@geastack/targets 0.1.87 → 0.1.88

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (80) hide show
  1. package/package.json +7 -7
  2. package/targets/esp32/app_project.cmake +18 -1
  3. package/targets/esp32/chip_bindings/audio/echo_cancellation.cpp +213 -0
  4. package/targets/esp32/chip_bindings/audio/echo_cancellation.h +12 -0
  5. package/targets/esp32/chip_bindings/audio/es8311.cpp +265 -45
  6. package/targets/esp32/chip_bindings/displays/rgb_panel.cpp +115 -12
  7. package/targets/esp32/chip_bindings/expanders/io_expander.h +2 -0
  8. package/targets/esp32/chip_bindings/expanders/tca9554.cpp +74 -0
  9. package/targets/esp32/chip_bindings/touch/ft3168.cpp +23 -1
  10. package/targets/esp32/chip_bindings/touch/ft3168_controller.h +1 -0
  11. package/targets/esp32/chip_bindings/touch/gt911.cpp +4 -3
  12. package/targets/esp32/components/gea_framework/CMakeLists.txt +8 -1
  13. package/targets/esp32/components/gea_juice/CMakeLists.txt +46 -0
  14. package/targets/esp32/components/gea_juice/README.md +20 -0
  15. package/targets/esp32/components/gea_juice/juice_memory.c +19 -0
  16. package/targets/esp32/components/gea_juice/juice_memory.h +16 -0
  17. package/targets/esp32/components/gea_juice/patch_threads.py +50 -0
  18. package/targets/esp32/components/gea_vp8_peer/CMakeLists.txt +16 -0
  19. package/targets/esp32/components/gea_vp8_peer/README.md +39 -0
  20. package/targets/esp32/components/gea_vp8_peer/gea_vp8_peer.h +21 -0
  21. package/targets/esp32/components/gea_vp8_peer/vp8_peer.cpp +538 -0
  22. package/targets/esp32/components/gea_vpx/CMakeLists.txt +235 -0
  23. package/targets/esp32/components/gea_vpx/README.md +133 -0
  24. package/targets/esp32/components/gea_vpx/i420_rgb565_s3.S +169 -0
  25. package/targets/esp32/components/gea_vpx/vp8_blocks_s3.c +42 -0
  26. package/targets/esp32/components/gea_vpx/vp8_copy_s3.S +39 -0
  27. package/targets/esp32/components/gea_vpx/vp8_entropy_s3.c +76 -0
  28. package/targets/esp32/components/gea_vpx/vp8_filter_s3.S +58 -0
  29. package/targets/esp32/components/gea_vpx/vp8_idct_s3.S +269 -0
  30. package/targets/esp32/components/gea_vpx/vp8_intrapred_s3.S +65 -0
  31. package/targets/esp32/components/gea_vpx/vp8_loopfilter_s3.S +333 -0
  32. package/targets/esp32/components/gea_vpx/vp8_loopfilter_s3.c +62 -0
  33. package/targets/esp32/components/gea_vpx/vp8_profile_s3.c +22 -0
  34. package/targets/esp32/components/gea_vpx/vp8_profile_s3.h +20 -0
  35. package/targets/esp32/components/gea_vpx/vp8_s3.c +120 -0
  36. package/targets/esp32/components/gea_vpx/vp8_s3.h +58 -0
  37. package/targets/esp32/components/gea_vpx/vp8_workspace_s3.c +14 -0
  38. package/targets/esp32/components/gea_vpx/vpx_memory.c +41 -0
  39. package/targets/esp32/connectivity/diagnostics_transport.cpp +18 -0
  40. package/targets/esp32/connectivity/wifi.cpp +26 -13
  41. package/targets/esp32/crypto/srtp_hmac_psa.c +118 -0
  42. package/targets/esp32/display.cpp +13 -0
  43. package/targets/esp32/gea_framework.cmake +7 -0
  44. package/targets/esp32/host_sources.cmake +2 -0
  45. package/targets/esp32/i2c_completion_order.cmake +57 -0
  46. package/targets/esp32/patch_streaming_dependencies.py +233 -0
  47. package/targets/esp32/services/device_control.cpp +141 -5
  48. package/targets/esp32/services/frame_scheduler.cpp +29 -1
  49. package/targets/esp32/services/opus_benchmark.h +82 -0
  50. package/targets/esp32/services/pcm_simd_check.h +73 -0
  51. package/targets/esp32/services/srtp_benchmark.h +106 -0
  52. package/targets/esp32/services/video_simd_benchmark.h +477 -0
  53. package/targets/esp32/services/vp8_benchmark.h +147 -0
  54. package/targets/esp32/streaming_dependency_patches.cmake +103 -0
  55. package/targets/esp32-m5stack-m5paper/main/idf_component.yml +2 -0
  56. package/targets/esp32-p4-m5stack-tab5/main/idf_component.yml +2 -0
  57. package/targets/esp32-p4-waveshare-touch-lcd-3.5/main/idf_component.yml +2 -0
  58. package/targets/esp32-p4-waveshare-touch-lcd-7/main/idf_component.yml +2 -0
  59. package/targets/esp32-s3-elecrow-rotary-2.1/main/idf_component.yml +2 -0
  60. package/targets/esp32-s3-epaper-1.54/main/idf_component.yml +2 -0
  61. package/targets/esp32-s3-lilygo-t-display-s3-long/main/idf_component.yml +2 -0
  62. package/targets/esp32-s3-m5stack-papers3/main/idf_component.yml +2 -0
  63. package/targets/esp32-s3-m5stack-sticks3/main/idf_component.yml +2 -0
  64. package/targets/esp32-s3-touch-amoled-1.75/main/idf_component.yml +2 -0
  65. package/targets/esp32-s3-touch-amoled-1.8/dependencies.lock +110 -2
  66. package/targets/esp32-s3-touch-amoled-1.8/main/CMakeLists.txt +29 -3
  67. package/targets/esp32-s3-touch-amoled-1.8/main/app_main.cpp +6 -3
  68. package/targets/esp32-s3-touch-amoled-1.8/main/idf_component.yml +11 -0
  69. package/targets/esp32-s3-touch-amoled-1.8/sdkconfig.defaults +9 -1
  70. package/targets/esp32-s3-touch-amoled-2.06/CMakeLists.txt +2 -0
  71. package/targets/esp32-s3-touch-amoled-2.06/dependencies.lock +171 -8
  72. package/targets/esp32-s3-touch-amoled-2.06/main/CMakeLists.txt +8 -1
  73. package/targets/esp32-s3-touch-amoled-2.06/main/app_main.cpp +10 -3
  74. package/targets/esp32-s3-touch-amoled-2.06/main/idf_component.yml +16 -1
  75. package/targets/esp32-s3-touch-amoled-2.06/main/include/board.h +3 -0
  76. package/targets/esp32-s3-touch-amoled-2.06/main/touch.cpp +7 -0
  77. package/targets/esp32-s3-touch-amoled-2.41/dependencies.lock +57 -1
  78. package/targets/esp32-s3-touch-amoled-2.41/main/idf_component.yml +10 -0
  79. package/targets/esp32-s3-waveshare-touch-lcd-4b.json +504 -0
  80. package/targets.json +12 -1
@@ -0,0 +1,76 @@
1
+ // SPDX-License-Identifier: BSD-3-Clause
2
+ // Differential verification of the S3 entropy optimization against libvpx's
3
+ // original arithmetic-bit algorithm (libvpx 1.16.0, BSD-licensed).
4
+ #include "vp8_s3.h"
5
+ #include "vp8/decoder/dboolhuff.h"
6
+ #include "esp_timer.h"
7
+ #include <string.h>
8
+
9
+ static __attribute__((noinline)) int reference_bit(BOOL_DECODER *br, int probability) {
10
+ const unsigned split = 1 + (((br->range - 1) * probability) >> 8);
11
+ if (br->count < 0) vp8dx_bool_decoder_fill(br);
12
+ VP8_BD_VALUE value = br->value;
13
+ int count = br->count;
14
+ const VP8_BD_VALUE bigsplit = (VP8_BD_VALUE)split << (VP8_BD_VALUE_SIZE - 8);
15
+ unsigned range = split;
16
+ int bit = 0;
17
+ if (value >= bigsplit) {
18
+ range = br->range - split;
19
+ value -= bigsplit;
20
+ bit = 1;
21
+ }
22
+ const unsigned shift = vp8_norm[(unsigned char)range];
23
+ br->range = range << shift;
24
+ br->value = value << shift;
25
+ br->count = count - shift;
26
+ return bit;
27
+ }
28
+
29
+ static void decrypt(void *opaque, const unsigned char *input, unsigned char *output, int size) {
30
+ const unsigned char key = *(const unsigned char *)opaque;
31
+ for (int i = 0; i < size; ++i) output[i] = input[i] ^ key;
32
+ }
33
+
34
+ int gea_vp8_entropy_check_s3(struct gea_vp8_entropy_result *result) {
35
+ memset(result, 0, sizeof(*result));
36
+ // Normalization is defined only for positive ranges. Every range reachable
37
+ // from either arithmetic branch is covered, including probability 0/255.
38
+ for (unsigned range = 1; range <= 255; ++range)
39
+ if (vp8_norm[range] != __builtin_clz(range) - 24) return 0;
40
+ uint32_t random = 0xa154f63du;
41
+ unsigned char data[128], probabilities[256], key = 0x53;
42
+ for (unsigned test = 0; test < 4096; ++test) {
43
+ for (unsigned i = 0; i < sizeof(probabilities); ++i) {
44
+ random ^= random << 13; random ^= random >> 17; random ^= random << 5;
45
+ probabilities[i] = (unsigned char)random;
46
+ if (i < sizeof(data)) data[i] = (unsigned char)(random >> 8);
47
+ }
48
+ // Includes zero-length, truncated and encrypted inputs. Compare the
49
+ // entire arithmetic state after every bit, not only the decoded output.
50
+ BOOL_DECODER scalar, native;
51
+ vpx_decrypt_cb callback = test & 1 ? decrypt : NULL;
52
+ const unsigned length = test % (sizeof(data) + 1);
53
+ vp8dx_start_decode(&scalar, data, length, callback, &key);
54
+ native = scalar;
55
+ for (unsigned i = 0; i < sizeof(probabilities); ++i) {
56
+ const int expected = reference_bit(&scalar, probabilities[i]);
57
+ const int actual = vp8dx_decode_bool(&native, probabilities[i]);
58
+ if (actual != expected || native.value != scalar.value ||
59
+ native.range != scalar.range || native.count != scalar.count ||
60
+ native.user_buffer != scalar.user_buffer ||
61
+ vp8dx_bool_error(&native) != vp8dx_bool_error(&scalar)) return 0;
62
+ }
63
+ ++result->cases;
64
+ // Time batches, excluding comparison and per-bit timer overhead.
65
+ vp8dx_start_decode(&scalar, data, sizeof(data), NULL, NULL);
66
+ native = scalar;
67
+ int64_t started = esp_timer_get_time();
68
+ for (unsigned i = 0; i < sizeof(probabilities); ++i) reference_bit(&scalar, probabilities[i]);
69
+ result->scalar_us += esp_timer_get_time() - started;
70
+ started = esp_timer_get_time();
71
+ for (unsigned i = 0; i < sizeof(probabilities); ++i) vp8dx_decode_bool(&native, probabilities[i]);
72
+ result->native_us += esp_timer_get_time() - started;
73
+ if (native.value != scalar.value || native.count != scalar.count || native.range != scalar.range) return 0;
74
+ }
75
+ return 1;
76
+ }
@@ -0,0 +1,58 @@
1
+ // SPDX-License-Identifier: Apache-2.0
2
+ // Eight parallel VP8 subpixel convolutions (bilinear or six-tap).
3
+ // Source has libvpx's padded border; destination is at least 4-byte aligned.
4
+ .section .rodata.gea_vp8_filter_constants,"a",@progbits
5
+ .balign 4
6
+ .Lfilter_constants:
7
+ .short 64, 255
8
+ .text
9
+ .align 4
10
+ .global gea_vp8_filter_rows_s3
11
+ .type gea_vp8_filter_rows_s3,@function
12
+ gea_vp8_filter_rows_s3:
13
+ // src, source row stride, coefficients, dst, destination row stride,
14
+ // packed: height<<8 | vertical<<4 | eight_lanes<<3 | taps.
15
+ // Constants stay in PIE registers across every row of this block.
16
+ entry a1, 32
17
+ movi a8, .Lfilter_constants
18
+ ee.vldbc.16.ip q6, a8, 2
19
+ ee.vldbc.16 q5, a8
20
+ srli a9, a7, 8
21
+ movi a14, 1
22
+ bbci a7, 4, .Lfilter_row
23
+ mov a14, a3
24
+ .Lfilter_row:
25
+ mov a11, a2
26
+ mov a12, a4
27
+ extui a10, a7, 0, 3
28
+ ee.mov.s16.qacc q6
29
+ loopnez a10, .Lfilter_tap_end
30
+ .Lfilter_tap:
31
+ mov a8, a11
32
+ ee.ld.128.usar.ip q1, a8, 16
33
+ ee.ld.128.usar.ip q2, a8, 16
34
+ ee.src.q.qup q0, q1, q2
35
+ ee.zero.q q3
36
+ ee.vzip.8 q0, q3
37
+ ee.vldbc.16.ip q7, a12, 2
38
+ ee.vmulas.s16.qacc q0, q7
39
+ add a11, a11, a14
40
+ .Lfilter_tap_end:
41
+ movi a8, 7
42
+ ee.srcmb.s16.qacc q0, a8, 0
43
+ ee.zero.q q4
44
+ ee.vmax.s16 q0, q0, q4
45
+ ee.vmin.s16 q0, q0, q5
46
+ ee.vunzip.8 q0, q4
47
+ ee.movi.32.a q0, a8, 0
48
+ s32i a8, a5, 0
49
+ bbci a7, 3, .Lfilter_next_row
50
+ ee.movi.32.a q0, a8, 1
51
+ s32i a8, a5, 4
52
+ .Lfilter_next_row:
53
+ add a2, a2, a3
54
+ add a5, a5, a6
55
+ addi a9, a9, -1
56
+ bnez a9, .Lfilter_row
57
+ retw
58
+ .size gea_vp8_filter_rows_s3, .-gea_vp8_filter_rows_s3
@@ -0,0 +1,269 @@
1
+ // SPDX-License-Identifier: Apache-2.0
2
+ // Bit-exact VP8 4x4 inverse DCT. Four transforms run in parallel in 32-bit
3
+ // lanes: no intermediate saturating 16-bit additions, even for extreme inputs.
4
+ .section .rodata.gea_vp8_idct_constants,"a",@progbits
5
+ .balign 4
6
+ .Lidct_constants:
7
+ .short 20091, -30068
8
+ .long 4, 255
9
+
10
+ .macro sx16 reg
11
+ ee.vsl.32 \reg, \reg
12
+ ee.vsr.32 \reg, \reg
13
+ .endm
14
+
15
+ .macro transpose4
16
+ ee.vzip.32 q0, q2
17
+ ee.vzip.32 q1, q3
18
+ ee.vzip.32 q0, q1
19
+ ee.vzip.32 q2, q3
20
+ .endm
21
+
22
+ .macro idct_pass
23
+ // SAR=16. sin coefficient 35468 = 65536-30068, exactly, including
24
+ // negative rounding: (x*35468)>>16 == x + ((x*-30068)>>16).
25
+ ee.vadds.s32 q5, q0, q2
26
+ ee.vsubs.s32 q0, q0, q2
27
+ ee.vldbc.16 q4, a11
28
+ ee.vmul.s16 q6, q1, q4
29
+ sx16 q6
30
+ ee.vadds.s32 q6, q6, q1
31
+ ee.vldbc.16 q4, a7
32
+ ee.vmul.s16 q7, q3, q4
33
+ sx16 q7
34
+ ee.vadds.s32 q7, q7, q3
35
+ ee.vsubs.s32 q6, q6, q7
36
+ ee.vmul.s16 q7, q1, q4
37
+ sx16 q7
38
+ ee.vadds.s32 q1, q1, q7
39
+ ee.vldbc.16 q4, a11
40
+ ee.vmul.s16 q7, q3, q4
41
+ sx16 q7
42
+ ee.vadds.s32 q3, q3, q7
43
+ ee.vadds.s32 q1, q1, q3
44
+ ee.vsubs.s32 q2, q0, q6
45
+ ee.vadds.s32 q0, q0, q6
46
+ ee.vsubs.s32 q3, q5, q1
47
+ ee.vadds.s32 q1, q5, q1
48
+ ee.orq q4, q0, q0
49
+ ee.orq q0, q1, q1
50
+ ee.orq q1, q4, q4
51
+ .endm
52
+
53
+ .macro load_row reg
54
+ ee.vld.l.64.ip \reg, a2, 8
55
+ ee.zero.q q4
56
+ ee.vzip.16 \reg, q4
57
+ sx16 \reg
58
+ .endm
59
+
60
+ .macro output_row reg
61
+ // Scalar loads/stores only transfer a packed 4-pixel word. Prediction,
62
+ // saturation and packing are vector operations.
63
+ l32i a12, a3, 0
64
+ ee.movi.32.q q4, a12, 0
65
+ ee.zero.q q5
66
+ ee.vzip.8 q4, q5
67
+ ee.zero.q q5
68
+ ee.vzip.16 q4, q5
69
+ // ZIP overwrites both operands; its upper-half result is not a zero vector.
70
+ ee.zero.q q5
71
+ ee.vadds.s32 \reg, \reg, q4
72
+ ee.vmax.s32 \reg, \reg, q5
73
+ ee.vmin.s32 \reg, \reg, q6
74
+ ee.vunzip.16 \reg, q5
75
+ ee.zero.q q5
76
+ ee.vunzip.8 \reg, q5
77
+ ee.movi.32.a \reg, a12, 0
78
+ s32i a12, a5, 0
79
+ add a3, a3, a4
80
+ add a5, a5, a6
81
+ .endm
82
+
83
+ .text
84
+ .align 4
85
+ .global vp8_short_idct4x4llm_s3
86
+ .type vp8_short_idct4x4llm_s3,@function
87
+ vp8_short_idct4x4llm_s3:
88
+ // input 8-byte aligned; prediction/destination 4-byte aligned and strides.
89
+ entry a1, 32
90
+ movi a7, .Lidct_constants
91
+ addi a11, a7, 2
92
+ movi a8, 16
93
+ wsr.sar a8
94
+ load_row q0
95
+ load_row q1
96
+ load_row q2
97
+ load_row q3
98
+ .Lidct_loaded:
99
+ idct_pass
100
+ // Upstream stores the first pass into short[16]: wrap, do not saturate.
101
+ sx16 q0
102
+ sx16 q1
103
+ sx16 q2
104
+ sx16 q3
105
+ transpose4
106
+ idct_pass
107
+ addi a7, a7, 4
108
+ ee.vldbc.32.ip q4, a7, 4
109
+ ee.vadds.s32 q0, q0, q4
110
+ ee.vadds.s32 q1, q1, q4
111
+ ee.vadds.s32 q2, q2, q4
112
+ ee.vadds.s32 q3, q3, q4
113
+ movi a9, 3
114
+ wsr.sar a9
115
+ ee.vsr.32 q0, q0
116
+ ee.vsr.32 q1, q1
117
+ ee.vsr.32 q2, q2
118
+ ee.vsr.32 q3, q3
119
+ wsr.sar a8
120
+ sx16 q0
121
+ sx16 q1
122
+ sx16 q2
123
+ sx16 q3
124
+ transpose4
125
+ ee.vldbc.32 q6, a7
126
+ output_row q0
127
+ output_row q1
128
+ output_row q2
129
+ output_row q3
130
+ retw
131
+ .size vp8_short_idct4x4llm_s3, .-vp8_short_idct4x4llm_s3
132
+
133
+ .align 4
134
+ .global vp8_dequant_idct_add_s3
135
+ .type vp8_dequant_idct_add_s3,@function
136
+ vp8_dequant_idct_add_s3:
137
+ // input, quantizers, pixels, stride. Keep dequantized coefficients in
138
+ // PIE registers: no write/reload of an intermediate PSRAM block.
139
+ entry a1, 32
140
+ mov a10, a2
141
+ mov a13, a4
142
+ mov a14, a5
143
+ movi a8, 0
144
+ wsr.sar a8
145
+ ee.vld.128.ip q0, a2, 16
146
+ ee.vld.128.ip q1, a2, 16
147
+ ee.vld.128.ip q4, a3, 16
148
+ ee.vld.128.ip q5, a3, 16
149
+ ee.vmul.s16 q0, q0, q4
150
+ ee.vmul.s16 q1, q1, q5
151
+ ee.zero.q q2
152
+ ee.zero.q q3
153
+ ee.vzip.16 q0, q2
154
+ ee.vzip.16 q1, q3
155
+ ee.orq q4, q1, q1
156
+ ee.orq q1, q2, q2
157
+ ee.orq q2, q4, q4
158
+ movi a8, 16
159
+ wsr.sar a8
160
+ sx16 q0
161
+ sx16 q1
162
+ sx16 q2
163
+ sx16 q3
164
+ // Match upstream's coefficient clearing, with exactly two native stores.
165
+ ee.zero.q q5
166
+ ee.vst.128.ip q5, a10, 16
167
+ ee.vst.128.ip q5, a10, 16
168
+ mov a3, a13
169
+ mov a4, a14
170
+ mov a5, a13
171
+ mov a6, a14
172
+ movi a7, .Lidct_constants
173
+ addi a11, a7, 2
174
+ j .Lidct_loaded
175
+ .size vp8_dequant_idct_add_s3, .-vp8_dequant_idct_add_s3
176
+
177
+ .align 4
178
+ .global vp8_dc_only_idct_add_s3
179
+ .type vp8_dc_only_idct_add_s3,@function
180
+ vp8_dc_only_idct_add_s3:
181
+ entry a1, 32
182
+ // DC-only residual is bounded to [-4096,4096] after rounding, so eight
183
+ // signed 16-bit lanes suffice even with the 255-valued prediction. Full
184
+ // transforms still use 32-bit lanes to preserve their extreme-input ABI.
185
+ addi a2, a2, 4
186
+ srai a2, a2, 3
187
+ s16i a2, a1, 0
188
+ ee.vldbc.16 q0, a1
189
+ movi a7, .Lidct_constants+8
190
+ ee.vldbc.16 q6, a7
191
+ ee.zero.q q7
192
+ movi a9, 4
193
+ .Ldc_row:
194
+ l32i a8, a3, 0
195
+ ee.movi.32.q q1, a8, 0
196
+ ee.zero.q q2
197
+ ee.vzip.8 q1, q2
198
+ ee.vadds.s16 q1, q1, q0
199
+ ee.vmax.s16 q1, q1, q7
200
+ ee.vmin.s16 q1, q1, q6
201
+ ee.zero.q q2
202
+ ee.vunzip.8 q1, q2
203
+ ee.movi.32.a q1, a8, 0
204
+ s32i a8, a5, 0
205
+ add a3, a3, a4
206
+ add a5, a5, a6
207
+ addi a9, a9, -1
208
+ bnez a9, .Ldc_row
209
+ retw
210
+ .size vp8_dc_only_idct_add_s3, .-vp8_dc_only_idct_add_s3
211
+
212
+ .macro walsh_pass
213
+ ee.vadds.s32 q4, q0, q3
214
+ ee.vadds.s32 q5, q1, q2
215
+ ee.vsubs.s32 q6, q1, q2
216
+ ee.vsubs.s32 q7, q0, q3
217
+ ee.vadds.s32 q0, q4, q5
218
+ ee.vadds.s32 q1, q6, q7
219
+ ee.vsubs.s32 q2, q4, q5
220
+ ee.vsubs.s32 q3, q7, q6
221
+ .endm
222
+ .macro scatter_dc reg
223
+ ee.movi.32.a \reg, a4, 0
224
+ s16i a4, a3, 0
225
+ ee.movi.32.a \reg, a4, 1
226
+ s16i a4, a3, 32
227
+ ee.movi.32.a \reg, a4, 2
228
+ s16i a4, a3, 64
229
+ ee.movi.32.a \reg, a4, 3
230
+ s16i a4, a3, 96
231
+ addi a3, a3, 128
232
+ .endm
233
+ .align 4
234
+ .global vp8_short_inv_walsh4x4_s3
235
+ .type vp8_short_inv_walsh4x4_s3,@function
236
+ vp8_short_inv_walsh4x4_s3:
237
+ entry a1, 32
238
+ movi a8, 16
239
+ wsr.sar a8
240
+ load_row q0
241
+ load_row q1
242
+ load_row q2
243
+ load_row q3
244
+ walsh_pass
245
+ sx16 q0
246
+ sx16 q1
247
+ sx16 q2
248
+ sx16 q3
249
+ transpose4
250
+ walsh_pass
251
+ movi a4, 3
252
+ s32i a4, a1, 0
253
+ ee.vldbc.32 q4, a1
254
+ ee.vadds.s32 q0, q0, q4
255
+ ee.vadds.s32 q1, q1, q4
256
+ ee.vadds.s32 q2, q2, q4
257
+ ee.vadds.s32 q3, q3, q4
258
+ wsr.sar a4
259
+ ee.vsr.32 q0, q0
260
+ ee.vsr.32 q1, q1
261
+ ee.vsr.32 q2, q2
262
+ ee.vsr.32 q3, q3
263
+ transpose4
264
+ scatter_dc q0
265
+ scatter_dc q1
266
+ scatter_dc q2
267
+ scatter_dc q3
268
+ retw
269
+ .size vp8_short_inv_walsh4x4_s3, .-vp8_short_inv_walsh4x4_s3
@@ -0,0 +1,65 @@
1
+ // SPDX-License-Identifier: Apache-2.0
2
+ // True-motion intra prediction: clip(left[row] + above[column] - top_left).
3
+ // Eight columns run in signed 16-bit PIE lanes, with exact saturation to bytes.
4
+ .section .rodata.gea_vp8_tm,"a",@progbits
5
+ .balign 4
6
+ .Ltm_max:
7
+ .short 255
8
+ .text
9
+ .align 4
10
+ .global gea_vp8_tm_s3
11
+ .type gea_vp8_tm_s3,@function
12
+ gea_vp8_tm_s3:
13
+ // dst (4-byte aligned), stride, aligned above, left, block size (4/8/16).
14
+ entry a1, 32
15
+ addi a9, a4, -1
16
+ l8ui a8, a9, 0
17
+ ee.zero.q q5
18
+ movi a9, .Ltm_max
19
+ ee.vldbc.16 q6, a9
20
+ beqi a6, 4, .Ltm_four
21
+ ee.vld.l.64.ip q0, a4, 8
22
+ ee.vzip.8 q0, q5
23
+ beqi a6, 8, .Ltm_ready
24
+ ee.zero.q q5
25
+ ee.vld.l.64.ip q1, a4, 8
26
+ ee.vzip.8 q1, q5
27
+ j .Ltm_ready
28
+ .Ltm_four:
29
+ l32i a9, a4, 0
30
+ ee.movi.32.q q0, a9, 0
31
+ ee.vzip.8 q0, q5
32
+ .Ltm_ready:
33
+ mov a10, a6
34
+ .Ltm_row:
35
+ l8ui a9, a5, 0
36
+ addi a5, a5, 1
37
+ sub a9, a9, a8
38
+ s16i a9, a1, 0
39
+ ee.vldbc.16 q2, a1
40
+ ee.zero.q q5
41
+ ee.vadds.s16 q3, q0, q2
42
+ ee.vmax.s16 q3, q3, q5
43
+ ee.vmin.s16 q3, q3, q6
44
+ ee.vunzip.8 q3, q5
45
+ ee.movi.32.a q3, a9, 0
46
+ s32i a9, a2, 0
47
+ beqi a6, 4, .Ltm_next
48
+ ee.movi.32.a q3, a9, 1
49
+ s32i a9, a2, 4
50
+ beqi a6, 8, .Ltm_next
51
+ ee.zero.q q5
52
+ ee.vadds.s16 q3, q1, q2
53
+ ee.vmax.s16 q3, q3, q5
54
+ ee.vmin.s16 q3, q3, q6
55
+ ee.vunzip.8 q3, q5
56
+ ee.movi.32.a q3, a9, 0
57
+ s32i a9, a2, 8
58
+ ee.movi.32.a q3, a9, 1
59
+ s32i a9, a2, 12
60
+ .Ltm_next:
61
+ add a2, a2, a3
62
+ addi a10, a10, -1
63
+ bnez a10, .Ltm_row
64
+ retw
65
+ .size gea_vp8_tm_s3, .-gea_vp8_tm_s3