fix: build errors from upstream merge — extern linkage + switch coverage (#48)

1. extern "C" GGML_API double-extern: wrap with extern "C" { } block
2. Missing turbo types in clamp switch: add TURBO3_0/TURBO4_0/TURBO2_0

Fixes #48.

Author: signalnine

ggml/src/ggml-cpu/ops.cpp

@@ -12,8 +12,6 @@
 #include <cfloat>
 #include <cmath>
 
-extern "C" GGML_API int turbo3_cpu_wht_group_size;
-
 // ggml_compute_forward_dup
 
 static void ggml_compute_forward_dup_same_cont(
@@ -682,6 +680,7 @@ void ggml_compute_forward_add(
         case GGML_TYPE_TQ2_0:
         case GGML_TYPE_TQ3_1S:
         case GGML_TYPE_TQ4_1S:
+        case GGML_TYPE_TQ4_0:
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
@@ -1134,6 +1133,7 @@ void ggml_compute_forward_add1(
         case GGML_TYPE_TQ2_0:
         case GGML_TYPE_TQ3_1S:
         case GGML_TYPE_TQ4_1S:
+        case GGML_TYPE_TQ4_0:
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
@@ -1265,6 +1265,7 @@ void ggml_compute_forward_acc(
         case GGML_TYPE_TQ2_0:
         case GGML_TYPE_TQ3_1S:
         case GGML_TYPE_TQ4_1S:
+        case GGML_TYPE_TQ4_0:
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
@@ -4355,6 +4356,7 @@ void ggml_compute_forward_out_prod(
         case GGML_TYPE_TQ2_0:
         case GGML_TYPE_TQ3_1S:
         case GGML_TYPE_TQ4_1S:
+        case GGML_TYPE_TQ4_0:
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
@@ -4633,6 +4635,7 @@ void ggml_compute_forward_set(
         case GGML_TYPE_TQ2_0:
         case GGML_TYPE_TQ3_1S:
         case GGML_TYPE_TQ4_1S:
+        case GGML_TYPE_TQ4_0:
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
@@ -4858,6 +4861,7 @@ void ggml_compute_forward_get_rows(
         case GGML_TYPE_TQ2_0:
         case GGML_TYPE_TQ3_1S:
         case GGML_TYPE_TQ4_1S:
+        case GGML_TYPE_TQ4_0:
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
@@ -4942,6 +4946,7 @@ static void ggml_compute_forward_set_rows_f32(
 
     // For turbo types: communicate WHT group size to the quantize function via global
     if (dst->type == GGML_TYPE_TURBO3_0 || dst->type == GGML_TYPE_TURBO4_0 || dst->type == GGML_TYPE_TURBO2_0) {
+        extern int turbo3_cpu_wht_group_size;
         int gs = 0;
         memcpy(&gs, dst->op_params, sizeof(int));
         turbo3_cpu_wht_group_size = (gs == 64 || gs == 128) ? gs : 0;
@@ -5592,6 +5597,7 @@ void ggml_compute_forward_clamp(
         case GGML_TYPE_TQ2_0:
         case GGML_TYPE_TQ3_1S:
         case GGML_TYPE_TQ4_1S:
+        case GGML_TYPE_TQ4_0:
         case GGML_TYPE_IQ2_XXS:
         case GGML_TYPE_IQ2_XS:
         case GGML_TYPE_IQ3_XXS:
@@ -9976,9 +9982,13 @@ static void ggml_compute_forward_rwkv_wkv6_f32(
     const int ith = params->ith;
     const int nth = params->nth;
 
-    const int h_start =  (HEADS * (ith    )) / nth;
-    const int h_end   = ((HEADS * (ith + 1)) / nth < HEADS) ?
-                         (HEADS * (ith + 1)) / nth : HEADS;
+    if (ith >= HEADS) {
+        return;
+    }
+
+    const int h_start = (HEADS * ith) / nth;
+    const int h_end = ((HEADS * (ith + 1)) / nth < HEADS) ?
+                (HEADS * (ith + 1)) / nth : HEADS;
 
     float * k =          (float *) dst->src[0]->data;
     float * v =          (float *) dst->src[1]->data;
@@ -10189,9 +10199,13 @@ static void ggml_compute_forward_gla_f32(
     const int ith = params->ith;
     const int nth = params->nth;
 
-    const int h_start =  (HEADS * (ith    )) / nth;
-    const int h_end   = ((HEADS * (ith + 1)) / nth < HEADS) ?
-                         (HEADS * (ith + 1)) / nth : HEADS;
+    if (ith >= HEADS) {
+        return;
+    }
+
+    const int h_start = (HEADS * ith) / nth;
+    const int h_end = ((HEADS * (ith + 1)) / nth < HEADS) ?
+                (HEADS * (ith + 1)) / nth : HEADS;
 
     float * k = (float *) dst->src[0]->data;
     float * v = (float *) dst->src[1]->data;
@@ -10746,9 +10760,13 @@ static void ggml_compute_forward_rwkv_wkv7_f32(
     const int ith = params->ith;
     const int nth = params->nth;
 
-    const int h_start =  (HEADS * (ith    )) / nth;
-    const int h_end   = ((HEADS * (ith + 1)) / nth < HEADS) ?
-                         (HEADS * (ith + 1)) / nth : HEADS;
+    if (ith >= HEADS) {
+        return;
+    }
+
+    const int h_start = (HEADS * ith) / nth;
+    const int h_end = ((HEADS * (ith + 1)) / nth < HEADS) ?
+                (HEADS * (ith + 1)) / nth : HEADS;
 
     float * r = (float *) dst->src[0]->data;
     float * w = (float *) dst->src[1]->data;