GCC Code Coverage Report

Directory: brotensor/
Coverage: low: ≥ 0% medium: ≥ 75.0% high: ≥ 90.0%
Coverage Exec / Excl / Total
Lines: 91.1% 102 / 0 / 112
Functions: 100.0% 11 / 0 / 11
Branches: 32.2% 94 / 0 / 292
src/api/native_tensor_misc.cpp
Line Branch Exec Source
1 // native_tensor_misc.cpp — bodies for the "misc" restored group: the
2 // gated-deltanet L2 norm pair, three remaining loss entry points and the
3 // diffusion sampler steps + timestep embedding. See native_tensor_misc_decl.h
4 // for the provenance of each function and native_tensor_nn.cpp for the
5 // mechanism these bodies copy (need() / BROTENSOR_API_TRY / the error slot).
6
7 #include "native_tensor_misc_decl.h"
8 #include "api_internal.h"
9 #include "native_register.h"
10
11 #include <string>
12
13 using namespace brotensor::api;
14
15 namespace {
16
17 // X is (L, numHeads*headDim), both positive.
18 6 bool needHeadCols(const char* L, const brotensor::Tensor* x, int32_t headDim, int32_t numHeads) {
19
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
 6 if (!needPositive(L, {headDim, numHeads})) return false;
20
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 4 times.
 6 if (static_cast<int64_t>(x->cols) != static_cast<int64_t>(numHeads) * headDim) {
21
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 2 setError(std::string(L) + ": X must have numHeads*headDim columns");
22 2 return false;
23 }
24 4 return true;
25 6 }
26
27 } // namespace
28
29 extern "C" {
30
31 // ---- L2 norm ---------------------------------------------------------------
32
33 4 void bro_tensor_l2NormForward(void* X, int32_t headDim, int32_t numHeads, double eps, void* Y) {
34
3/4
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 2 times.
 4 if (!need("l2NormForward", {X, Y}) || !needHeadCols("l2NormForward", toTensor(X), headDim, numHeads)) return;
35 BROTENSOR_API_TRY
36
3/6
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
 2 brotensor::l2_norm_forward(*toTensor(X), headDim, numHeads, static_cast<float>(eps), *toTensor(Y));
37
0/12
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 2 BROTENSOR_API_CATCH("l2NormForward")
38 4 }
39
40 2 void bro_tensor_l2NormBackward(void* X, int32_t headDim, int32_t numHeads, double eps, void* dY, void* dX) {
41
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 2 if (!need("l2NormBackward", {X, dY, dX}) || !needHeadCols("l2NormBackward", toTensor(X), headDim, numHeads)) return;
42
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!needPair("l2NormBackward", "dY", toTensor(dY), toTensor(X))) return;
43 BROTENSOR_API_TRY
44
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 4 brotensor::l2_norm_backward(*toTensor(X), headDim, numHeads, static_cast<float>(eps),
45
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 2 *toTensor(dY), *toTensor(dX));
46
0/12
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 2 BROTENSOR_API_CATCH("l2NormBackward")
47 2 }
48
49 // ---- losses ----------------------------------------------------------------
50
51 6 void bro_tensor_bceWithLogitsFusedBatched(void* logits_BL, void* target_BL, uint64_t mask_bits,
52 double posWeight, void* probs_BL, void* dLogits_BL,
53 void* lossPerSample) {
54 6 const char* L = "bceWithLogitsFusedBatched";
55
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
 6 if (!need(L, {logits_BL, target_BL, probs_BL, dLogits_BL, lossPerSample})) return;
56
1/2
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
 6 if (!needPair(L, "target_BL", toTensor(target_BL), toTensor(logits_BL))) return;
57
1/2
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
 6 if (!needMask(L, tensorFromValue(mask_bits), toTensor(logits_BL)->size())) return;
58 BROTENSOR_API_TRY
59
3/6
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 6 times.
✗ Branch 5 not taken.
 12 brotensor::bce_with_logits_fused_batched(*toTensor(logits_BL), *toTensor(target_BL),
60
1/2
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
 6 maskPtr(mask_bits), static_cast<float>(posWeight),
61
2/4
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 6 times.
✗ Branch 3 not taken.
 6 *toTensor(probs_BL), *toTensor(dLogits_BL),
62
1/2
✓ Branch 0 taken 6 times.
✗ Branch 1 not taken.
 6 *toTensor(lossPerSample));
63
0/12
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 6 BROTENSOR_API_CATCH("bceWithLogitsFusedBatched")
64 6 }
65
66 // Host buffers: the op takes raw `const float*` / `float*` (CPU-only), so the
67 // four arrays cross as f32[] pairs and probs / dLogits are written in place.
68 // An empty mask array is "no mask" — the old binding's `mask|null`.
69 4 double bro_tensor_softmaxXentSegment(const float* logits, uint32_t logits_len,
70 const float* target, uint32_t target_len,
71 float* probs, uint32_t probs_len,
72 float* dLogits, uint32_t dLogits_len,
73 int32_t n,
74 const float* mask, uint32_t mask_len) {
75
4/8
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 4 times.
 4 if (!logits || !target || !probs || !dLogits) {
76 ✗ setError("softmaxXentSegment: logits/target/probs/dLogits must be Float32Array");
77 ✗ return 0.0;
78 }
79
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (n <= 0) {
80 ✗ setError("softmaxXentSegment: n must be positive");
81 ✗ return 0.0;
82 }
83 4 const auto un = static_cast<uint32_t>(n);
84
4/8
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 4 times.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✓ Branch 7 taken 4 times.
 4 if (logits_len < un || target_len < un || probs_len < un || dLogits_len < un) {
85 ✗ setError("softmaxXentSegment: a buffer holds fewer than n=" + std::to_string(n) + " elements");
86 ✗ return 0.0;
87 }
88
3/4
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 2 times.
 4 const float* mask_ptr = (mask && mask_len > 0) ? mask : nullptr;
89
3/4
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 4 if (mask_ptr && mask_len < un) {
90 ✗ setError("softmaxXentSegment: mask holds fewer than n=" + std::to_string(n) + " elements");
91 ✗ return 0.0;
92 }
93 BROTENSOR_API_TRY
94
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
 4 return brotensor::softmax_xent_segment(logits, target, probs, dLogits, n, mask_ptr);
95 ✗ BROTENSOR_API_CATCH("softmaxXentSegment")
96 ✗ return 0.0;
97 4 }
98
99 // mseScalar(pred, target) -> [loss, dPred]: the pair as a length-2 f64[].
100 4 void bro_tensor_mseScalar(double pred, double target, bronze_native_buffer* out) {
101 static thread_local double pair[2];
102
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
 4 if (!out) return;
103 4 out->data = nullptr;
104 4 out->length = 0;
105 4 out->release = nullptr;
106 4 out->ctx = nullptr;
107 BROTENSOR_API_TRY
108 4 float dPred = 0.0f;
109
1/2
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
 4 const float loss = brotensor::mse_scalar(static_cast<float>(pred), static_cast<float>(target), dPred);
110 4 pair[0] = loss;
111 4 pair[1] = dPred;
112 4 out->data = pair;
113 4 out->length = 2;
114
0/12
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 4 BROTENSOR_API_CATCH("mseScalar")
115 4 }
116
117 // ---- diffusion sampler steps + timestep embedding --------------------------
118
119 2 void bro_tensor_ddimStep(void* x_t, void* eps_pred, double alphaT, double alphaPrev,
120 double sigmaT, void* x_prev) {
121
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!need("ddimStep", {x_t, eps_pred, x_prev})) return;
122
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!needPair("ddimStep", "eps_pred", toTensor(eps_pred), toTensor(x_t))) return;
123 BROTENSOR_API_TRY
124
3/6
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
 2 brotensor::ddim_step(*toTensor(x_t), *toTensor(eps_pred), static_cast<float>(alphaT),
125
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 static_cast<float>(alphaPrev), static_cast<float>(sigmaT), *toTensor(x_prev));
126
0/12
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 2 BROTENSOR_API_CATCH("ddimStep")
127 2 }
128
129 2 void bro_tensor_eulerStep(void* x_t, void* eps_pred, double sigmaT, double sigmaPrev, void* x_prev) {
130
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!need("eulerStep", {x_t, eps_pred, x_prev})) return;
131
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!needPair("eulerStep", "eps_pred", toTensor(eps_pred), toTensor(x_t))) return;
132 BROTENSOR_API_TRY
133
3/6
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
 2 brotensor::euler_step(*toTensor(x_t), *toTensor(eps_pred), static_cast<float>(sigmaT),
134
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 static_cast<float>(sigmaPrev), *toTensor(x_prev));
135
0/12
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 2 BROTENSOR_API_CATCH("eulerStep")
136 2 }
137
138 2 void bro_tensor_dpmpp2mStep(void* x_t, void* eps_pred, void* x0_prev, double sigmaT,
139 double c_xt, double c_x0t, double c_x0prev,
140 void* x_prev, void* x0_out) {
141
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!need("dpmpp2mStep", {x_t, eps_pred, x0_prev, x_prev, x0_out})) return;
142
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 2 if (!needPair("dpmpp2mStep", "eps_pred", toTensor(eps_pred), toTensor(x_t)) ||
143 2 !needPair("dpmpp2mStep", "x0_prev", toTensor(x0_prev), toTensor(x_t))) return;
144 BROTENSOR_API_TRY
145
4/8
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 2 times.
✗ Branch 7 not taken.
 4 brotensor::dpmpp_2m_step(*toTensor(x_t), *toTensor(eps_pred), *toTensor(x0_prev),
146 2 static_cast<float>(sigmaT), static_cast<float>(c_xt),
147 2 static_cast<float>(c_x0t), static_cast<float>(c_x0prev),
148
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 2 *toTensor(x_prev), *toTensor(x0_out));
149
0/12
✗ Branch 0 not taken.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 2 BROTENSOR_API_CATCH("dpmpp2mStep")
150 2 }
151
152 8 void bro_tensor_timestepEmbedding(void* timesteps, int32_t dim, double maxPeriod, void* Y) {
153
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 8 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 8 times.
 8 if (!need("timestepEmbedding", {timesteps, Y}) || !needNonNegative("timestepEmbedding", {dim})) return;
154 BROTENSOR_API_TRY
155
4/6
✓ Branch 0 taken 8 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 8 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 6 times.
✓ Branch 5 taken 2 times.
 8 brotensor::timestep_embedding(*toTensor(timesteps), dim, static_cast<float>(maxPeriod), *toTensor(Y));
156
5/12
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 2 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 2 times.
✗ Branch 7 not taken.
✓ Branch 8 taken 2 times.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
 8 BROTENSOR_API_CATCH("timestepEmbedding")
157 10 }
158
159 } // extern "C"
160
161 namespace brotensor::api {
162
163 3 bool registerTensorNatives_misc(std::string* error) {
164 using namespace brotensor::api::reg;
165 3 return
166 6 fn("__bro_native.tensor.l2NormForward", p(&bro_tensor_l2NormForward), "void",
167
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, "i32", "i32", "f64", kTensorCls}, error) &&
168 6 fn("__bro_native.tensor.l2NormBackward", p(&bro_tensor_l2NormBackward), "void",
169
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, "i32", "i32", "f64", kTensorCls, kTensorCls}, error) &&
170 6 fn("__bro_native.tensor.bceWithLogitsFusedBatched", p(&bro_tensor_bceWithLogitsFusedBatched), "void",
171
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, kDyn, "f64", kTensorCls, kTensorCls, kTensorCls}, error) &&
172 6 fn("__bro_native.tensor.softmaxXentSegment", p(&bro_tensor_softmaxXentSegment), "f64",
173
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {"f32[]", "f32[]", "f32[]", "f32[]", "i32", "f32[]"}, error) &&
174 6 fn("__bro_native.tensor.mseScalar", p(&bro_tensor_mseScalar), "f64[]",
175
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {"f64", "f64"}, error) &&
176 6 fn("__bro_native.tensor.ddimStep", p(&bro_tensor_ddimStep), "void",
177
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, "f64", "f64", "f64", kTensorCls}, error) &&
178 6 fn("__bro_native.tensor.eulerStep", p(&bro_tensor_eulerStep), "void",
179
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, "f64", "f64", kTensorCls}, error) &&
180 6 fn("__bro_native.tensor.dpmpp2mStep", p(&bro_tensor_dpmpp2mStep), "void",
181
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 3 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 3 times.
 6 {kTensorCls, kTensorCls, kTensorCls, "f64", "f64", "f64", "f64", kTensorCls, kTensorCls}, error) &&
182 6 fn("__bro_native.tensor.timestepEmbedding", p(&bro_tensor_timestepEmbedding), "void",
183 3 {kTensorCls, "i32", "f64", kTensorCls}, error);
184 }
185
186 } // namespace brotensor::api
187