mirror of
https://github.com/ml-explore/mlx.git
synced 2025-06-30 13:41:14 +08:00
e7a2a3dcd1
* fix jit reduce for RMS norm * make strides a single buffer * better eval error message * fix compiling with inf and bf16 * fix cpu compile with bf16
36 lines
620 B
Bash
36 lines
620 B
Bash
#!/bin/bash
|
|
#
|
|
# This script generates a C++ function that provides the CPU
|
|
# code for use with kernel generation.
|
|
#
|
|
# Copyright © 2023-24 Apple Inc.
|
|
|
|
|
|
OUTPUT_FILE=$1
|
|
GCC=$2
|
|
SRCDIR=$3
|
|
CLANG=$4
|
|
|
|
if [ "$CLANG" = "TRUE" ]; then
|
|
read -r -d '' INCLUDES <<- EOM
|
|
#include <cmath>
|
|
#include <complex>
|
|
#include <cstdint>
|
|
#include <vector>
|
|
EOM
|
|
|
|
fi
|
|
|
|
CONTENT=$($GCC -I $SRCDIR -E $SRCDIR/mlx/backend/common/compiled_preamble.h 2>/dev/null)
|
|
|
|
cat << EOF > "$OUTPUT_FILE"
|
|
const char* get_kernel_preamble() {
|
|
return R"preamble(
|
|
$INCLUDES
|
|
$CONTENT
|
|
using namespace mlx::core;
|
|
using namespace mlx::core::detail;
|
|
)preamble";
|
|
}
|
|
EOF
|