Enable q6_0 for flash attention (#101)

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

11 files changed, 120 insertions, 16 deletions

diff --git a/Makefile b/Makefile
index ae636f50..a7a53b1a 100644
--- a/Makefile
+++ b/Makefile
@@ -600,6 +600,8 @@ else
 	OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*f16-f16.cu))
 	OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*q8_0-iq4_nl.cu))
 	OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*:iq4_nl-iq4_nl.cu))
+	OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*:q6_0-q5_0.cu))
+	OBJ_CUDA_TMPL += $(patsubst %.cu,%.o,$(wildcard ggml/src/ggml-cuda/template-instances/fattn-vec*:q8_0-q6_0.cu))
 endif # GGML_CUDA_FA_ALL_QUANTS
 
 ifdef GGML_CUDA
diff --git a/ggml/src/CMakeLists.txt b/ggml/src/CMakeLists.txt
index eb6d457c..da1746c8 100644
--- a/ggml/src/CMakeLists.txt
+++ b/ggml/src/CMakeLists.txt
@@ -332,6 +332,10 @@ if (GGML_CUDA)
             list(APPEND GGML_SOURCES_CUDA ${SRCS})
             file(GLOB   SRCS "ggml-cuda/template-instances/fattn-vec*iq4_nl-iq4_nl.cu")
             list(APPEND GGML_SOURCES_CUDA ${SRCS})
+            file(GLOB   SRCS "ggml-cuda/template-instances/fattn-vec*q6_0-q5_0.cu")
+            list(APPEND GGML_SOURCES_CUDA ${SRCS})
+            file(GLOB   SRCS "ggml-cuda/template-instances/fattn-vec*q8_0-q6_0.cu")
+            list(APPEND GGML_SOURCES_CUDA ${SRCS})
         endif()
 
         list(APPEND GGML_CDEF_PUBLIC GGML_USE_CUDA)
diff --git a/ggml/src/ggml-cuda/fattn-common.cuh b/ggml/src/ggml-cuda/fattn-common.cuh
index 1984c838..0a664dbd 100644
--- a/ggml/src/ggml-cuda/fattn-common.cuh
+++ b/ggml/src/ggml-cuda/fattn-common.cuh
@@ -277,6 +277,49 @@ static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q5_1(
     return sum;
 }
 
+template<typename T, int D>
+static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q6_0(
+    const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
+
+    const block_q6_0 * K_q6_0 = (const block_q6_0 *) K_c;
+    GGML_UNUSED(Q_v);
+
+    T sum = 0.0f;
+
+#pragma unroll
+    for (int k_KQ_0 = 0; k_KQ_0 < D/sizeof(int); k_KQ_0 += WARP_SIZE) {
+        const int k_KQ = k_KQ_0 + threadIdx.x;
+
+        const int ib    = k_KQ /  QI8_1;
+        const int iqs4  = k_KQ %  QI6_0;  // 0...3
+        const int shift = k_KQ & (QI8_1/2);
+
+        const int vh = (get_int_b2(K_q6_0[ib].qh, iqs4%2) >> (4*(iqs4/2) + shift/2)) & 0x03030303;
+        const int vl = (get_int_b2(K_q6_0[ib].qs, iqs4) >> shift) & 0x0F0F0F0F;
+        const int v  = vl | (vh << 4);
+
+        const int u = Q_q8[k_KQ_0/WARP_SIZE];
+
+        const int sumi = ggml_cuda_dp4a(v, u, 0);
+
+#ifdef FP16_AVAILABLE
+        if (std::is_same<T, half>::value) {
+            const half2  * Q_ds = (const half2  *) Q_ds_v;
+
+            const half2 sum2 = __half2half2(K_q6_0[ib].d) * Q_ds[k_KQ_0/WARP_SIZE];
+            sum += (T) (((half) sumi)*__low2half(sum2) - __high2half(sum2)*__float2half(4.0f)) /* *32/QI8_1 == 4 */;
+        } else
+#endif // FP16_AVAILABLE
+        {
+            const float2 * Q_ds = (const float2 *) Q_ds_v;
+
+            sum += (T) (__half2float(K_q6_0[ib].d) * (sumi*Q_ds[k_KQ_0/WARP_SIZE].x - (32/QI8_1)*Q_ds[k_KQ_0/WARP_SIZE].y));
+        }
+    }
+
+    return sum;
+}
+
 template <typename T, int D>
 static __device__ __forceinline__ T vec_dot_fattn_vec_KQ_q8_0(
     const char * __restrict__ K_c, const void * __restrict__ Q_v, const int * __restrict__ Q_q8, const void * __restrict__ Q_ds_v) {
@@ -511,6 +554,30 @@ static __device__ __forceinline__ T dequantize_1_q5_1(const void * __restrict__
 }
 
 template <typename T>
+static __device__ __forceinline__ T dequantize_1_q6_0(const void * __restrict__ vx, const int64_t i) {
+    const block_q6_0 * x = (const block_q6_0 *) vx;
+
+    const int64_t ib    =  i  /  QK6_0;
+    const int     idq   =  i  %  QK6_0;
+    const int     iqs   =  i  % (QK6_0/2);
+    const int     shift = idq / (QK6_0/2);
+    //const int     shift = (i % QK6_0) / (QK6_0/2);
+
+    const T   d  = x[ib].d;
+    const int ql = x[ib].qs[iqs] >> 4*shift;
+    const int qh = x[ib].qh[idq%(QK6_0/4)] >> (4*((idq/(QK6_0/4))%2) + 2*shift);
+    const int q  = ((ql & 0x0f) | ((qh & 0x03) << 4)) - 32;
+
+#ifdef FP16_AVAILABLE
+    if (std::is_same<T, half>::value) {
+        return ((half) d)*((half) q);
+    }
+#endif // FP16_AVAILABLE
+
+    return ((float) d)*((float) q);
+}
+
+template <typename T>
 static __device__ __forceinline__ T dequantize_1_q8_0(const void * __restrict__ vx, const int64_t i) {
     const block_q8_0 * x = (const block_q8_0 *) vx;
 
@@ -543,6 +610,7 @@ constexpr __device__ vec_dot_KQ_f16_t get_vec_dot_KQ_f16(ggml_type type_K) {
            type_K == GGML_TYPE_IQ4_NL ? vec_dot_fattn_vec_KQ_iq4_nl<half, D> :
            type_K == GGML_TYPE_Q5_0   ? vec_dot_fattn_vec_KQ_q5_0<half, D>   :
            type_K == GGML_TYPE_Q5_1   ? vec_dot_fattn_vec_KQ_q5_1<half, D>   :
+           type_K == GGML_TYPE_Q6_0   ? vec_dot_fattn_vec_KQ_q6_0<half, D>   :
            type_K == GGML_TYPE_Q8_0   ? vec_dot_fattn_vec_KQ_q8_0<half, D>   :
            type_K == GGML_TYPE_F16    ? vec_dot_fattn_vec_KQ_f16<half, D>    :
            nullptr;
@@ -555,6 +623,7 @@ constexpr __device__ vec_dot_KQ_f32_t get_vec_dot_KQ_f32(ggml_type type_K) {
            type_K == GGML_TYPE_IQ4_NL ? vec_dot_fattn_vec_KQ_iq4_nl<float, D> :
            type_K == GGML_TYPE_Q5_0   ? vec_dot_fattn_vec_KQ_q5_0<float, D>   :
            type_K == GGML_TYPE_Q5_1   ? vec_dot_fattn_vec_KQ_q5_1<float, D>   :
+           type_K == GGML_TYPE_Q6_0   ? vec_dot_fattn_vec_KQ_q6_0<float, D>   :
            type_K == GGML_TYPE_Q8_0   ? vec_dot_fattn_vec_KQ_q8_0<float, D>   :
            type_K == GGML_TYPE_F16    ? vec_dot_fattn_vec_KQ_f16<float, D>    :
            nullptr;
@@ -565,6 +634,7 @@ constexpr __device__ dequantize_1_f16_t get_dequantize_1_f16(ggml_type type_V) {
            type_V == GGML_TYPE_Q4_1   ? dequantize_1_q4_1<half> :
            type_V == GGML_TYPE_Q5_0   ? dequantize_1_q5_0<half> :
            type_V == GGML_TYPE_Q5_1   ? dequantize_1_q5_1<half> :
+           type_V == GGML_TYPE_Q6_0   ? dequantize_1_q6_0<half> :
            type_V == GGML_TYPE_Q8_0   ? dequantize_1_q8_0<half> :
            type_V == GGML_TYPE_IQ4_NL ? dequantize_1_iq4_nl<half> :
            type_V == GGML_TYPE_F16    ? dequantize_1_f16<half> :
@@ -576,6 +646,7 @@ constexpr __device__ dequantize_1_f32_t get_dequantize_1_f32(ggml_type type_V) {
            type_V == GGML_TYPE_Q4_1   ? dequantize_1_q4_1<float> :
            type_V == GGML_TYPE_Q5_0   ? dequantize_1_q5_0<float> :
            type_V == GGML_TYPE_Q5_1   ? dequantize_1_q5_1<float> :
+           type_V == GGML_TYPE_Q6_0   ? dequantize_1_q6_0<float> :
            type_V == GGML_TYPE_Q8_0   ? dequantize_1_q8_0<float> :
            type_V == GGML_TYPE_IQ4_NL ? dequantize_1_iq4_nl<float> :
            type_V == GGML_TYPE_F16    ? dequantize_1_f16<float> :
@@ -635,11 +706,13 @@ static void on_no_fattn_vec_case(const int D) {
     } else if (D == 128) {
         fprintf(stderr, "Unsupported KV type combination for head_size 128.\n");
         fprintf(stderr, "Supported combinations:\n");
-        fprintf(stderr, "  - K == q4_0,   V == q4_0,   4.50 BPV\n");
-        fprintf(stderr, "  - K == iq4_nl, V == iq4_nl, 4.50 BPV\n");
-        fprintf(stderr, "  - K == q8_0,   V == iq4_nl, 6.50 BPV\n");
-        fprintf(stderr, "  - K == q8_0,   V == q8_0,   8.50 BPV\n");
-        fprintf(stderr, "  - K == f16,    V == f16,   16.00 BPV\n");
+        fprintf(stderr, "  - K == q4_0,   V == q4_0,   4.5 BPV\n");
+        fprintf(stderr, "  - K == iq4_nl, V == iq4_nl, 4.5 BPV\n");
+        fprintf(stderr, "  - K == q6_0,   V == q5_0,   6.0 BPV\n");
+        fprintf(stderr, "  - K == q8_0,   V == iq4_nl, 6.5 BPV\n");
+        fprintf(stderr, "  - K == q8_0,   V == q6_0,   7.5 BPV\n");
+        fprintf(stderr, "  - K == q8_0,   V == q8_0,   8.5 BPV\n");
+        fprintf(stderr, "  - K == f16,    V == f16,   16.0 BPV\n");
         fprintf(stderr, "Compile with GGML_CUDA_FA_ALL_QUANTS for all combinations of q4_0, q4_1, iq4_nl, q5_0, q5_1, q8_0, and f16.\n");
         GGML_ABORT("fatal error");
     } else {
diff --git a/ggml/src/ggml-cuda/fattn.cu b/ggml/src/ggml-cuda/fattn.cu
index c5540161..1dfb24f9 100644
--- a/ggml/src/ggml-cuda/fattn.cu
+++ b/ggml/src/ggml-cuda/fattn.cu
@@ -208,13 +208,11 @@ static void ggml_cuda_flash_attn_ext_vec_f16(ggml_backend_cuda_context & ctx, gg
 
     FATTN_VEC_F16_CASE(256, GGML_TYPE_F16, GGML_TYPE_F16)
 
-    //FATTN_VEC_F16_CASE( 64, GGML_TYPE_F16,  GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE(128, GGML_TYPE_F16,  GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE(256, GGML_TYPE_F16,  GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE( 64, GGML_TYPE_Q8_0, GGML_TYPE_IQ4_NL)
     FATTN_VEC_F16_CASE(128, GGML_TYPE_IQ4_NL, GGML_TYPE_IQ4_NL)
     FATTN_VEC_F16_CASE(128, GGML_TYPE_Q8_0,   GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE(256, GGML_TYPE_Q8_0, GGML_TYPE_IQ4_NL)
+    FATTN_VEC_F16_CASE(128, GGML_TYPE_Q6_0,   GGML_TYPE_Q5_0)
+    FATTN_VEC_F16_CASE(128, GGML_TYPE_Q6_0,   GGML_TYPE_Q6_0)
+    FATTN_VEC_F16_CASE(128, GGML_TYPE_Q8_0,   GGML_TYPE_Q6_0)
 #else
     FATTN_VEC_F16_CASE(128, GGML_TYPE_Q4_0, GGML_TYPE_Q4_0)
 
@@ -224,13 +222,10 @@ static void ggml_cuda_flash_attn_ext_vec_f16(ggml_backend_cuda_context & ctx, gg
     FATTN_VEC_F16_CASE(128, GGML_TYPE_F16, GGML_TYPE_F16)
     FATTN_VEC_F16_CASE(256, GGML_TYPE_F16, GGML_TYPE_F16)
 
-    //FATTN_VEC_F16_CASE( 64, GGML_TYPE_F16,  GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE(128, GGML_TYPE_F16,  GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE(256, GGML_TYPE_F16,  GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE( 64, GGML_TYPE_Q8_0, GGML_TYPE_IQ4_NL)
     FATTN_VEC_F16_CASE(128, GGML_TYPE_IQ4_NL, GGML_TYPE_IQ4_NL)
     FATTN_VEC_F16_CASE(128, GGML_TYPE_Q8_0,   GGML_TYPE_IQ4_NL)
-    //FATTN_VEC_F16_CASE(256, GGML_TYPE_Q8_0, GGML_TYPE_IQ4_NL)
+    FATTN_VEC_F16_CASE(128, GGML_TYPE_Q6_0,   GGML_TYPE_Q5_0)
+    FATTN_VEC_F16_CASE(128, GGML_TYPE_Q8_0,   GGML_TYPE_Q6_0)
 #endif // GGML_CUDA_FA_ALL_QUANTS
 
     on_no_fattn_vec_case(Q->ne[0]);
diff --git a/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q6_0-q5_0.cu b/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q6_0-q5_0.cu
new file mode 100644
index 00000000..d1ecb548
--- /dev/null
+++ b/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q6_0-q5_0.cu
@@ -0,0 +1,5 @@
+// This file has been autogenerated by generate_cu_files.py, do not edit manually.
+
+#include "../fattn-vec-f16.cuh"
+
+DECL_FATTN_VEC_F16_CASE(128, GGML_TYPE_Q6_0, GGML_TYPE_Q5_0);
diff --git a/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q6_0-q6_0.cu b/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q6_0-q6_0.cu
new file mode 100644
index 00000000..e605e7a6
--- /dev/null
+++ b/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q6_0-q6_0.cu
@@ -0,0 +1,5 @@
+// This file has been autogenerated by generate_cu_files.py, do not edit manually.
+
+#include "../fattn-vec-f16.cuh"
+
+DECL_FATTN_VEC_F16_CASE(128, GGML_TYPE_Q6_0, GGML_TYPE_Q6_0);
diff --git a/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q6_0.cu b/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q6_0.cu
new file mode 100644
index 00000000..80539daf
--- /dev/null
+++ b/ggml/src/ggml-cuda/template-instances/fattn-vec-f16-instance-hs128-q8_0-q6_0.cu
@@ -0,0 +1,5 @@
+// This file has been autogenerated by generate_cu_files.py, do not edit manually.
+
+#include "../fattn-vec-f16.cuh"
+
+DECL_FATTN_VEC_F16_CASE(128, GGML_TYPE_Q8_0, GGML_TYPE_Q6_0);
diff --git a/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q6_0-q5_0.cu b/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q6_0-q5_0.cu
new file mode 100644
index 00000000..78eca1c2
--- /dev/null
+++ b/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q6_0-q5_0.cu
@@ -0,0 +1,5 @@
+// This file has been autogenerated by generate_cu_files.py, do not edit manually.
+
+#include "../fattn-vec-f32.cuh"
+
+DECL_FATTN_VEC_F32_CASE(128, GGML_TYPE_Q6_0, GGML_TYPE_Q5_0);
diff --git a/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q6_0-q6_0.cu b/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q6_0-q6_0.cu
new file mode 100644
index 00000000..fa16ddbc
--- /dev/null
+++ b/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q6_0-q6_0.cu
@@ -0,0 +1,5 @@
+// This file has been autogenerated by generate_cu_files.py, do not edit manually.
+
+#include "../fattn-vec-f32.cuh"
+
+DECL_FATTN_VEC_F32_CASE(128, GGML_TYPE_Q6_0, GGML_TYPE_Q6_0);
diff --git a/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q6_0.cu b/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q6_0.cu
new file mode 100644
index 00000000..d25d482b
--- /dev/null
+++ b/ggml/src/ggml-cuda/template-instances/fattn-vec-f32-instance-hs128-q8_0-q6_0.cu
@@ -0,0 +1,5 @@
+// This file has been autogenerated by generate_cu_files.py, do not edit manually.
+
+#include "../fattn-vec-f32.cuh"
+
+DECL_FATTN_VEC_F32_CASE(128, GGML_TYPE_Q8_0, GGML_TYPE_Q6_0);
diff --git a/ggml/src/ggml-cuda/template-instances/generate_cu_files.py b/ggml/src/ggml-cuda/template-instances/generate_cu_files.py
index 1186112e..4f7489d5 100755
--- a/ggml/src/ggml-cuda/template-instances/generate_cu_files.py
+++ b/ggml/src/ggml-cuda/template-instances/generate_cu_files.py
@@ -3,7 +3,7 @@
 from glob import glob
 import os
 
-TYPES_KV = ["GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0", "GGML_TYPE_IQ4_NL", "GGML_TYPE_F16"]
+TYPES_KV = ["GGML_TYPE_Q4_0", "GGML_TYPE_Q4_1", "GGML_TYPE_Q5_0", "GGML_TYPE_Q5_1", "GGML_TYPE_Q8_0", "GGML_TYPE_IQ4_NL", "GGML_TYPE_Q6_0", "GGML_TYPE_F16"]
 
 SOURCE_FATTN_VEC = """// This file has been autogenerated by generate_cu_files.py, do not edit manually.