Merge pull request #68 from mmwillet/reciprocal-stride-hack

Reciprocal stride hack

Author: mmwillet

src/util.cpp

@@ -75,11 +75,21 @@ float round_to_float(double v) {
     return roundf(v * powl(10, 6)) / powl(10, 6);
 }
 
+struct ggml_tensor * reciprocal(ggml_context * ctx, struct ggml_tensor * x) {
+    TTS_ASSERT(x->ne[0] == 1);
+    static constexpr float one = 1.0f;
+    ggml_tensor * numerator = ggml_new_tensor_2d(ctx, GGML_TYPE_F32, 1, x->ne[1]);
+    // stride trick so that the scalar numerator can be divided by x.
+    numerator->nb[1] = 0;
+    numerator->data = const_cast<float *>(&one);
+    return ggml_div(ctx, numerator, x);
+}
+
 // Described in https://arxiv.org/abs/2006.08195
 // Snake1d is a common tunable activation function used in the DAC model.
 struct ggml_tensor * snake_1d(ggml_context * ctx, struct ggml_tensor * alpha, struct ggml_tensor * a) {
     assert(a->ne[2] == 1 && a->ne[3] == 1);
-    return ggml_add(ctx, a, ggml_mul(ctx, ggml_sqr(ctx, ggml_sin(ctx, ggml_mul(ctx, a, alpha))), ggml_reciprocal(ctx, alpha)));
+    return ggml_add(ctx, a, ggml_mul(ctx, ggml_sqr(ctx, ggml_sin(ctx, ggml_mul(ctx, a, alpha))), reciprocal(ctx, alpha)));
 }
 
 bool has_suffix(std::string value, std::string suffix) {

src/util.h

@@ -50,6 +50,8 @@ void uv_noise_compute(struct ggml_tensor * dst, const struct ggml_tensor * a, co
 // This is a custom op for logit correction in the Dia model.
 void cfg_scale(struct ggml_tensor * dst, const struct ggml_tensor * a, const struct ggml_tensor * b, int ith, int nth, void * userdata);
 
+struct ggml_tensor * reciprocal(ggml_context * ctx, struct ggml_tensor * x);
+
 bool has_suffix(std::string value, std::string suffix);
 bool has_prefix(std::string value, std::string prefix);