ == 1. && beta_ == 1.) ? "_add" : "_axpby");
// Encode and dispatch kernel