ggml : suppress Windows compiler warnings

This commit suppresses compiler warnings emitted from ggml on Windows.

The motivation for these changes is that some compilers generate
warnings for these conversion, for example Windows MSVC, and there
are quite a few of them. This makes it a little difficult to spot new
warnings that may be introduced and also can be difficult for
users/embedders of ggml where these warnings are hard to separate from
their own warnings.

Author: danbev

ggml/src/ggml-backend-reg.cpp

@@ -453,7 +453,7 @@ static fs::path get_executable_path() {
     return base_path + "/";
 #elif defined(_WIN32)
     std::vector<wchar_t> path(MAX_PATH);
-    DWORD len = GetModuleFileNameW(NULL, path.data(), path.size());
+    DWORD len = GetModuleFileNameW(NULL, path.data(), (DWORD)path.size());
     if (len == 0) {
         return {};
     }

ggml/src/ggml-cpu/ggml-cpu-quants.c

@@ -1,4 +1,8 @@
 #define GGML_COMMON_IMPL_C
+// assert.h is included before ggml-common.h to a warning about
+// redefinition of static_assert.
+#include <assert.h>
+
 #include "ggml-common.h"
 
 #include "ggml-quants.h"
@@ -9,7 +13,6 @@
 
 #include <math.h>
 #include <string.h>
-#include <assert.h>
 #include <float.h>
 #include <stdlib.h> // for qsort
 #include <stdio.h>  // for GGML_ASSERT

ggml/src/ggml-opt.cpp

@@ -596,8 +596,8 @@ static void ggml_opt_eval_graph(ggml_opt_context_t opt_ctx, ggml_cgraph * graph,
         GGML_ASSERT(opt_pars.adamw.wd    <= 1.0f);
 
         // beta1, beta2 after applying warmup
-        const float beta1h = 1.0f/(1.0f - powf(opt_pars.adamw.beta1, opt_ctx->iter));
-        const float beta2h = 1.0f/(1.0f - powf(opt_pars.adamw.beta2, opt_ctx->iter));
+        const float beta1h = 1.0f/(1.0f - powf(opt_pars.adamw.beta1, (float)opt_ctx->iter));
+        const float beta2h = 1.0f/(1.0f - powf(opt_pars.adamw.beta2, (float)opt_ctx->iter));
 
         float * adamw_par_data = ggml_get_data_f32(opt_ctx->adamw_params);
         adamw_par_data[0] = opt_pars.adamw.alpha;
@@ -807,7 +807,7 @@ void ggml_opt_fit(
     int64_t epoch = 1;
 
     ggml_opt_params params = ggml_opt_default_params(backend_sched, ctx_compute, inputs, outputs, loss_type);
-    params.opt_period      = opt_period;
+    params.opt_period      = (int32_t)opt_period;
     params.get_opt_pars    = get_opt_pars;
     params.get_opt_pars_ud = &epoch;
     ggml_opt_context_t opt_ctx = ggml_opt_init(params);

ggml/src/ggml-quants.c

@@ -1,4 +1,8 @@
 #define GGML_COMMON_IMPL_C
+// assert.h is included before ggml-common.h to a warning about
+// redefinition of static_assert.
+#include <assert.h>
+
 #include "ggml-common.h"
 
 #include "ggml-quants.h"
@@ -8,7 +12,6 @@
 
 #include <math.h>
 #include <string.h>
-#include <assert.h>
 #include <float.h>
 #include <stdlib.h> // for qsort
 #include <stdio.h>  // for GGML_ASSERT

ggml/src/ggml.c

@@ -1965,7 +1965,7 @@ static struct ggml_tensor * ggml_acc_impl(
 
     struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
 
-    int32_t params[] = { nb1, nb2, nb3, offset, inplace ? 1 : 0 };
+    int32_t params[] = { (int32_t)nb1, (int32_t)nb2, (int32_t)nb3, (int32_t)offset, inplace ? 1 : 0 };
     ggml_set_op_params(result, params, sizeof(params));
 
     result->op     = GGML_OP_ACC;
@@ -2881,7 +2881,7 @@ static struct ggml_tensor * ggml_set_impl(
     struct ggml_tensor * result = inplace ? ggml_view_tensor(ctx, a) : ggml_dup_tensor(ctx, a);
 
     GGML_ASSERT(offset < (size_t)(1 << 30));
-    int32_t params[] = { nb1, nb2, nb3, offset, inplace ? 1 : 0 };
+    int32_t params[] = { (int32_t)nb1, (int32_t)nb2, (int32_t)nb3, (int32_t)offset, inplace ? 1 : 0 };
     ggml_set_op_params(result, params, sizeof(params));
 
     result->op     = GGML_OP_SET;
@@ -3269,15 +3269,15 @@ struct ggml_tensor * ggml_permute(
     int ne[GGML_MAX_DIMS];
     int nb[GGML_MAX_DIMS];
 
-    ne[axis0] = a->ne[0];
-    ne[axis1] = a->ne[1];
-    ne[axis2] = a->ne[2];
-    ne[axis3] = a->ne[3];
+    ne[axis0] = (int)a->ne[0];
+    ne[axis1] = (int)a->ne[1];
+    ne[axis2] = (int)a->ne[2];
+    ne[axis3] = (int)a->ne[3];
 
-    nb[axis0] = a->nb[0];
-    nb[axis1] = a->nb[1];
-    nb[axis2] = a->nb[2];
-    nb[axis3] = a->nb[3];
+    nb[axis0] = (int)a->nb[0];
+    nb[axis1] = (int)a->nb[1];
+    nb[axis2] = (int)a->nb[2];
+    nb[axis3] = (int)a->nb[3];
 
     result->ne[0] = ne[0];
     result->ne[1] = ne[1];
@@ -3916,7 +3916,7 @@ struct ggml_tensor* ggml_conv_1d_ph(
         struct ggml_tensor  * b,
         int                   s,
         int                   d) {
-    return ggml_conv_1d(ctx, a, b, s, a->ne[0] / 2, d);
+    return ggml_conv_1d(ctx, a, b, s, (int)a->ne[0] / 2, d);
 }
 
 // ggml_conv_1d_dw
@@ -3948,7 +3948,7 @@ struct ggml_tensor * ggml_conv_1d_dw_ph(
         struct ggml_tensor  * b,
         int                   s0,
         int                   d0) {
-    return ggml_conv_1d_dw(ctx, a, b, s0, a->ne[0] / 2, d0);
+    return ggml_conv_1d_dw(ctx, a, b, s0, (int)a->ne[0] / 2, d0);
 }
 
 // ggml_conv_transpose_1d
@@ -4022,7 +4022,7 @@ struct ggml_tensor * ggml_conv_2d_sk_p0(
         struct ggml_context * ctx,
         struct ggml_tensor  * a,
         struct ggml_tensor  * b) {
-    return ggml_conv_2d(ctx, a, b, a->ne[0], a->ne[1], 0, 0, 1, 1);
+    return ggml_conv_2d(ctx, a, b, (int)a->ne[0], (int)a->ne[1], 0, 0, 1, 1);
 }
 
 // ggml_conv_2d_s1_ph
@@ -4031,7 +4031,7 @@ struct ggml_tensor * ggml_conv_2d_s1_ph(
         struct ggml_context * ctx,
         struct ggml_tensor  * a,
         struct ggml_tensor  * b) {
-    return ggml_conv_2d(ctx, a, b, 1, 1, a->ne[0] / 2, a->ne[1] / 2, 1, 1);
+    return ggml_conv_2d(ctx, a, b, 1, 1, (int)a->ne[0] / 2, (int)a->ne[1] / 2, 1, 1);
 }
 
 // ggml_conv_2d_dw
@@ -4132,7 +4132,7 @@ struct ggml_tensor * ggml_conv_transpose_2d_p0(
 // ggml_pool_*
 
 static int64_t ggml_calc_pool_output_size(int64_t ins, int ks, int s, float p) {
-    return (ins + 2 * p - ks) / s + 1;
+    return (int64_t)(ins + 2 * p - ks) / s + 1;
 }
 
 // ggml_pool_1d
@@ -4145,7 +4145,7 @@ struct ggml_tensor * ggml_pool_1d(
         int                   s0,
         int                   p0) {
     const int64_t ne[4] = {
-        ggml_calc_pool_output_size(a->ne[0], k0, s0, p0),
+        ggml_calc_pool_output_size(a->ne[0], k0, s0, (float)p0),
         a->ne[1],
         a->ne[2],
         a->ne[3],
@@ -4182,7 +4182,7 @@ struct ggml_tensor * ggml_pool_2d(
     };
     result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, ne);
 
-    int32_t params[] = { op, k0, k1, s0, s1, p0, p1 };
+    int32_t params[] = { op, k0, k1, s0, s1, (int)p0, (int)p1 };
     ggml_set_op_params(result, params, sizeof(params));
 
     result->op     = GGML_OP_POOL_2D;
@@ -4205,7 +4205,7 @@ struct ggml_tensor * ggml_pool_2d_back(
     struct ggml_tensor * result;
     result = ggml_new_tensor(ctx, GGML_TYPE_F32, 4, af->ne);
 
-    int32_t params[] = { op, k0, k1, s0, s1, p0, p1 };
+    int32_t params[] = { op, k0, k1, s0, s1, (int)p0, (int)p1 };
     ggml_set_op_params(result, params, sizeof(params));
 
     result->op     = GGML_OP_POOL_2D_BACK;
@@ -4245,7 +4245,7 @@ struct ggml_tensor * ggml_upscale(
         struct ggml_tensor  * a,
         int                   scale_factor,
         enum ggml_scale_mode  mode) {
-    return ggml_upscale_impl(ctx, a, a->ne[0] * scale_factor, a->ne[1] * scale_factor, a->ne[2], a->ne[3], mode);
+    return ggml_upscale_impl(ctx, a, (int)a->ne[0] * scale_factor, (int)a->ne[1] * scale_factor, (int)a->ne[2], (int)a->ne[3], mode);
 }
 
 struct ggml_tensor * ggml_upscale_ext(
@@ -4617,8 +4617,8 @@ struct ggml_tensor * ggml_win_part(
     const int px = (w - a->ne[1]%w)%w;
     const int py = (w - a->ne[2]%w)%w;
 
-    const int npx = (px + a->ne[1])/w;
-    const int npy = (py + a->ne[2])/w;
+    const int npx = (int)(px + a->ne[1])/w;
+    const int npy = (int)(py + a->ne[2])/w;
     const int np  = npx*npy;
 
     const int64_t ne[4] = { a->ne[0], w, w, np, };
@@ -5689,7 +5689,7 @@ static void ggml_compute_backward(
                 const      int32_t      p0 = ggml_get_op_params_i32(tensor, 5);
                 const      int32_t      p1 = ggml_get_op_params_i32(tensor, 6);
 
-                ggml_add_or_set(ctx, cgraph, isrc0, ggml_pool_2d_back(ctx, grad, src0, op, k0, k1, s0, s1, p0, p1));
+                ggml_add_or_set(ctx, cgraph, isrc0, ggml_pool_2d_back(ctx, grad, src0, op, k0, k1, s0, s1, (float)p0, (float)p1));
             }
         } break;
         case GGML_OP_WIN_PART:
@@ -5967,7 +5967,7 @@ struct ggml_cgraph * ggml_new_graph_custom(struct ggml_context * ctx, size_t siz
     assert(obj_size == (size_t)((char *)p - (char *)cgraph));
 
     *cgraph = (struct ggml_cgraph) {
-        /*.size         =*/ size,
+        /*.size         =*/ (int)size,
         /*.n_nodes      =*/ 0,
         /*.n_leafs      =*/ 0,
         /*.nodes        =*/ nodes_ptr,

ggml/src/gguf.cpp

@@ -454,7 +454,7 @@ struct gguf_context * gguf_init_from_file_impl(FILE * file, struct gguf_init_par
         }
         GGML_ASSERT(int64_t(ctx->kv.size()) == n_kv);
 
-        const int alignment_idx = gguf_find_key(ctx, GGUF_KEY_GENERAL_ALIGNMENT);
+        const int alignment_idx = (int)gguf_find_key(ctx, GGUF_KEY_GENERAL_ALIGNMENT);
         ctx->alignment = alignment_idx == -1 ? GGUF_DEFAULT_ALIGNMENT : gguf_get_val_u32(ctx, alignment_idx);
 
         if (ctx->alignment == 0 || (ctx->alignment & (ctx->alignment - 1)) != 0) {
@@ -589,7 +589,7 @@ struct gguf_context * gguf_init_from_file_impl(FILE * file, struct gguf_init_par
     GGML_ASSERT(int64_t(ctx->info.size()) == n_tensors);
 
     // we require the data section to be aligned, so take into account any padding
-    if (fseek(file, GGML_PAD(ftell(file), ctx->alignment), SEEK_SET) != 0) {
+    if (fseek(file, (long)GGML_PAD(ftell(file), ctx->alignment), SEEK_SET) != 0) {
         fprintf(stderr, "%s: failed to seek to beginning of data section\n", __func__);
         gguf_free(ctx);
         return nullptr;