Mirror/AI-on-the-edge-device
1
0
Fork 0
mirror of https://github.com/jomjol/AI-on-the-edge-device.git synced 2025-12-09 13:06:54 +03:00
Code Issues Packages Projects Releases Wiki Activity

Rolling 20220716_2

Browse Source
This commit is contained in:
jomjol
2022-07-16 20:59:09 +02:00
parent 0b039e8d8c
commit 17fd0f96df
166 changed files with 10471 additions and 1691 deletions
Download Patch File Download Diff File Expand all files Collapse all files

63
code/components/esp-nn/src/common/common_functions.h
View File

@@ -41,15 +41,39 @@
__NN_FORCE_INLINE__ int32_t esp_nn_clz32(uint32_t in)
{
#if CONFIG_IDF_TARGET_ARCH_XTENSA
__asm__ volatile("nsau %0, %0" : "+r" (in));
return in;
}
__NN_FORCE_INLINE__ int32_t esp_nn_pick_sat_high32_of64(int64_t val64)
{
int32_t sign = (int32_t) (val64 >> 63);
int32_t to_add = sign & ((1ul << 31) - 1);
return (int32_t) ((int64_t) (val64 + to_add) >> 31);
#elif defined(__GNUC__)
return __builtin_clz(in);
#else
int32_t count = 32;
uint32_t x = in, y = in >> 16;
if (y != 0) {
count -= 16;
x = y;
}
y = x >> 8;
if (y != 0) {
count -= 8;
x = y;
}
y = x >> 4;
if (y != 0) {
count -= 4;
x = y;
}
y = x >> 2;
if (y != 0) {
count -= 2;
x = y;
}
y = x >> 1;
if (y != 0) {
return count - 2;
}
return count - x;
#endif
}
/**
@@ -57,8 +81,19 @@ __NN_FORCE_INLINE__ int32_t esp_nn_pick_sat_high32_of64(int64_t val64)
*/
__NN_FORCE_INLINE__ int32_t esp_nn_saturate8(int32_t in)
{
#if CONFIG_IDF_TARGET_ARCH_XTENSA
__asm__ volatile("clamps %0, %0, 7" : "+a"(in));
return in;
#else
return max(INT8_MIN, min(in, INT8_MAX));
#endif
}
__NN_FORCE_INLINE__ int32_t esp_nn_pick_sat_high32_of64(int64_t val64)
{
int32_t sign = (int32_t) (val64 >> 63);
int32_t to_add = sign & ((1ul << 31) - 1);
return (int32_t) ((int64_t) (val64 + to_add) >> 31);
}
__NN_FORCE_INLINE__ int32_t esp_nn_sat_round_doubling_high_mul(int32_t in0, int32_t in1)
@@ -144,7 +179,7 @@ static void esp_nn_aligned_s8_pad_with_value(const int8_t *src, int8_t *dst,
const uint16_t pad_ht)
{
/* memset with pad_val */
memset(dst, pad_val, ((input_wd + 2 * pad_wd) * (input_ht + 2 * pad_ht)) * channels * 2);
memset(dst, pad_val, ((input_wd + 2 * pad_wd) * (input_ht + 2 * pad_ht)) * channels);
dst += (pad_wd + input_wd + pad_wd) * channels;
for (int i = 0; i < input_ht; i++) {
@@ -156,7 +191,6 @@ static void esp_nn_aligned_s8_pad_with_value(const int8_t *src, int8_t *dst,
}
}
#if 0
static void esp_nn_aligned_s8_pad_end_with_value(const int8_t *src, int8_t *dst,
const uint16_t input_wd,
const uint16_t input_ht,
@@ -169,13 +203,16 @@ static void esp_nn_aligned_s8_pad_end_with_value(const int8_t *src, int8_t *dst,
for (int j = 0; j < input_wd * channels; j++) {
*dst++ = *src++;
}
memset(dst, pad_val, pad_wd * channels);
dst += pad_wd * channels;
if (pad_wd) {
memset(dst, pad_val, pad_wd * channels);
dst += pad_wd * channels;
}
}
/* pad end `pad_ht` lines at end */
memset(dst, pad_val, (input_wd + pad_wd) * pad_ht * channels);
if (pad_ht) {
memset(dst, pad_val, (input_wd + pad_wd) * pad_ht * channels);
}
}
#endif
/**
* @brief convert 8 bit input data to 16 bit