Const-me

#include <stdlib.h>
#include <stdio.h>
#include <random>
#include <vector>
#include <unordered_map>
#include <algorithm>
#include <optional>
#include <intrin.h>
#include <inttypes.h>

Const-me

using System.Linq.Expressions;
using System.Reflection;
using System.Runtime.CompilerServices;

static class ReflectTest
{
	/// <summary>Generic method to call</summary>
	public static T GetValue<T>( T value )
	{
		return value;

Const-me

#include <stdlib.h>
#include <vector>
#include <intrin.h>
#include <stdint.h>
#include <inttypes.h>

std::vector<char> makeTestVector( bool random )
{
	std::vector<char> result;
	result.resize( 1024 * 16 );

Const-me

// ==== AVX2 decompressor for Q4_0 and Q4_1 compressed blocks ====
#include <array>
#include <immintrin.h>
#include <assert.h>
#include <float.h>

// Unpack 32 4-bit fields into 32 bytes
// The output vector contains 32 bytes, each one in [ 0 .. 15 ] interval
inline __m256i bytesFromNibbles( const uint8_t* rsi )
{

Const-me

// ==== AVX2 decompressor for Q4_0 and Q4_1 compressed blocks ====
#include <array>
#include <immintrin.h>
#include <assert.h>
#include <float.h>

// Unpack 32 4-bit fields into 32 bytes
// The output vector contains 32 bytes, each one in [ 0 .. 15 ] interval
inline __m256i bytesFromNibbles( const uint8_t* rsi )
{

Const-me

// ==== AVX2 decompressor for Q4_0 and Q4_1 compressed blocks ====
#include <array>
#include <immintrin.h>
#include <assert.h>
#include <float.h>

// Unpack 32 4-bit fields into 32 bytes
// The output vector contains 32 bytes, each one in [ 0 .. 15 ] interval
inline __m256i bytesFromNibbles( const uint8_t* rsi )
{

Const-me

// ==== AVX2 decompressor for Q4_0 and Q4_1 compressed blocks ====
#include <array>
#include <immintrin.h>

// Unpack 32 4-bit fields into 32 bytes
// The output vector contains 32 bytes, each one in [ 0 .. 15 ] interval
inline __m256i bytesFromNibbles( const uint8_t* rsi )
{
	// Load 16 bytes from memory
	__m128i tmp = _mm_loadu_si128( ( const __m128i* )rsi );

Const-me

#include <immintrin.h>

// Compute product of width*16 column major matrix by vector of length `width`,
// the result is a vector of length 16
// BTW, according to godbolt.org, gcc does better than clang for this code.
void multiplyInner_avx16( const float* mat, const float* vec, size_t width, float* rdi )
{
	// Using 4 accumulators per row, 4*16=64 scalars in 8 AVX vectors
	__m256 a00 = _mm256_setzero_ps();
	__m256 a01 = _mm256_setzero_ps();

Const-me

#include <immintrin.h>

// Compute product of width*16 column major matrix by vector of length `width`,
// the result is a vector of length 16
void multiplyInner_avx16( const float* mat, const float* vec, size_t width, float* rdi )
{
	// Using 2 accumulators per row to workaround data dependency on the accumulators

	// Initialize the accumulators
	__m256 a00 = _mm256_setzero_ps();

Const-me

using System.Runtime.InteropServices;
using Whisper;

/// <summary>This class demonstrates how to implement iAudioBuffer COM interface in C#, to supply audio samples produced by managed code</summary>
/// <remarks>The library requires these samples to be <c>float</c> numbers @ 16 kHz sample rate</remarks>
sealed class AudioBuffer: iAudioBuffer
{
	void IDisposable.Dispose()
	{
		free();