zhangyiss/mlx
1
0
Fork 0
mirror of https://github.com/ml-explore/mlx.git synced 2025-12-16 01:49:05 +08:00
Code Issues Actions 1 Packages Projects Releases Wiki Activity

redesign for faster cpu/gpu synch (#1869)

Browse Source 
* redesign for faster cpu/gpu synch

* load + more async CPU

* use command encoder API and move more ops to use it

* make fence back-end generic + CPU only fence

* faster build

* fix async eval

* fixes + handle temporaries

* fix / improve cpu conv

* remove unused status, fix siblings

* fix extensions

* fix

* fix no cpu build

* format

* comments

* fix perf regression, remove unecessary abort

* fix events, task limit cpu

* fix waiting

* fix donation / temporaries in normalization
This commit is contained in:
Awni Hannun
2025-03-06 19:23:38 -08:00
committed by GitHub
parent 5245f12a46
commit c4230747a1
103 changed files with 5013 additions and 3873 deletions
Download Patch File Download Diff File Expand all files Collapse all files

69
mlx/backend/cpu/arange.h
View File

@@ -2,76 +2,27 @@
#pragma once
#include "mlx/allocator.h"
#include "mlx/array.h"
#include "mlx/backend/cpu/encoder.h"
namespace mlx::core {
namespace {
template <typename T>
void arange(T start, T next, array& out, size_t size) {
void arange(T start, T next, array& out, size_t size, Stream stream) {
auto ptr = out.data<T>();
auto step_size = next - start;
for (int i = 0; i < size; ++i) {
ptr[i] = start;
start += step_size;
}
auto& encoder = cpu::get_command_encoder(stream);
encoder.set_output_array(out);
encoder.dispatch([ptr, start, step_size, size]() mutable {
for (int i = 0; i < size; ++i) {
ptr[i] = start;
start += step_size;
}
});
}
} // namespace
void arange(
const std::vector<array>& inputs,
array& out,
double start,
double step) {
assert(inputs.size() == 0);
out.set_data(allocator::malloc_or_wait(out.nbytes()));
switch (out.dtype()) {
case bool_:
throw std::runtime_error("Bool type unsupported for arange.");
break;
case uint8:
arange<uint8_t>(start, start + step, out, out.size());
break;
case uint16:
arange<uint16_t>(start, start + step, out, out.size());
break;
case uint32:
arange<uint32_t>(start, start + step, out, out.size());
break;
case uint64:
arange<uint64_t>(start, start + step, out, out.size());
break;
case int8:
arange<int8_t>(start, start + step, out, out.size());
break;
case int16:
arange<int16_t>(start, start + step, out, out.size());
break;
case int32:
arange<int32_t>(start, start + step, out, out.size());
break;
case int64:
arange<int64_t>(start, start + step, out, out.size());
break;
case float16:
arange<float16_t>(start, start + step, out, out.size());
break;
case float32:
arange<float>(start, start + step, out, out.size());
break;
case float64:
arange<double>(start, start + step, out, out.size());
break;
case bfloat16:
arange<bfloat16_t>(start, start + step, out, out.size());
break;
case complex64:
arange<complex64_t>(start, start + step, out, out.size());
break;
}
}
} // namespace mlx::core