[auto] Sync version 2304231232.0.0+llamacpp-release.master-53c8434

KerfuffleV2 · Apr 23, 2023 · 53c4b18 · 53c4b18
1 parent 9190d3c
commit 53c4b18
Show file tree

Hide file tree

Showing 5 changed files with 9 additions and 13 deletions.
diff --git a/Cargo.toml b/Cargo.toml
@@ -1,6 +1,6 @@
 [package]
 name = "ggml-sys-bleedingedge"
-version = "2304230824.0.0+llamacpp-release.master-c9e2c26"
+version = "2304231232.0.0+llamacpp-release.master-53c8434"
 description = "Bleeding edge low-level bindings to GGML. "
 repository = "https://github.com/KerfuffleV2/ggml-bleeding-edge"
 keywords = ["deep-learning", "machine-learning", "tensors", "ggml", "ml"]

diff --git a/VERSION.txt b/VERSION.txt
@@ -1 +1 @@
-2304230824.0.0+llamacpp-release.master-c9e2c26
+2304231232.0.0+llamacpp-release.master-53c8434
diff --git a/ggml-src/ggml.c b/ggml-src/ggml.c
@@ -2947,16 +2947,16 @@ static void ggml_vec_dot_q4_3_q8_0(const int n, float * restrict s, const void *
 #elif defined(__AVX2__)
     // Initialize accumulator with zeros
     __m256 acc = _mm256_setzero_ps();
+    float summs = 0.0f;
 
     // Main loop
     for (int i = 0; i < nb; i++) {
         const __m128 d0 = _mm_set1_ps(GGML_FP16_TO_FP32(x[2*i + 0].d));
         const __m128 d1 = _mm_set1_ps(GGML_FP16_TO_FP32(x[2*i + 1].d));
         const __m256 dx = _mm256_set_m128(d1, d0);
 
-        const __m128 m0 = _mm_set1_ps(GGML_FP16_TO_FP32(x[2*i + 0].m));
-        const __m128 m1 = _mm_set1_ps(GGML_FP16_TO_FP32(x[2*i + 1].m));
-        const __m256 mx = _mm256_set_m128(m1, m0);
+        summs += GGML_FP16_TO_FP32(x[2*i + 0].m) * y[i].s0
+               + GGML_FP16_TO_FP32(x[2*i + 1].m) * y[i].s1;
 
         const __m128i bx0 = bytes_from_nibbles_16(x[2*i + 0].qs);
         const __m128i bx1 = bytes_from_nibbles_16(x[2*i + 1].qs);
@@ -2965,16 +2965,12 @@ static void ggml_vec_dot_q4_3_q8_0(const int n, float * restrict s, const void *
         const __m256 dy = _mm256_broadcast_ss(&y[i].d);
         const __m256i by = _mm256_loadu_si256((const __m256i *)y[i].qs);
 
-        const __m256i syi = _mm256_maddubs_epi16(_mm256_set1_epi8(1), by);
-        const __m256 syf = sum_i16_pairs_float(syi);
-
         const __m256 q = mul_sum_i8_pairs_float(bx, by);
 
-        const __m256 sxy = _mm256_fmadd_ps(q, dx, _mm256_mul_ps(mx, syf));
-        acc = _mm256_fmadd_ps(sxy, dy, acc);
+        acc = _mm256_fmadd_ps(q, _mm256_mul_ps(dx, dy), acc);
     }
 
-    *s = hsum_float_8(acc);
+    *s = hsum_float_8(acc) + summs;
 #else
     // scalar
     float sumf = 0.0;

diff --git a/ggml-tag-current.txt b/ggml-tag-current.txt
@@ -1 +1 @@
-master-c9e2c26
+master-53c8434
diff --git a/ggml-tag-previous.txt b/ggml-tag-previous.txt
@@ -1 +1 @@
-master-0e018fe
+master-c9e2c26
Original file line number	Diff line number	Diff line change
		@@ -1 +1 @@
		2304230824.0.0+llamacpp-release.master-c9e2c26
		2304231232.0.0+llamacpp-release.master-53c8434