Skip to content

Commit e028871

Browse files
shlmregevrameshkunasiveblushyinzara
authored
Missing signal opts (#3264)
* Add missing optimizations for xtensa/ and hexagon/ to forntend ops under signal/ * Revert change to copyright year * Fix wrong include paths * Reorder include paths * Move static inline functions out of the namespace block. * Fix path to KissFFT * Remove file that was added by mistake * Realized a lot of these optimizations were already added under src/ * Reverted a few more files * Fix compilation errors * Porting Reduce_All reference operator porting from TFLite to TFLM (#3269) * Sync files related to Reverse_V2 from TFLite #3110 * PRelu Int16x8 support in RefC * Fix code style in prelu_test.cc * 1. Reverted the copyright year * Resolved compilation error for Int8x8 test case * Add Dynamic_Update_Slice support to TFLM * Code style error correction * Code style correction * Replaced hard coded MaxDimensions to RuntimeShape::kMaxSmallSize * 1. Added more test cases \n2.Removed unused code * Updates for test failure on ARM * Code style updates * Updates on test case failure for ARM * Updates on test case failure for ARM * Code style updates * Add Reduce_All reference operator support to TFLM * Resolving HiFi build errors --------- Co-authored-by: Esun Kim <veblush@google.com> * Provide default values for uninitialized variable (#3282) Switch-cases in decode_state_lut.cc don't assign a default value to a local variables. On one version of ARM GCC (building for ARM cortex m33, this results in a compiler error [-Werror=maybe-uninitialized]. BUG=451462435 * Add Ingenic MIPS port to README.md (#3255) * Add Ingenic MIPS port to README.md * Update README.md Changed order of new board to be alphabetical * Update README.md Removed extra brackets making things look wrong. --------- Co-authored-by: Esun Kim <veblush@google.com> * Solve compiler errors in decode op (#3284) 1. When building with old Xtensa toolchains, the compiler throw an unused variable warning which is treated as error (-Werror is defined by default). The cause is ScopedMicroProfile instantiation in the decoder op. Added a dummy reference to mute the warning. 2. Old pre C++14 Xtensa compilers don't support ticks in preprocessor constants, e.g. 0x8000'0000 Removed the ticks. BUG=451462435 * Fix coding style * Match FFT int16 test's tolerance with audiofrontend's * Fix failure in micro_speech example * Increase tolerance of feature extraction * Incerased error interval for micro speech detection * Incerased error interval for micro speech detection * Incerased error interval for micro speech detection * Incerased error interval for micro speech detection * Comment on looser tolerance in micro speech test and limit it to Xtensa --------- Co-authored-by: Kunasi Ramesh <28750242+rameshkunasi@users.noreply.github.com> Co-authored-by: Esun Kim <veblush@google.com> Co-authored-by: Matthew Vance <yinzara@gmail.com>
1 parent e1a6329 commit e028871

9 files changed

Lines changed: 1257 additions & 2 deletions

File tree

‎signal/micro/kernels/fft_test.cc‎

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -275,7 +275,7 @@ TF_LITE_MICRO_TEST(RfftTestSize512Int16) {
275275
tflite::tflm_signal::Register_RFFT_INT16();
276276
// See (b/287518815) for why this is needed.
277277
#if defined(HIFI3) || defined(HIFI4) || defined(HIFI5)
278-
int tolerance = 9;
278+
int tolerance = 16;
279279
#else // defined(HIFI3) || defined(HIFI4) || defined(HIFI5)
280280
int tolerance = 3;
281281
#endif // defined(HIFI3) || defined(HIFI4) || defined(HIFI5)

‎signal/micro/kernels/hexagon/hexagon_square_root.S‎

Lines changed: 110 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,110 @@
1+
/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.
2+
3+
Licensed under the Apache License, Version 2.0 (the "License");
4+
you may not use this file except in compliance with the License.
5+
You may obtain a copy of the License at
6+
7+
http://www.apache.org/licenses/LICENSE-2.0
8+
9+
Unless required by applicable law or agreed to in writing, software
10+
distributed under the License is distributed on an "AS IS" BASIS,
11+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12+
See the License for the specific language governing permissions and
13+
limitations under the License.
14+
==============================================================================*/
15+
16+
.section .note.GNU-stack,"",@progbits
17+
18+
// SignalHexagonSqrt32
19+
// input: R0 unsigned 32-bit
20+
// output: R0 unsigned 32-bit
21+
//
22+
// The assembly routine below implements the following:
23+
//
24+
// uint16_t Sqrt32(uint32_t num) {
25+
// uint32_t res = 0;
26+
// uint32_t bit = ((int32_t)1) << 30U;
27+
// while (bit > num)
28+
// bit >>= 2;
29+
// while (bit != 0) {
30+
// if (num >= res + bit) {
31+
// num -= res + bit;
32+
// res = (res >> 1U) + bit;
33+
// } else {
34+
// res >>= 1U;
35+
// }
36+
// bit >>= 2U;
37+
// }
38+
// // Do rounding
39+
// if (num > res && num != 0xFFFF)
40+
// ++res;
41+
// return res;
42+
// }
43+
44+
.text
45+
.p2align 2
46+
.p2align 4,,15
47+
.globl SignalHexagonSqrt32
48+
.type SignalHexagonSqrt32, @function
49+
50+
// Register mnemonics
51+
#define num R0 // input - as in loop above
52+
#define res R1 // as in loop aboe
53+
#define bit R2 // as in loop above
54+
#define temp R3 // the quantity bit + res
55+
#define zcount R4 // leading zeroes
56+
#define res_shift R5 // the quantity res >> 1
57+
#define bit_shift R6 // the quantity bit >> 2
58+
59+
SignalHexagonSqrt32:
60+
// Set bit to the largest even-power of two
61+
// that is less than or equal to the input
62+
{
63+
res = #0 // return value
64+
bit = ##1073741824 // 2^30
65+
zcount = cl0(num) // count leading zeroes
66+
}
67+
zcount = clrbit(zcount, #0) // even power of 2
68+
{
69+
bit = lsr(bit, zcount) // 2^30 right shifted
70+
if (cmp.eq(bit.new, #0)) jump:nt .done // return if bit == 0
71+
}
72+
.falign
73+
.loop:
74+
{
75+
// Calculate quantities to be used in the conditional below
76+
temp = add(bit, res)
77+
res_shift = lsr(res, #1)
78+
bit_shift = lsr(bit, #2)
79+
}
80+
{
81+
// Conditionally assign to num and res
82+
p0 = cmp.ltu(temp, num)
83+
if (p0.new) num = sub(num, temp)
84+
if (p0.new) res = add(res_shift, bit)
85+
if (!p0.new) res = res_shift
86+
}
87+
{
88+
// Advance bit >> 2 and exit loop if done
89+
bit = bit_shift
90+
if (cmp.gt(bit.new, #0)) jump:t .loop
91+
}
92+
.falign
93+
.done:
94+
// if (num > res && res != 0xffff) {
95+
// ++res
96+
// }
97+
// return res in num (R0)
98+
{
99+
temp = ##65535
100+
}
101+
{
102+
p0 = cmp.gt(num, res)
103+
p0 = !cmp.eq(res, temp)
104+
if (p0.new) num = add(res, #1)
105+
if (!p0.new) num = res
106+
}
107+
{
108+
jumpr r31
109+
}
110+
.size SignalHexagonSqrt32, .-SignalHexagonSqrt32

‎signal/micro/kernels/hexagon/rfft_int16.cc‎

Lines changed: 187 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,187 @@
1+
/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.
2+
3+
Licensed under the Apache License, Version 2.0 (the "License");
4+
you may not use this file except in compliance with the License.
5+
You may obtain a copy of the License at
6+
7+
http://www.apache.org/licenses/LICENSE-2.0
8+
9+
Unless required by applicable law or agreed to in writing, software
10+
distributed under the License is distributed on an "AS IS" BASIS,
11+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12+
See the License for the specific language governing permissions and
13+
limitations under the License.
14+
==============================================================================*/
15+
#include <stddef.h>
16+
#include <stdint.h>
17+
#include <string.h>
18+
19+
#include "signal/src/msb.h"
20+
#include "signal/src/rfft.h"
21+
22+
// Do not reorder these headers. "typedef.h" must appear before rfft.h
23+
#include "typedef.h"
24+
extern "C" {
25+
#include "rfft.h"
26+
}
27+
28+
namespace tflm_signal {
29+
30+
// TODO(b/467010877) The twiddle tables should come from the tflite file
31+
// This array includes the first 3*512/4=384 elements in the twiddle table in:
32+
// HEXAGON_Tools/<TOOL_VERSION>/Examples/libcore/SigProc/cxFFT_IFFT/include
33+
const uint16_t twiddles[] __attribute__((aligned(8))) = {
34+
0x7fff, 0x0000, 0x0000, 0x8000, 0x0000, 0x8000, 0x0000, 0x8000, 0xa57e,
35+
0xa57e, 0xa57e, 0x5a82, 0x5a82, 0xa57e, 0xcf05, 0x89bf, 0x89bf, 0xcf05,
36+
0xa57e, 0xa57e, 0x89bf, 0xcf05, 0x30fc, 0x7642, 0x7642, 0xcf05, 0xe708,
37+
0x8276, 0xb8e4, 0x9593, 0xcf05, 0x89bf, 0x9593, 0xb8e4, 0xe708, 0x7d8a,
38+
0x30fc, 0x89bf, 0xb8e4, 0x9593, 0x8276, 0x18f9, 0x89bf, 0xcf05, 0x8276,
39+
0xe708, 0x6a6e, 0x471d, 0x7d8a, 0xe708, 0xf375, 0x809e, 0xdad8, 0x8583,
40+
0xe708, 0x8276, 0x9d0e, 0xaecd, 0xc3aa, 0x70e3, 0x471d, 0x9593, 0xc3aa,
41+
0x8f1e, 0x809e, 0xf375, 0x9593, 0xb8e4, 0x8583, 0xdad8, 0x5134, 0x62f2,
42+
0x6a6e, 0xb8e4, 0xdad8, 0x8583, 0x9d0e, 0xaecd, 0xb8e4, 0x9593, 0x8f1e,
43+
0xc3aa, 0x0c8c, 0x7f62, 0x18f9, 0x8276, 0xaecd, 0x9d0e, 0x8f1e, 0x3c57,
44+
0x8276, 0xe708, 0x809e, 0xf375, 0x7a7d, 0x2528, 0x7f62, 0xf375, 0xf9b9,
45+
0x8028, 0xed38, 0x8163, 0xf375, 0x809e, 0xa129, 0xaa0b, 0xb3c1, 0x66d0,
46+
0x5134, 0x9d0e, 0xc946, 0x8c4b, 0x83d7, 0xe0e7, 0x9d0e, 0xaecd, 0x877c,
47+
0xd4e1, 0x41ce, 0x6dca, 0x70e3, 0xc3aa, 0xe0e7, 0x83d7, 0xaa0b, 0xa129,
48+
0xc3aa, 0x8f1e, 0x9236, 0xbe32, 0xf9b9, 0x7fd9, 0x2528, 0x8583, 0xb3c1,
49+
0x9931, 0x877c, 0x2b1f, 0x8583, 0xdad8, 0x8163, 0xed38, 0x73b6, 0x36ba,
50+
0x7a7d, 0xdad8, 0xed38, 0x8163, 0xc946, 0x8c4b, 0xdad8, 0x8583, 0x9931,
51+
0xb3c1, 0xd4e1, 0x7885, 0x3c57, 0x8f1e, 0xbe32, 0x9236, 0x8028, 0x0648,
52+
0x8f1e, 0xc3aa, 0x83d7, 0xe0e7, 0x5ed7, 0x55f6, 0x62f2, 0xaecd, 0xd4e1,
53+
0x877c, 0x9236, 0xbe32, 0xaecd, 0x9d0e, 0x8c4b, 0xc946, 0x1f1a, 0x7c2a,
54+
0x0c8c, 0x809e, 0xaa0b, 0xa129, 0x9931, 0x4c40, 0x809e, 0xf375, 0x8028,
55+
0xf9b9, 0x7e9d, 0x12c8, 0x7fd9, 0xf9b9, 0xfcdc, 0x800a, 0xf696, 0x8059,
56+
0xf9b9, 0x8028, 0xa34c, 0xa7be, 0xac65, 0x60ec, 0x55f6, 0xa129, 0xcc22,
57+
0x8afc, 0x8676, 0xd7da, 0xa129, 0xaa0b, 0x8894, 0xd1ef, 0x398d, 0x7255,
58+
0x73b6, 0xc946, 0xe3f5, 0x831d, 0xb141, 0x9b18, 0xc946, 0x8c4b, 0x93dc,
59+
0xbb86, 0xf055, 0x7f0a, 0x2b1f, 0x877c, 0xb64c, 0x975a, 0x84a3, 0x2224,
60+
0x877c, 0xd4e1, 0x81e3, 0xea1e, 0x6f5f, 0x3f17, 0x7c2a, 0xe0e7, 0xf055,
61+
0x80f7, 0xd1ef, 0x8894, 0xe0e7, 0x83d7, 0x9b18, 0xb141, 0xcc22, 0x7505,
62+
0x41ce, 0x9236, 0xc0e9, 0x90a1, 0x800a, 0xfcdc, 0x9236, 0xbe32, 0x84a3,
63+
0xdddd, 0x5843, 0x5cb4, 0x66d0, 0xb3c1, 0xd7da, 0x8676, 0x975a, 0xb64c,
64+
0xb3c1, 0x9931, 0x8dab, 0xc674, 0x15e2, 0x7e1e, 0x12c8, 0x8163, 0xac65,
65+
0x9f14, 0x93dc, 0x447b, 0x8163, 0xed38, 0x8059, 0xf696, 0x7ce4, 0x1c0c,
66+
0x7e9d, 0xed38, 0xf696, 0x8059, 0xe3f5, 0x831d, 0xed38, 0x8163, 0x9f14,
67+
0xac65, 0xbb86, 0x6c24, 0x4c40, 0x9931, 0xc674, 0x8dab, 0x81e3, 0xea1e,
68+
0x9931, 0xb3c1, 0x8676, 0xd7da, 0x49b4, 0x68a7, 0x6dca, 0xbe32, 0xdddd,
69+
0x84a3, 0xa34c, 0xa7be, 0xbe32, 0x9236, 0x90a1, 0xc0e9, 0x0324, 0x7ff6,
70+
0x1f1a, 0x83d7, 0xb141, 0x9b18, 0x8afc, 0x33df, 0x83d7, 0xe0e7, 0x80f7,
71+
0xf055, 0x776c, 0x2e11, 0x7885, 0xd4e1, 0xea1e, 0x81e3, 0xc0e9, 0x90a1,
72+
0xd4e1, 0x877c, 0x975a, 0xb64c, 0xdddd, 0x7b5d, 0x36ba, 0x8c4b, 0xbb86,
73+
0x93dc, 0x80f7, 0x0fab, 0x8c4b, 0xc946, 0x831d, 0xe3f5, 0x64e9, 0x4ec0,
74+
0x5ed7, 0xaa0b, 0xd1ef, 0x8894, 0x8dab, 0xc674, 0xaa0b, 0xa129, 0x8afc,
75+
0xcc22, 0x2827, 0x798a, 0x0648, 0x8028, 0xa7be, 0xa34c, 0x9f14, 0x539b,
76+
0x8028, 0xf9b9, 0x800a, 0xfcdc, 0x7fa7, 0x096b, 0x7ff6, 0xfcdc, 0xfe6e,
77+
0x8003, 0xfb4a, 0x8017, 0xfcdc, 0x800a, 0xa463, 0xa69c, 0xa8e3, 0x5dc8,
78+
0x5843, 0xa34c, 0xcd92, 0x8a5b, 0x8806, 0xd368, 0xa34c, 0xa7be, 0x8927,
79+
0xd079, 0x354e, 0x7460, 0x7505, 0xcc22, 0xe57e, 0x82c7, 0xb505, 0x9843,
80+
0xcc22, 0x8afc, 0x94b6, 0xba33, 0xebab, 0x7e60, 0x2e11, 0x8894, 0xb797,
81+
0x9674, 0x8377, 0x1d93, 0x8894, 0xd1ef, 0x822a, 0xe893, 0x6cf9, 0x4326,
82+
0x7ce4, 0xe3f5, 0xf1e5, 0x80c8, 0xd65d, 0x86f7, 0xe3f5, 0x831d, 0x9c11,
83+
0xb005, 0xc7dc, 0x7308, 0x447b, 0x93dc, 0xc248, 0x8fdd, 0x803e, 0xf827,
84+
0x93dc, 0xbb86, 0x8511, 0xdc5a, 0x54ca, 0x5fe4, 0x68a7, 0xb64c, 0xd958,
85+
0x85fb, 0x9a23, 0xb27f, 0xb64c, 0x975a, 0x8e62, 0xc50e, 0x113a, 0x7ed6,
86+
0x15e2, 0x81e3, 0xad97, 0x9e0f, 0x916a, 0x4074, 0x81e3, 0xea1e, 0x8079,
87+
0xf505, 0x7bc6, 0x209f, 0x7f0a, 0xf055, 0xf827, 0x803e, 0xe893, 0x822a,
88+
0xf055, 0x80f7, 0xa01d, 0xab36, 0xb797, 0x698c, 0x4ec0, 0x9b18, 0xc7dc,
89+
0x8cf9, 0x82c7, 0xe57e, 0x9b18, 0xb141, 0x86f7, 0xd65d, 0x45cd, 0x6b4b,
90+
0x6f5f, 0xc0e9, 0xdf61, 0x843b, 0xa69c, 0xa463, 0xc0e9, 0x90a1, 0x916a,
91+
0xbf8d, 0xfe6e, 0x7ffe, 0x2224, 0x84a3, 0xb27f, 0x9a23, 0x8927, 0x2f87,
92+
0x84a3, 0xdddd, 0x812b, 0xeec7, 0x75a6, 0x326e, 0x798a, 0xd7da, 0xebab,
93+
0x81a1, 0xc50e, 0x8e62, 0xd7da, 0x8676, 0x9843, 0xb505, 0xd958, 0x7a06,
94+
0x398d, 0x8dab, 0xbcdb, 0x9307, 0x8079, 0x0afb, 0x8dab, 0xc674, 0x8377,
95+
0xe26d, 0x61f1, 0x5269, 0x60ec, 0xac65, 0xd368, 0x8806, 0x8fdd, 0xc248,
96+
0xac65, 0x9f14, 0x8ba1, 0xcab3, 0x23a7, 0x7aef, 0x096b, 0x8059, 0xa8e3,
97+
0xa239, 0x9c11, 0x4ffb, 0x8059, 0xf696, 0x8017, 0xfb4a, 0x7f38, 0x0e1c,
98+
0x7fa7, 0xf696, 0xfb4a, 0x8017, 0xf1e5, 0x80c8, 0xf696, 0x8059, 0xa239,
99+
0xa8e3, 0xb005, 0x63ef, 0x539b, 0x9f14, 0xcab3, 0x8ba1, 0x8511, 0xdc5a,
100+
0x9f14, 0xac65, 0x8806, 0xd368, 0x3db8, 0x7023, 0x7255, 0xc674, 0xe26d,
101+
0x8377, 0xad97, 0x9e0f, 0xc674, 0x8dab, 0x9307, 0xbcdb, 0xf505, 0x7f87,
102+
0x2827, 0x8676, 0xb505, 0x9843, 0x85fb, 0x26a8, 0x8676, 0xd7da, 0x81a1,
103+
0xebab, 0x719e, 0x3af3, 0x7b5d, 0xdddd, 0xeec7, 0x812b, 0xcd92, 0x8a5b,
104+
0xdddd, 0x84a3, 0x9a23, 0xb27f, 0xd079, 0x76d9, 0x3f17, 0x90a1, 0xbf8d,
105+
0x916a, 0x8003, 0x0192, 0x90a1, 0xc0e9, 0x843b, 0xdf61, 0x5b9d, 0x5964,
106+
0x64e9, 0xb141, 0xd65d, 0x86f7, 0x94b6, 0xba33, 0xb141, 0x9b18, 0x8cf9,
107+
0xc7dc, 0x1a83, 0x7d3a, 0x0fab, 0x80f7, 0xab36, 0xa01d, 0x9674, 0x486a,
108+
0x80f7, 0xf055, 0x803e, 0xf827, 0x7dd6, 0x176e, 0x7e1e, 0xea1e, 0xf505,
109+
0x8079, 0xdf61, 0x843b, 0xea1e, 0x81e3, 0x9e0f, 0xad97, 0xbf8d, 0x6e97,
110+
0x49b4, 0x975a, 0xc50e, 0x8e62, 0x812b, 0xeec7, 0x975a, 0xb64c, 0x85fb,
111+
0xd958, 0x4d81, 0x65de, 0x6c24, 0xbb86, 0xdc5a, 0x8511, 0xa01d, 0xab36,
112+
0xbb86, 0x93dc, 0x8fdd, 0xc248, 0x07d9, 0x7fc2, 0x1c0c, 0x831d, 0xb005,
113+
0x9c11, 0x8cf9, 0x3825, 0x831d, 0xe3f5, 0x80c8, 0xf1e5, 0x790a, 0x29a4,
114+
0x776c, 0xd1ef, 0xe893, 0x822a, 0xbcdb, 0x9307, 0xd1ef, 0x8894, 0x9674,
115+
0xb797, 0xe26d, 0x7c89, 0x33df, 0x8afc, 0xba33, 0x94b6, 0x81a1, 0x1455,
116+
0x8afc, 0xcc22, 0x82c7, 0xe57e, 0x67bd, 0x4afb, 0x5cb4, 0xa7be, 0xd079,
117+
0x8927, 0x8ba1, 0xcab3, 0xa7be, 0xa34c, 0x8a5b, 0xcd92, 0x2c99, 0x77fb,
118+
0x0324, 0x800a, 0xa69c, 0xa463, 0xa239, 0x571e, 0x800a, 0xfcdc, 0x8003,
119+
0xfe6e, 0x7fea, 0x04b6};
120+
121+
// Twiddle factors used for the last stage of N-point real FFT
122+
// generated as j*W^k, k=1, 2, ... N/4
123+
// That's 128 complex int16_t elements
124+
// Or 256 real int16_t elements
125+
const uint16_t rtwiddles[] __attribute__((aligned(8))) = {
126+
0x0192, 0x7ffe, 0x0324, 0x7ff6, 0x04b6, 0x7fea, 0x0648, 0x7fd9, 0x07d9,
127+
0x7fc2, 0x096b, 0x7fa7, 0x0afb, 0x7f87, 0x0c8c, 0x7f62, 0x0e1c, 0x7f38,
128+
0x0fab, 0x7f0a, 0x113a, 0x7ed6, 0x12c8, 0x7e9d, 0x1455, 0x7e60, 0x15e2,
129+
0x7e1e, 0x176e, 0x7dd6, 0x18f9, 0x7d8a, 0x1a83, 0x7d3a, 0x1c0c, 0x7ce4,
130+
0x1d93, 0x7c89, 0x1f1a, 0x7c2a, 0x209f, 0x7bc6, 0x2224, 0x7b5d, 0x23a7,
131+
0x7aef, 0x2528, 0x7a7d, 0x26a8, 0x7a06, 0x2827, 0x798a, 0x29a4, 0x790a,
132+
0x2b1f, 0x7885, 0x2c99, 0x77fb, 0x2e11, 0x776c, 0x2f87, 0x76d9, 0x30fc,
133+
0x7642, 0x326e, 0x75a6, 0x33df, 0x7505, 0x354e, 0x7460, 0x36ba, 0x73b6,
134+
0x3825, 0x7308, 0x398d, 0x7255, 0x3af3, 0x719e, 0x3c57, 0x70e3, 0x3db8,
135+
0x7023, 0x3f17, 0x6f5f, 0x4074, 0x6e97, 0x41ce, 0x6dca, 0x4326, 0x6cf9,
136+
0x447b, 0x6c24, 0x45cd, 0x6b4b, 0x471d, 0x6a6e, 0x486a, 0x698c, 0x49b4,
137+
0x68a7, 0x4afb, 0x67bd, 0x4c40, 0x66d0, 0x4d81, 0x65de, 0x4ec0, 0x64e9,
138+
0x4ffb, 0x63ef, 0x5134, 0x62f2, 0x5269, 0x61f1, 0x539b, 0x60ec, 0x54ca,
139+
0x5fe4, 0x55f6, 0x5ed7, 0x571e, 0x5dc8, 0x5843, 0x5cb4, 0x5964, 0x5b9d,
140+
0x5a82, 0x5a82, 0x5b9d, 0x5964, 0x5cb4, 0x5843, 0x5dc8, 0x571e, 0x5ed7,
141+
0x55f6, 0x5fe4, 0x54ca, 0x60ec, 0x539b, 0x61f1, 0x5269, 0x62f2, 0x5134,
142+
0x63ef, 0x4ffb, 0x64e9, 0x4ec0, 0x65de, 0x4d81, 0x66d0, 0x4c40, 0x67bd,
143+
0x4afb, 0x68a7, 0x49b4, 0x698c, 0x486a, 0x6a6e, 0x471d, 0x6b4b, 0x45cd,
144+
0x6c24, 0x447b, 0x6cf9, 0x4326, 0x6dca, 0x41ce, 0x6e97, 0x4074, 0x6f5f,
145+
0x3f17, 0x7023, 0x3db8, 0x70e3, 0x3c57, 0x719e, 0x3af3, 0x7255, 0x398d,
146+
0x7308, 0x3825, 0x73b6, 0x36ba, 0x7460, 0x354e, 0x7505, 0x33df, 0x75a6,
147+
0x326e, 0x7642, 0x30fc, 0x76d9, 0x2f87, 0x776c, 0x2e11, 0x77fb, 0x2c99,
148+
0x7885, 0x2b1f, 0x790a, 0x29a4, 0x798a, 0x2827, 0x7a06, 0x26a8, 0x7a7d,
149+
0x2528, 0x7aef, 0x23a7, 0x7b5d, 0x2224, 0x7bc6, 0x209f, 0x7c2a, 0x1f1a,
150+
0x7c89, 0x1d93, 0x7ce4, 0x1c0c, 0x7d3a, 0x1a83, 0x7d8a, 0x18f9, 0x7dd6,
151+
0x176e, 0x7e1e, 0x15e2, 0x7e60, 0x1455, 0x7e9d, 0x12c8, 0x7ed6, 0x113a,
152+
0x7f0a, 0x0fab, 0x7f38, 0x0e1c, 0x7f62, 0x0c8c, 0x7f87, 0x0afb, 0x7fa7,
153+
0x096b, 0x7fc2, 0x07d9, 0x7fd9, 0x0648, 0x7fea, 0x04b6, 0x7ff6, 0x0324,
154+
0x7ffe, 0x0192, 0x7fff, 0x0000};
155+
156+
struct RfftState {
157+
int16_t* aligned_input;
158+
int32_t fft_length;
159+
};
160+
161+
size_t RfftInt16GetNeededMemory(int32_t fft_length) {
162+
return sizeof(RfftState) + 2 * sizeof(int16_t) * fft_length;
163+
}
164+
165+
void* RfftInt16Init(int32_t fft_length, void* state, size_t state_size) {
166+
RfftState* rfft_state = (RfftState*)state;
167+
int16_t* unaligned_buffer = (int16_t*)(rfft_state + 1);
168+
rfft_state->aligned_input =
169+
(int16_t*)((uint32_t)(unaligned_buffer + fft_length) &
170+
~((1 << tflite::tflm_signal::MostSignificantBit32(
171+
fft_length)) -
172+
1));
173+
rfft_state->fft_length = fft_length;
174+
return state;
175+
}
176+
177+
void RfftInt16Apply(void* state, const int16_t* input,
178+
Complex<int16_t>* output) {
179+
RfftState* rfft_state = (RfftState*)state;
180+
memcpy(rfft_state->aligned_input, input,
181+
rfft_state->fft_length * sizeof(int16_t));
182+
rfft(rfft_state->aligned_input, rfft_state->fft_length, (CWord2x16*)twiddles,
183+
(CWord2x16*)rtwiddles, (CWord2x16*)output);
184+
return;
185+
}
186+
187+
} // namespace tflm_signal

‎signal/micro/kernels/hexagon/square_root_32.cc‎

Lines changed: 28 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,28 @@
1+
/* Copyright 2021 The TensorFlow Authors. All Rights Reserved.
2+
3+
Licensed under the Apache License, Version 2.0 (the "License");
4+
you may not use this file except in compliance with the License.
5+
You may obtain a copy of the License at
6+
7+
http://www.apache.org/licenses/LICENSE-2.0
8+
9+
Unless required by applicable law or agreed to in writing, software
10+
distributed under the License is distributed on an "AS IS" BASIS,
11+
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
12+
See the License for the specific language governing permissions and
13+
limitations under the License.
14+
==============================================================================*/
15+
#include "signal/src/square_root.h"
16+
17+
extern "C" uint16_t SignalHexagonSqrt32(uint32_t num);
18+
19+
namespace tflite {
20+
namespace tflm_signal {
21+
22+
// SignalHexagonSqrt32() is defined in assembly. This C wrapper is only
23+
// necessary to force TFLM's source specialization to pick up the optimized
24+
// Hexagon implementation instead of the portable one.
25+
uint16_t Sqrt32(uint32_t num) { return SignalHexagonSqrt32(num); }
26+
27+
} // namespace tflm_signal
28+
} // namespace tflite

0 commit comments

Comments
 (0)