whisper.rn 0.4.0-rc.10 → 0.4.0-rc.11

package/cpp/ggml-common.h

@@ -6,7 +6,20 @@
 typedef uint16_t wsp_ggml_half;
 typedef uint32_t wsp_ggml_half2;
 
-#define WSP_GGML_COMMON_AGGR
+#define WSP_GGML_COMMON_AGGR_U
+#define WSP_GGML_COMMON_AGGR_S
+
+#define WSP_GGML_COMMON_DECL
+#elif defined(WSP_GGML_COMMON_DECL_CPP)
+#include <cstdint>
+
+typedef uint16_t wsp_ggml_half;
+typedef uint32_t wsp_ggml_half2;
+
+// std-c++ allow anonymous unions but some compiler warn on it
+#define WSP_GGML_COMMON_AGGR_U data
+// std-c++ do not allow it.
+#define WSP_GGML_COMMON_AGGR_S data
 
 #define WSP_GGML_COMMON_DECL
 #elif defined(WSP_GGML_COMMON_DECL_METAL)
@@ -15,7 +28,8 @@ typedef uint32_t wsp_ggml_half2;
 typedef half  wsp_ggml_half;
 typedef half2 wsp_ggml_half2;
 
-#define WSP_GGML_COMMON_AGGR
+#define WSP_GGML_COMMON_AGGR_U
+#define WSP_GGML_COMMON_AGGR_S
 
 #define WSP_GGML_COMMON_DECL
 #elif defined(WSP_GGML_COMMON_DECL_CUDA)
@@ -29,7 +43,8 @@ typedef half2 wsp_ggml_half2;
 typedef half  wsp_ggml_half;
 typedef half2 wsp_ggml_half2;
 
-#define WSP_GGML_COMMON_AGGR data
+#define WSP_GGML_COMMON_AGGR_U
+#define WSP_GGML_COMMON_AGGR_S data
 
 #define WSP_GGML_COMMON_DECL
 #elif defined(WSP_GGML_COMMON_DECL_HIP)
@@ -39,7 +54,8 @@ typedef half2 wsp_ggml_half2;
 typedef half  wsp_ggml_half;
 typedef half2 wsp_ggml_half2;
 
-#define WSP_GGML_COMMON_AGGR data
+#define WSP_GGML_COMMON_AGGR_U
+#define WSP_GGML_COMMON_AGGR_S data
 
 #define WSP_GGML_COMMON_DECL
 #elif defined(WSP_GGML_COMMON_DECL_SYCL)
@@ -49,7 +65,8 @@ typedef half2 wsp_ggml_half2;
 typedef sycl::half  wsp_ggml_half;
 typedef sycl::half2 wsp_ggml_half2;
 
-#define WSP_GGML_COMMON_AGGR data
+#define WSP_GGML_COMMON_AGGR_U
+#define WSP_GGML_COMMON_AGGR_S data
 
 #define WSP_GGML_COMMON_DECL
 #endif
@@ -154,9 +171,9 @@ typedef struct {
         struct {
             wsp_ggml_half d; // delta
             wsp_ggml_half m; // min
-        } WSP_GGML_COMMON_AGGR;
+        } WSP_GGML_COMMON_AGGR_S;
         wsp_ggml_half2 dm;
-    };
+    } WSP_GGML_COMMON_AGGR_U;
     uint8_t qs[QK4_1 / 2]; // nibbles / quants
 } block_q4_1;
 static_assert(sizeof(block_q4_1) == 2 * sizeof(wsp_ggml_half) + QK4_1 / 2, "wrong q4_1 block size/padding");
@@ -175,9 +192,9 @@ typedef struct {
         struct {
             wsp_ggml_half d; // delta
             wsp_ggml_half m; // min
-        } WSP_GGML_COMMON_AGGR;
+        } WSP_GGML_COMMON_AGGR_S;
         wsp_ggml_half2 dm;
-    };
+    } WSP_GGML_COMMON_AGGR_U;
     uint8_t qh[4];         // 5-th bit of quants
     uint8_t qs[QK5_1 / 2]; // nibbles / quants
 } block_q5_1;
@@ -196,37 +213,13 @@ typedef struct {
         struct {
             wsp_ggml_half d; // delta
             wsp_ggml_half s; // d * sum(qs[i])
-        } WSP_GGML_COMMON_AGGR;
+        } WSP_GGML_COMMON_AGGR_S;
         wsp_ggml_half2 ds;
-    };
+    } WSP_GGML_COMMON_AGGR_U;
     int8_t qs[QK8_1]; // quants
 } block_q8_1;
 static_assert(sizeof(block_q8_1) == 2*sizeof(wsp_ggml_half) + QK8_1, "wrong q8_1 block size/padding");
 
-typedef struct {
-    wsp_ggml_half d[4];        // deltas for 4 q4_0 blocks
-    uint8_t qs[QK4_0 * 2]; // nibbles / quants for 4 q4_0 blocks
-} block_q4_0x4;
-static_assert(sizeof(block_q4_0x4) == 4 * sizeof(wsp_ggml_half) + QK4_0 * 2, "wrong q4_0x4 block size/padding");
-
-typedef struct {
-    wsp_ggml_half d[8];        // deltas for 8 q4_0 blocks
-    uint8_t qs[QK4_0 * 4]; // nibbles / quants for 8 q4_0 blocks
-} block_q4_0x8;
-static_assert(sizeof(block_q4_0x8) == 8 * sizeof(wsp_ggml_half) + QK4_0 * 4, "wrong q4_0x8 block size/padding");
-
-typedef struct {
-    wsp_ggml_half d[4];        // deltas for 4 q8_0 blocks
-    int8_t qs[QK8_0 * 4];  // quants for 4 q8_0 blocks
-} block_q8_0x4;
-static_assert(sizeof(block_q8_0x4) == 4 * sizeof(wsp_ggml_half) + QK8_0 * 4, "wrong q8_0x4 block size/padding");
-
-typedef struct {
-    wsp_ggml_half d[8];        // deltas for 8 q8_0 blocks
-    int8_t qs[QK8_0 * 8];  // quants for 8 q8_0 blocks
-} block_q8_0x8;
-static_assert(sizeof(block_q8_0x8) == 8 * sizeof(wsp_ggml_half) + QK8_0 * 8, "wrong q8_0x8 block size/padding");
-
 //
 // Ternary quantization
 //
@@ -261,9 +254,9 @@ typedef struct {
         struct {
             wsp_ggml_half d;    // super-block scale for quantized scales
             wsp_ggml_half dmin; // super-block scale for quantized mins
-        } WSP_GGML_COMMON_AGGR;
+        } WSP_GGML_COMMON_AGGR_S;
         wsp_ggml_half2 dm;
-    };
+    } WSP_GGML_COMMON_AGGR_U;
 } block_q2_K;
 static_assert(sizeof(block_q2_K) == 2*sizeof(wsp_ggml_half) + QK_K/16 + QK_K/4, "wrong q2_K block size/padding");
 
@@ -288,9 +281,9 @@ typedef struct {
         struct {
             wsp_ggml_half d;    // super-block scale for quantized scales
             wsp_ggml_half dmin; // super-block scale for quantized mins
-        } WSP_GGML_COMMON_AGGR;
+        } WSP_GGML_COMMON_AGGR_S;
         wsp_ggml_half2 dm;
-    };
+    } WSP_GGML_COMMON_AGGR_U;
     uint8_t scales[K_SCALE_SIZE]; // scales and mins, quantized with 6 bits
     uint8_t qs[QK_K/2];           // 4--bit quants
 } block_q4_K;
@@ -305,9 +298,9 @@ typedef struct {
         struct {
             wsp_ggml_half d;    // super-block scale for quantized scales
             wsp_ggml_half dmin; // super-block scale for quantized mins
-        } WSP_GGML_COMMON_AGGR;
+        } WSP_GGML_COMMON_AGGR_S;
         wsp_ggml_half2 dm;
-    };
+    } WSP_GGML_COMMON_AGGR_U;
     uint8_t scales[K_SCALE_SIZE]; // scales and mins, quantized with 6 bits
     uint8_t qh[QK_K/8];           // quants, high bit
     uint8_t qs[QK_K/2];           // quants, low 4 bits
@@ -431,6 +424,13 @@ static_assert(sizeof(block_iq4_xs) == sizeof(wsp_ggml_half) + sizeof(uint16_t) +
 #define WSP_GGML_TABLE_BEGIN(type, name, size) static const type name[size] = {
 #define WSP_GGML_TABLE_END() };
 
+#define WSP_GGML_COMMON_IMPL
+#elif defined(WSP_GGML_COMMON_IMPL_CPP)
+#include <cstdint>
+
+#define WSP_GGML_TABLE_BEGIN(type, name, size) static const type name[size] = {
+#define WSP_GGML_TABLE_END() };
+
 #define WSP_GGML_COMMON_IMPL
 #elif defined(WSP_GGML_COMMON_IMPL_METAL)
 #include <metal_stdlib>
@@ -473,7 +473,6 @@ WSP_GGML_TABLE_BEGIN(uint8_t, ksigns_iq2xs, 128)
     240, 113, 114, 243, 116, 245, 246, 119, 120, 249, 250, 123, 252, 125, 126, 255,
 WSP_GGML_TABLE_END()
 
-//#if __CUDA_ARCH__ >= MIN_CC_DP4A // lowest compute capability for integer intrinsics
 WSP_GGML_TABLE_BEGIN(uint64_t, ksigns64, 128)
     0x0000000000000000, 0xff000000000000ff, 0xff0000000000ff00, 0x000000000000ffff,
     0xff00000000ff0000, 0x0000000000ff00ff, 0x0000000000ffff00, 0xff00000000ffffff,
@@ -508,7 +507,6 @@ WSP_GGML_TABLE_BEGIN(uint64_t, ksigns64, 128)
     0x00ffffffff000000, 0xffffffffff0000ff, 0xffffffffff00ff00, 0x00ffffffff00ffff,
     0xffffffffffff0000, 0x00ffffffffff00ff, 0x00ffffffffffff00, 0xffffffffffffffff,
 WSP_GGML_TABLE_END()
-//#endif
 
 
 WSP_GGML_TABLE_BEGIN(uint64_t, iq2xxs_grid, 256)

package/cpp/ggml-cpp.h

@@ -7,6 +7,7 @@
 #include "ggml.h"
 #include "ggml-alloc.h"
 #include "ggml-backend.h"
+#include "gguf.h"
 #include <memory>
 
 // Smart pointers for ggml types