Add AVX2 support for x86 architectures thanks to @Const-me !

2 files changed, 166 insertions, 3 deletions

diff --git a/README.md b/README.md
index 73862d8..1770441 100644
--- a/README.md
+++ b/README.md
@@ -17,6 +17,7 @@ The main goal is to run the model using 4-bit quantization on a MacBook.
 
 - Plain C/C++ implementation without dependencies
 - Apple silicon first-class citizen - optimized via Arm Neon and Accelerate framework
+- AVX2 support for x86 architectures
 - Mixed F16 / F32 precision
 - 4-bit quantization support
 - Runs on the CPU
@@ -185,9 +186,6 @@ When running the larger models, make sure you have enough disk space to store al
   In general, it seems to work, but I think it fails for unicode character support. Hopefully, someone can help with that
 - I don't know yet how much the quantization affects the quality of the generated text
 - Probably the token sampling can be improved
-- x86 quantization support [not yet ready](https://github.com/ggerganov/ggml/pull/27). Basically, you want to run this
-  on Apple Silicon. For now, on Linux and Windows you can use the F16 `ggml-model-f16.bin` model, but it will be much
-  slower.
 - The Accelerate framework is actually currently unused since I found that for tensor shapes typical for the Decoder,
   there is no benefit compared to the ARM_NEON intrinsics implementation. Of course, it's possible that I simlpy don't
   know how to utilize it properly. But in any case, you can even disable it with `LLAMA_NO_ACCELERATE=1 make` and the
diff --git a/ggml.c b/ggml.c
index bb714e2..71c3028 100644
--- a/ggml.c
+++ b/ggml.c
@@ -359,6 +359,45 @@ static const size_t CACHE_LINE_SIZE_F32 = CACHE_LINE_SIZE/sizeof(float);
 
 #define QK 32
 
+// AVX routines provided by GH user Const-me
+// ref: https://github.com/ggerganov/ggml/pull/27#issuecomment-1464934600
+#if __AVX2__
+// 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 );
+
+    // Expand bytes into uint16_t values
+    __m256i bytes = _mm256_cvtepu8_epi16( tmp );
+
+    // Unpack values into individual bytes
+    const __m256i lowMask = _mm256_set1_epi8( 0xF );
+    __m256i high = _mm256_andnot_si256( lowMask, bytes );
+    __m256i low = _mm256_and_si256( lowMask, bytes );
+    high = _mm256_slli_epi16( high, 4 );
+    bytes = _mm256_or_si256( low, high );
+    return bytes;
+}
+
+inline __m128i packNibbles( __m256i bytes )
+{
+    // Move bits within 16-bit lanes from 0000_abcd_0000_efgh into 0000_0000_abcd_efgh
+    const __m256i lowByte = _mm256_set1_epi16( 0xFF );
+    __m256i high = _mm256_andnot_si256( lowByte, bytes );
+    __m256i low = _mm256_and_si256( lowByte, bytes );
+    high = _mm256_srli_epi16( high, 4 );
+    bytes = _mm256_or_si256( low, high );
+
+    // Compress uint16_t lanes into bytes
+    __m128i r0 = _mm256_castsi256_si128( bytes );
+    __m128i r1 = _mm256_extracti128_si256( bytes, 1 );
+    return _mm_packus_epi16( r0, r1 );
+}
+#endif
+
+
 // method 5
 // blocks of QK elements
 // represented with a single float (delta) and QK/2 8-bit ints (i.e QK 4-bit signed integer factors)
@@ -414,6 +453,77 @@ void quantize_row_q4_0(const float * restrict x, void * restrict y, int k) {
 #else
 #error "not implemented for QK"
 #endif
+#elif defined(__AVX2__)
+#if QK == 32
+    for (int i = 0; i < nb; i++) {
+        // Load elements into 4 AVX vectors
+        __m256 v0 = _mm256_loadu_ps( x );
+        __m256 v1 = _mm256_loadu_ps( x + 8 );
+        __m256 v2 = _mm256_loadu_ps( x + 16 );
+        __m256 v3 = _mm256_loadu_ps( x + 24 );
+        x += 32;
+
+        // Compute max(abs(e)) for the block
+        const __m256 signBit = _mm256_set1_ps( -0.0f );
+        __m256 maxAbs = _mm256_andnot_ps( signBit, v0 );
+        maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v1 ) );
+        maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v2 ) );
+        maxAbs = _mm256_max_ps( maxAbs, _mm256_andnot_ps( signBit, v3 ) );
+
+        __m128 max4 = _mm_max_ps( _mm256_extractf128_ps( maxAbs, 1 ), _mm256_castps256_ps128( maxAbs ) );
+        max4 = _mm_max_ps( max4, _mm_movehl_ps( max4, max4 ) );
+        max4 = _mm_max_ss( max4, _mm_movehdup_ps( max4 ) );
+        const float maxScalar = _mm_cvtss_f32( max4 );
+
+        // Quantize these floats
+        const float d = maxScalar / 7.0f;
+        *(float *)pd = d;
+        pd += bs;
+        const float id = ( maxScalar != 0.0f ) ? 7.0f / maxScalar : 0.0f;
+        const __m256 mul = _mm256_set1_ps( id );
+
+        // Apply the multiplier
+        v0 = _mm256_mul_ps( v0, mul );
+        v1 = _mm256_mul_ps( v1, mul );
+        v2 = _mm256_mul_ps( v2, mul );
+        v3 = _mm256_mul_ps( v3, mul );
+
+        // Round to nearest integer
+        v0 = _mm256_round_ps( v0, _MM_ROUND_NEAREST );
+        v1 = _mm256_round_ps( v1, _MM_ROUND_NEAREST );
+        v2 = _mm256_round_ps( v2, _MM_ROUND_NEAREST );
+        v3 = _mm256_round_ps( v3, _MM_ROUND_NEAREST );
+
+        // Convert floats to integers
+        __m256i i0 = _mm256_cvtps_epi32( v0 );
+        __m256i i1 = _mm256_cvtps_epi32( v1 );
+        __m256i i2 = _mm256_cvtps_epi32( v2 );
+        __m256i i3 = _mm256_cvtps_epi32( v3 );
+
+        // Convert int32 to int16
+        i0 = _mm256_packs_epi32( i0, i1 );	// 0, 1, 2, 3,  8, 9, 10, 11,  4, 5, 6, 7, 12, 13, 14, 15
+        i2 = _mm256_packs_epi32( i2, i3 );	// 16, 17, 18, 19,  24, 25, 26, 27,  20, 21, 22, 23, 28, 29, 30, 31
+                                            // Convert int16 to int8
+        i0 = _mm256_packs_epi16( i0, i2 );	// 0, 1, 2, 3,  8, 9, 10, 11,  16, 17, 18, 19,  24, 25, 26, 27,  4, 5, 6, 7, 12, 13, 14, 15, 20, 21, 22, 23, 28, 29, 30, 31
+
+        // We got our precious signed bytes, but the order is now wrong
+        // These AVX2 pack instructions process 16-byte pieces independently
+        // The following instruction is fixing the order
+        const __m256i perm = _mm256_setr_epi32( 0, 4, 1, 5, 2, 6, 3, 7 );
+        i0 = _mm256_permutevar8x32_epi32( i0, perm );
+
+        // Apply offset to translate the range from [ -7 .. +7 ] into [ +1 .. +15 ]
+        const __m256i off = _mm256_set1_epi8( 8 );
+        i0 = _mm256_add_epi8( i0, off );
+
+        // Compress the vector into 4 bit/value, and store
+        __m128i res = packNibbles( i0 );
+        _mm_storeu_si128( ( __m128i* )pb, res );
+        pb += bs;
+    }
+#else
+#error "not implemented for QK"
+#endif
 #elif defined(__wasm_simd128__)
 #if QK == 32
     for (int i = 0; i < nb; i++) {
@@ -1285,6 +1395,61 @@ inline static void ggml_vec_dot_q4_0(const int n, float * restrict s, const void
 #else
 #error "not implemented for QK"
 #endif
+#elif defined(__AVX2__)
+#if QK == 32
+    const size_t countBlocks = nb;
+
+    // Initialize accumulator with zeros
+    __m256 acc = _mm256_setzero_ps();
+
+    // Main loop
+    for (int i = 0; i < nb; ++i) {
+        const float * d0_0 = (const float *) (pd0 + i*bs);
+        const float * d1_0 = (const float *) (pd1 + i*bs);
+
+        const uint8_t * restrict p0 = pb0 + i*bs;
+        const uint8_t * restrict p1 = pb1 + i*bs;
+
+        // Compute combined scale for the block
+        const __m256 scale = _mm256_mul_ps( _mm256_broadcast_ss( d0_0 ), _mm256_broadcast_ss( d1_0 ) );
+
+        // Load 16 bytes, and unpack 4 bit fields into bytes, making 32 bytes
+        __m256i bx = bytesFromNibbles( p0 );
+        __m256i by = bytesFromNibbles( p1 );
+
+        // Now we have a vector with bytes in [ 0 .. 15 ] interval. Offset them into [ -8 .. +7 ] interval.
+        const __m256i off = _mm256_set1_epi8( 8 );
+        bx = _mm256_sub_epi8( bx, off );
+        by = _mm256_sub_epi8( by, off );
+
+        // Sign-extend first 16 signed bytes into int16_t
+        __m256i x16 = _mm256_cvtepi8_epi16( _mm256_castsi256_si128( bx ) );
+        __m256i y16 = _mm256_cvtepi8_epi16( _mm256_castsi256_si128( by ) );
+        // Compute products of int16_t integers, add pairwise
+        __m256i i32 = _mm256_madd_epi16( x16, y16 );
+
+        // Sign-extend last 16 signed bytes into int16_t vectors
+        x16 = _mm256_cvtepi8_epi16( _mm256_extracti128_si256( bx, 1 ) );
+        y16 = _mm256_cvtepi8_epi16( _mm256_extracti128_si256( by, 1 ) );
+        // Accumulate products of int16_t integers
+        i32 = _mm256_add_epi32( i32, _mm256_madd_epi16( x16, y16 ) );
+
+        // Convert int32_t to float
+        __m256 p = _mm256_cvtepi32_ps( i32 );
+        // Apply the scale, and accumulate
+        acc = _mm256_fmadd_ps( scale, p, acc );
+    }
+
+    // Return horizontal sum of the acc vector
+    __m128 res = _mm256_extractf128_ps( acc, 1 );
+    res = _mm_add_ps( res, _mm256_castps256_ps128( acc ) );
+    res = _mm_add_ps( res, _mm_movehl_ps( res, res ) );
+    res = _mm_add_ss( res, _mm_movehdup_ps( res ) );
+
+    sumf = _mm_cvtss_f32( res );
+#else
+#error "not implemented for QK"
+#endif
 #elif defined(__wasm_simd128__)
 #if QK == 32
     // wasm simd