Q2_K_R4 (#136)

* q2_k_r4: Zen4

PP-512(LLaMA-3.1-8B) = 256 t/s

* q3_k_r4: AVX2

* q2_k_r4: AVX2

We get PP-512(LLaMA-3.1-8B) = 287 t/s.

Also cherry-picked the q3_k_r4 AVX2 adaptation that I somehow
forgot to push upstream.

* q2_k_r4: NEON

We get PP-512(LLaMA-3.1-8B) = 106.2 t/s.
TG-128 is 36.02 t/s, which is ~10% higher than q2_K_S.

* Make sure rows per thread are a multiple of 4

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>

1 files changed, 7 insertions, 0 deletions

diff --git a/ggml/src/ggml-common.h b/ggml/src/ggml-common.h
index bc34718e..61e8dfd3 100644
--- a/ggml/src/ggml-common.h
+++ b/ggml/src/ggml-common.h
@@ -276,6 +276,13 @@ typedef struct {
 } block_q2_K;
 static_assert(sizeof(block_q2_K) == 2*sizeof(ggml_half) + QK_K/16 + QK_K/4, "wrong q2_K block size/padding");
 
+typedef struct {
+    ggml_half d[8];
+    uint8_t scales[QK_K/4]; // scales and mins, quantized with 4 bits
+    uint8_t qs[QK_K];       // quants
+} block_q2_k_r4;
+static_assert(sizeof(block_q2_k_r4) == 8*sizeof(ggml_half) + QK_K/4 + QK_K, "wrong q2_k_r4 block size/padding");
+
 // 3-bit quantization
 // weight is represented as x = a * q
 // 16 blocks of 16 elements each