GCC Code Coverage Report

Directory: brotensor/
Coverage: low: ≥ 0% medium: ≥ 75.0% high: ≥ 90.0%
Coverage Exec / Excl / Total
Lines: 100.0% 138 / 0 / 138
Functions: 100.0% 16 / 0 / 16
Branches: 34.9% 150 / 0 / 430
src/api/native_tensor_batched.cpp
Line Branch Exec Source
1 // native_tensor_batched.cpp — bodies for the restored "batched" group.
2 //
3 // Ported from the QuickJS binding (bro-quickjs-oracle
4 // src/js/tensor_bindings.cpp, `nngpu::` namespace) onto the op signatures that
5 // exist today under include/brotensor/ops/. A native is called straight from
6 // generated code, so nothing may unwind out of one: every body that reaches
7 // into brotensor is wrapped in BROTENSOR_API_TRY / BROTENSOR_API_CATCH, which
8 // records the message for js/tensor_batched.js to throw after the call.
9
10 #include "native_tensor_batched_decl.h"
11 #include "api_internal.h"
12 #include "native_register.h"
13 #include "native_tensor_extra_util.h"
14
15 #include <cstdint>
16 #include <string>
17 #include <vector>
18
19 using namespace brotensor::api;
20 using brotensor::Tensor;
21 using brotensor::Dtype;
22
23 // gather_rows / scatter_rows_add take Idx as an (M,1) INT32 tensor whose
24 // VALUES address rows on the device, unchecked by the kernels. The native
25 // reads the indices back, range-checks them, and hands the op an INT32
26 // operand (the tensor itself, or an INT32 copy of a whole-number FP32 one —
27 // GpuTensor.prototype.upload always lands FP32). See native_tensor_extra_util.h.
28
29 namespace {
30
31 124 Tensor* T(void* p) { return toTensor(p); }
32
33 // X_BD is (B, W.cols).
34 12 bool needBatchIn(const char* L, const Tensor* W, const Tensor* X) {
35
2/2
✓ Branch 0 taken 10 times.
✓ Branch 1 taken 2 times.
 12 if (X->cols != W->cols) {
36
5/10
✓ 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.
✓ Branch 8 taken 2 times.
✗ Branch 9 not taken.
 2 setError(std::string(L) + ": X_BD must be (B, " + std::to_string(W->cols) + ") to match W's in-dim");
37 2 return false;
38 }
39 10 return true;
40 12 }
41
42 } // namespace
43
44 extern "C" {
45
46 // ---- batched forwards ------------------------------------------------------
47
48 // W may be FP32/FP16/BF16; bias, X and Y are FP32.
49 6 void bro_tensor_linearForwardBatched(void* W, void* bias, void* X_BD, void* Y_BD) {
50 6 const char* L = "linearForwardBatched";
51
3/4
✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 4 times.
 6 if (!need(L, {W, bias, X_BD, Y_BD}) || !needBatchIn(L, T(W), T(X_BD))) return;
52
3/4
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 4 if (!needMask(L, T(bias), T(W)->rows, "bias") || !needMask(L, T(X_BD), 0, "X_BD")) return;
53 BROTENSOR_API_TRY
54
5/10
✓ 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.
✓ Branch 8 taken 2 times.
✗ Branch 9 not taken.
 2 brotensor::linear_forward_batched(*toTensor(W), *toTensor(bias), *toTensor(X_BD), *toTensor(Y_BD));
55
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(L)
56 6 }
57
58 4 void bro_tensor_linearForwardBatchedFp16(void* W, uint64_t bias_bits, void* X_BD, void* Y_BD) {
59 4 const char* L = "linearForwardBatchedFp16";
60
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
 4 if (!need(L, {W, X_BD, Y_BD}) || !needBatchIn(L, T(W), T(X_BD))) return;
61
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
 4 if (!needOperands(L, T(W), {{"X_BD", T(X_BD), 0}, {"bias", tensorFromValue(bias_bits), T(W)->rows}})) return;
62 BROTENSOR_API_TRY
63
3/6
✓ 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.
 8 brotensor::linear_forward_batched_fp16(*toTensor(W), tensorFromValue(bias_bits),
64
2/4
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
 4 *toTensor(X_BD), *toTensor(Y_BD));
65
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(L)
66 4 }
67
68 2 void bro_tensor_reluForwardBatched(void* X_BD, void* Y_BD) {
69
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!need("reluForwardBatched", {X_BD, Y_BD})) return;
70 BROTENSOR_API_TRY
71
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::relu_forward_batched(*toTensor(X_BD), *toTensor(Y_BD));
72
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("reluForwardBatched")
73 2 }
74
75 2 void bro_tensor_tanhForwardBatched(void* X_BD, void* Y_BD) {
76
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!need("tanhForwardBatched", {X_BD, Y_BD})) return;
77 BROTENSOR_API_TRY
78
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::tanh_forward_batched(*toTensor(X_BD), *toTensor(Y_BD));
79
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("tanhForwardBatched")
80 2 }
81
82 4 void bro_tensor_addInplaceBatched(void* Y_BD, void* X_BD) {
83
3/4
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 2 times.
 4 if (!need("addInplaceBatched", {Y_BD, X_BD}) || !needPair("addInplaceBatched", "X_BD", T(X_BD), T(Y_BD))) return;
84 BROTENSOR_API_TRY
85
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::add_inplace_batched(*toTensor(Y_BD), *toTensor(X_BD));
86
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("addInplaceBatched")
87 4 }
88
89 // ---- batched-training backwards --------------------------------------------
90
91 2 void bro_tensor_linearBackwardBatched(void* W, void* X_BD, void* dY_BD, void* dX_BD, void* dW, void* dB) {
92 2 const char* L = "linearBackwardBatched";
93
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 2 if (!need(L, {W, X_BD, dY_BD, dX_BD, dW, dB}) || !needBatchIn(L, T(W), T(X_BD))) return;
94 2 const Tensor* w = T(W);
95 // dW / dB accumulate — "caller zeros".
96
3/6
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
✗ Branch 4 not taken.
✓ Branch 5 taken 2 times.
 6 if (!needOperands(L, w, {{"X_BD", T(X_BD), 0}, {"dY_BD", T(dY_BD), elems({T(X_BD)->rows, w->rows})},
97 4 {"dW", T(dW), elems({w->rows, w->cols})}, {"dB", T(dB), w->rows}})) return;
98 BROTENSOR_API_TRY
99
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::linear_backward_batched(*toTensor(W), *toTensor(X_BD), *toTensor(dY_BD),
100
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 *toTensor(dX_BD), *toTensor(dW), *toTensor(dB));
101
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("linearBackwardBatched")
102 2 }
103
104 2 void bro_tensor_reluBackwardBatched(void* X_BD, void* dY_BD, void* dX_BD) {
105
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 2 if (!need("reluBackwardBatched", {X_BD, dY_BD, dX_BD}) ||
106 2 !needPair("reluBackwardBatched", "dY_BD", T(dY_BD), T(X_BD))) return;
107 BROTENSOR_API_TRY
108
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.
 2 brotensor::relu_backward_batched(*toTensor(X_BD), *toTensor(dY_BD), *toTensor(dX_BD));
109
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("reluBackwardBatched")
110 2 }
111
112 2 void bro_tensor_tanhBackwardBatched(void* Y_BD, void* dY_BD, void* dX_BD) {
113
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 2 if (!need("tanhBackwardBatched", {Y_BD, dY_BD, dX_BD}) ||
114 2 !needPair("tanhBackwardBatched", "dY_BD", T(dY_BD), T(Y_BD))) return;
115 BROTENSOR_API_TRY
116
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.
 2 brotensor::tanh_backward_batched(*toTensor(Y_BD), *toTensor(dY_BD), *toTensor(dX_BD));
117
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("tanhBackwardBatched")
118 2 }
119
120 // ---- row gather / scatter / top-k ------------------------------------------
121
122 10 void bro_tensor_gatherRows(void* X, void* Idx, void* Y) {
123 10 const char* L = "gatherRows";
124
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 10 times.
 10 if (!need(L, {X, Idx, Y})) return;
125 10 const Tensor* idx = T(Idx);
126 BROTENSOR_API_TRY
127 10 std::vector<int32_t> vals;
128
3/6
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
 10 if (!extra::hostInt32(L, "Idx", idx, idx->size(), vals)) return;
129
5/8
✓ Branch 0 taken 10 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 10 times.
✗ Branch 3 not taken.
✓ Branch 4 taken 10 times.
✗ Branch 5 not taken.
✓ Branch 6 taken 6 times.
✓ Branch 7 taken 4 times.
 10 if (!extra::needIndicesIn(L, "Idx", vals, idx->size(), 0, T(X)->rows)) return;
130 6 Tensor scratch;
131
4/8
✓ 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.
✓ Branch 6 taken 6 times.
✗ Branch 7 not taken.
 6 brotensor::gather_rows(*toTensor(X), extra::int32Operand(*idx, vals, scratch), *toTensor(Y));
132
2/15
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
✓ Branch 2 taken 6 times.
✗ 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.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
 10 BROTENSOR_API_CATCH(L)
133 10 }
134
135 // dY is (M, C) with M = Idx's length; every index names one of dX's R rows.
136 4 void bro_tensor_scatterRowsAdd(void* dY, void* Idx, int32_t R, void* dX) {
137 4 const char* L = "scatterRowsAdd";
138
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 4 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 4 times.
 4 if (!need(L, {dY, Idx, dX}) || !needNonNegative(L, {R})) return;
139 4 const Tensor* idx = T(Idx);
140 4 const Tensor* dy = T(dY);
141 BROTENSOR_API_TRY
142 4 std::vector<int32_t> vals;
143
2/4
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 4 times.
✗ Branch 3 not taken.
 4 if (!extra::hostInt32(L, "Idx", idx, dy->rows, vals)) return;
144
3/4
✓ Branch 0 taken 4 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✓ Branch 3 taken 2 times.
 4 if (!extra::needIndicesIn(L, "Idx", vals, dy->rows, 0, R)) return;
145 2 Tensor scratch;
146
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.
 2 brotensor::scatter_rows_add(*toTensor(dY), extra::int32Operand(*idx, vals, scratch), R, *toTensor(dX));
147
2/15
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✓ Branch 2 taken 2 times.
✗ 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.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
✗ Branch 14 not taken.
 4 BROTENSOR_API_CATCH(L)
148 4 }
149
150 6 void bro_tensor_topKRows(void* X, int32_t k, void* Vals, void* Idx) {
151
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 6 times.
 6 if (!need("topKRows", {X, Vals, Idx})) return;
152 BROTENSOR_API_TRY
153
5/8
✓ 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.
✓ Branch 6 taken 4 times.
✓ Branch 7 taken 2 times.
 6 brotensor::top_k_rows(*toTensor(X), k, *toTensor(Vals), *toTensor(Idx));
154
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.
 6 BROTENSOR_API_CATCH("topKRows")
155 8 }
156
157 // ---- batched LayerNorm with training caches --------------------------------
158
159 2 void bro_tensor_layernormForwardBatchedWithCaches(void* X, void* gamma, void* beta, void* Y, void* Xhat, void* Mean, void* Rstd, double eps) {
160 2 const char* L = "layernormForwardBatchedWithCaches";
161
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!need(L, {X, gamma, beta, Y, Xhat, Mean, Rstd})) return;
162
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!needOperands(L, T(X), {{"gamma", T(gamma), T(X)->cols}, {"beta", T(beta), T(X)->cols}})) return;
163 BROTENSOR_API_TRY
164
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::layernorm_forward_batched_with_caches(*toTensor(X), *toTensor(gamma), *toTensor(beta),
165
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 2 *toTensor(Y), *toTensor(Xhat),
166
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 2 *toTensor(Mean), *toTensor(Rstd),
167 2 static_cast<float>(eps));
168
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("layernormForwardBatchedWithCaches")
169 2 }
170
171 2 void bro_tensor_layernormBackwardBatchedWithCaches(void* dY, void* Xhat, void* gamma, void* Rstd, void* dX, void* dGamma, void* dBeta) {
172 2 const char* L = "layernormBackwardBatchedWithCaches";
173
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
 2 if (!need(L, {dY, Xhat, gamma, Rstd, dX, dGamma, dBeta})) return;
174 // dY/Xhat (R,D); gamma, dGamma, dBeta (D,) — the last two accumulate; Rstd (R,1) FP32.
175 2 const Tensor* dy = T(dY);
176
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!needPair(L, "Xhat", T(Xhat), dy)) return;
177
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 2 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 2 times.
 4 if (!needOperands(L, dy, {{"gamma", T(gamma), dy->cols}, {"dGamma", T(dGamma), dy->cols},
178 2 {"dBeta", T(dBeta), dy->cols}})) return;
179
1/2
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
 2 if (!needMask(L, T(Rstd), dy->rows, "Rstd")) return;
180 BROTENSOR_API_TRY
181
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::layernorm_backward_batched_with_caches(*toTensor(dY), *toTensor(Xhat), *toTensor(gamma),
182
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 2 *toTensor(Rstd), *toTensor(dX),
183
2/4
✓ Branch 0 taken 2 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 2 times.
✗ Branch 3 not taken.
 2 *toTensor(dGamma), *toTensor(dBeta));
184
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("layernormBackwardBatchedWithCaches")
185 2 }
186
187 } // extern "C"
188
189 namespace brotensor::api {
190
191 3 bool registerTensorNatives_batched(std::string* error) {
192 using namespace brotensor::api::reg;
193 3 return
194 6 fn("__bro_native.tensor.linearForwardBatched", p(&bro_tensor_linearForwardBatched), "void",
195
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, kTensorCls, kTensorCls}, error) &&
196 6 fn("__bro_native.tensor.linearForwardBatchedFp16", p(&bro_tensor_linearForwardBatchedFp16), "void",
197
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kDyn, kTensorCls, kTensorCls}, error) &&
198 6 fn("__bro_native.tensor.reluForwardBatched", p(&bro_tensor_reluForwardBatched), "void",
199
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls}, error) &&
200 6 fn("__bro_native.tensor.tanhForwardBatched", p(&bro_tensor_tanhForwardBatched), "void",
201
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls}, error) &&
202 6 fn("__bro_native.tensor.addInplaceBatched", p(&bro_tensor_addInplaceBatched), "void",
203
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls}, error) &&
204 6 fn("__bro_native.tensor.linearBackwardBatched", p(&bro_tensor_linearBackwardBatched), "void",
205
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls}, error) &&
206 6 fn("__bro_native.tensor.reluBackwardBatched", p(&bro_tensor_reluBackwardBatched), "void",
207
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, kTensorCls}, error) &&
208 6 fn("__bro_native.tensor.tanhBackwardBatched", p(&bro_tensor_tanhBackwardBatched), "void",
209
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, kTensorCls}, error) &&
210 6 fn("__bro_native.tensor.gatherRows", p(&bro_tensor_gatherRows), "void",
211
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, kTensorCls}, error) &&
212 6 fn("__bro_native.tensor.scatterRowsAdd", p(&bro_tensor_scatterRowsAdd), "void",
213
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, kTensorCls, "i32", kTensorCls}, error) &&
214 6 fn("__bro_native.tensor.topKRows", p(&bro_tensor_topKRows), "void",
215
2/4
✓ Branch 0 taken 3 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 3 times.
✗ Branch 3 not taken.
 6 {kTensorCls, "i32", kTensorCls, kTensorCls}, error) &&
216 6 fn("__bro_native.tensor.layernormForwardBatchedWithCaches", p(&bro_tensor_layernormForwardBatchedWithCaches), "void",
217
2/4
✗ Branch 0 not taken.
✓ Branch 1 taken 3 times.
✗ Branch 2 not taken.
✓ Branch 3 taken 3 times.
 6 {kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls, "f64"}, error) &&
218 6 fn("__bro_native.tensor.layernormBackwardBatchedWithCaches", p(&bro_tensor_layernormBackwardBatchedWithCaches), "void",
219 3 {kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls, kTensorCls}, error);
220 }
221
222 } // namespace brotensor::api
223