mirror of
https://github.com/openmv/openmv.git
synced 2025-11-04 14:49:50 +08:00
Merge pull request #2228 from kwagyeman/kwabena/fix_missing_invalidates
imlib/draw: Add missing speculative read invalidates.
This commit is contained in:
commit
f922b76ae3
@ -801,7 +801,7 @@ void imlib_draw_row_setup(imlib_draw_row_data_t *data) {
|
|||||||
cfg.pCLUT = clut;
|
cfg.pCLUT = clut;
|
||||||
cfg.CLUTColorMode = DMA2D_CCM_ARGB8888;
|
cfg.CLUTColorMode = DMA2D_CCM_ARGB8888;
|
||||||
cfg.Size = 255;
|
cfg.Size = 255;
|
||||||
#if defined(MCU_SERIES_F7) || defined(MCU_SERIES_H7)
|
#if __DCACHE_PRESENT
|
||||||
SCB_CleanDCache_by_Addr(clut, 256 * sizeof(uint32_t));
|
SCB_CleanDCache_by_Addr(clut, 256 * sizeof(uint32_t));
|
||||||
#endif
|
#endif
|
||||||
HAL_DMA2D_CLUTLoad(&data->dma2d, cfg, 1);
|
HAL_DMA2D_CLUTLoad(&data->dma2d, cfg, 1);
|
||||||
@ -828,6 +828,10 @@ void imlib_draw_row_setup(imlib_draw_row_data_t *data) {
|
|||||||
data->dma2d.LayerCfg[1].ChromaSubSampling = DMA2D_NO_CSS;
|
data->dma2d.LayerCfg[1].ChromaSubSampling = DMA2D_NO_CSS;
|
||||||
#endif
|
#endif
|
||||||
HAL_DMA2D_ConfigLayer(&data->dma2d, 1);
|
HAL_DMA2D_ConfigLayer(&data->dma2d, 1);
|
||||||
|
#if __DCACHE_PRESENT
|
||||||
|
// SCB_InvalidateDCache_by_Addr does nothing if dsize is 0.
|
||||||
|
data->dma2d_invalidate_dsize = 0;
|
||||||
|
#endif
|
||||||
} else {
|
} else {
|
||||||
data->row_buffer[1] = data->row_buffer[0];
|
data->row_buffer[1] = data->row_buffer[0];
|
||||||
}
|
}
|
||||||
@ -866,6 +870,10 @@ void imlib_draw_row_teardown(imlib_draw_row_data_t *data) {
|
|||||||
if (data->dma2d_initialized) {
|
if (data->dma2d_initialized) {
|
||||||
if (!data->callback) {
|
if (!data->callback) {
|
||||||
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
||||||
|
#if __DCACHE_PRESENT
|
||||||
|
// Ensures any cached reads to dst16 are dropped.
|
||||||
|
SCB_InvalidateDCache_by_Addr(data->dma2d_invalidate_addr, data->dma2d_invalidate_dsize);
|
||||||
|
#endif
|
||||||
}
|
}
|
||||||
HAL_DMA2D_DeInit(&data->dma2d);
|
HAL_DMA2D_DeInit(&data->dma2d);
|
||||||
if (data->src_img_pixfmt == PIXFORMAT_GRAYSCALE) {
|
if (data->src_img_pixfmt == PIXFORMAT_GRAYSCALE) {
|
||||||
@ -2379,15 +2387,24 @@ void imlib_draw_row(int x_start, int x_end, int y_row, imlib_draw_row_data_t *da
|
|||||||
if (data->dma2d_enabled) {
|
if (data->dma2d_enabled) {
|
||||||
if (!data->callback) {
|
if (!data->callback) {
|
||||||
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
||||||
|
#if __DCACHE_PRESENT
|
||||||
|
// Ensures any cached reads to dst16 are dropped.
|
||||||
|
SCB_InvalidateDCache_by_Addr(data->dma2d_invalidate_addr,
|
||||||
|
data->dma2d_invalidate_dsize);
|
||||||
|
#endif
|
||||||
}
|
}
|
||||||
#if defined(MCU_SERIES_F7) || defined(MCU_SERIES_H7)
|
#if __DCACHE_PRESENT
|
||||||
// Memory referenced by src8 between (x_end - x_start) may or may not be
|
// Memory referenced by src8 between (x_end - x_start) may or may not be
|
||||||
// cache algined. However, after being flushed it shouldn't change again
|
// cache algined. However, after being flushed it shouldn't change again
|
||||||
// so DMA2D can safety read the line of pixels.
|
// so DMA2D can safety read the line of pixels.
|
||||||
SCB_CleanDCache_by_Addr((uint32_t *) src8, (x_end - x_start) * sizeof(uint8_t));
|
SCB_CleanDCache_by_Addr((uint32_t *) src8, (x_end - x_start) * sizeof(uint8_t));
|
||||||
// DMA2D will overwrite this area. dst16 (x_end - x_start) must be cache
|
// DMA2D will overwrite this area. dst16 (x_end - x_start) must be cache
|
||||||
// aligned or the line of pixels will be corrutped.
|
// aligned or the line of pixels will be corrutped.
|
||||||
SCB_InvalidateDCache_by_Addr((uint32_t *) dst16, (x_end - x_start) * sizeof(uint16_t));
|
// Ensures any cached writes to dst16 are dropped.
|
||||||
|
data->dma2d_invalidate_addr = (uint32_t *) dst16;
|
||||||
|
data->dma2d_invalidate_dsize = (x_end - x_start) * sizeof(uint16_t);
|
||||||
|
SCB_InvalidateDCache_by_Addr(data->dma2d_invalidate_addr,
|
||||||
|
data->dma2d_invalidate_dsize);
|
||||||
#endif
|
#endif
|
||||||
HAL_DMA2D_BlendingStart(&data->dma2d,
|
HAL_DMA2D_BlendingStart(&data->dma2d,
|
||||||
(uint32_t) src8,
|
(uint32_t) src8,
|
||||||
@ -2397,6 +2414,11 @@ void imlib_draw_row(int x_start, int x_end, int y_row, imlib_draw_row_data_t *da
|
|||||||
1);
|
1);
|
||||||
if (data->callback) {
|
if (data->callback) {
|
||||||
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
||||||
|
#if __DCACHE_PRESENT
|
||||||
|
// Ensures any cached reads to dst16 are dropped.
|
||||||
|
SCB_InvalidateDCache_by_Addr(data->dma2d_invalidate_addr,
|
||||||
|
data->dma2d_invalidate_dsize);
|
||||||
|
#endif
|
||||||
}
|
}
|
||||||
} else if (data->smuad_alpha_palette) {
|
} else if (data->smuad_alpha_palette) {
|
||||||
#else
|
#else
|
||||||
@ -2550,15 +2572,24 @@ void imlib_draw_row(int x_start, int x_end, int y_row, imlib_draw_row_data_t *da
|
|||||||
if (data->dma2d_enabled) {
|
if (data->dma2d_enabled) {
|
||||||
if (!data->callback) {
|
if (!data->callback) {
|
||||||
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
||||||
|
#if __DCACHE_PRESENT
|
||||||
|
// Ensures any cached reads to dst16 are dropped.
|
||||||
|
SCB_InvalidateDCache_by_Addr(data->dma2d_invalidate_addr,
|
||||||
|
data->dma2d_invalidate_dsize);
|
||||||
|
#endif
|
||||||
}
|
}
|
||||||
#if defined(MCU_SERIES_F7) || defined(MCU_SERIES_H7)
|
#if __DCACHE_PRESENT
|
||||||
// Memory referenced by src16 between (x_end - x_start) may or may not be
|
// Memory referenced by src16 between (x_end - x_start) may or may not be
|
||||||
// cache algined. However, after being flushed it shouldn't change again
|
// cache algined. However, after being flushed it shouldn't change again
|
||||||
// so DMA2D can safety read the line of pixels.
|
// so DMA2D can safety read the line of pixels.
|
||||||
SCB_CleanDCache_by_Addr((uint32_t *) src16, (x_end - x_start) * sizeof(uint16_t));
|
SCB_CleanDCache_by_Addr((uint32_t *) src16, (x_end - x_start) * sizeof(uint16_t));
|
||||||
// DMA2D will overwrite this area. dst16 (x_end - x_start) must be cache
|
// DMA2D will overwrite this area. dst16 (x_end - x_start) must be cache
|
||||||
// aligned or the line of pixels will be corrutped.
|
// aligned or the line of pixels will be corrutped.
|
||||||
SCB_InvalidateDCache_by_Addr((uint32_t *) dst16, (x_end - x_start) * sizeof(uint16_t));
|
// Ensures any cached writes to dst16 are dropped.
|
||||||
|
data->dma2d_invalidate_addr = (uint32_t *) dst16;
|
||||||
|
data->dma2d_invalidate_dsize = (x_end - x_start) * sizeof(uint16_t);
|
||||||
|
SCB_InvalidateDCache_by_Addr(data->dma2d_invalidate_addr,
|
||||||
|
data->dma2d_invalidate_dsize);
|
||||||
#endif
|
#endif
|
||||||
HAL_DMA2D_BlendingStart(&data->dma2d,
|
HAL_DMA2D_BlendingStart(&data->dma2d,
|
||||||
(uint32_t) src16,
|
(uint32_t) src16,
|
||||||
@ -2568,6 +2599,11 @@ void imlib_draw_row(int x_start, int x_end, int y_row, imlib_draw_row_data_t *da
|
|||||||
1);
|
1);
|
||||||
if (data->callback) {
|
if (data->callback) {
|
||||||
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
HAL_DMA2D_PollForTransfer(&data->dma2d, 1000);
|
||||||
|
#if __DCACHE_PRESENT
|
||||||
|
// Ensures any cached reads to dst16 are dropped.
|
||||||
|
SCB_InvalidateDCache_by_Addr(data->dma2d_invalidate_addr,
|
||||||
|
data->dma2d_invalidate_dsize);
|
||||||
|
#endif
|
||||||
}
|
}
|
||||||
} else if (!data->black_background) {
|
} else if (!data->black_background) {
|
||||||
#else
|
#else
|
||||||
|
|||||||
@ -1143,6 +1143,10 @@ typedef struct imlib_draw_row_data {
|
|||||||
bool dma2d_enabled; // private
|
bool dma2d_enabled; // private
|
||||||
bool dma2d_initialized; // private
|
bool dma2d_initialized; // private
|
||||||
DMA2D_HandleTypeDef dma2d; // private
|
DMA2D_HandleTypeDef dma2d; // private
|
||||||
|
#if __DCACHE_PRESENT
|
||||||
|
void *dma2d_invalidate_addr; // private
|
||||||
|
int32_t dma2d_invalidate_dsize; // private
|
||||||
|
#endif
|
||||||
#endif
|
#endif
|
||||||
long smuad_alpha; // private
|
long smuad_alpha; // private
|
||||||
uint32_t *smuad_alpha_palette; // private
|
uint32_t *smuad_alpha_palette; // private
|
||||||
|
|||||||
@ -729,7 +729,7 @@ static void spi_tv_display(image_t *src_img, int dst_x_start, int dst_y_start, f
|
|||||||
|
|
||||||
#ifdef __DCACHE_PRESENT
|
#ifdef __DCACHE_PRESENT
|
||||||
// Flush data for DMA
|
// Flush data for DMA
|
||||||
SCB_CleanDCache();
|
SCB_CleanDCache_by_Addr((uint32_t *) dst_img.data, image_size(&dst_img));
|
||||||
#endif
|
#endif
|
||||||
|
|
||||||
// Update head which means a new image is ready.
|
// Update head which means a new image is ready.
|
||||||
|
|||||||
@ -81,6 +81,11 @@ static void jpeg_compress_get_data(JPEG_HandleTypeDef *hjpeg, uint32_t NbDecoded
|
|||||||
}
|
}
|
||||||
|
|
||||||
static void jpeg_compress_data_ready(JPEG_HandleTypeDef *hjpeg, uint8_t *pDataOut, uint32_t OutDataLength) {
|
static void jpeg_compress_data_ready(JPEG_HandleTypeDef *hjpeg, uint8_t *pDataOut, uint32_t OutDataLength) {
|
||||||
|
if ((!(((uint32_t) pDataOut) % __SCB_DCACHE_LINE_SIZE)) && (OutDataLength == JPEG_OUTPUT_CHUNK_SIZE)) {
|
||||||
|
// Ensure any cached reads are dropped.
|
||||||
|
SCB_InvalidateDCache_by_Addr((uint32_t *) pDataOut, JPEG_OUTPUT_CHUNK_SIZE);
|
||||||
|
}
|
||||||
|
|
||||||
// We have received this much data.
|
// We have received this much data.
|
||||||
JPEG_state.out_data_len += OutDataLength;
|
JPEG_state.out_data_len += OutDataLength;
|
||||||
|
|
||||||
@ -96,7 +101,7 @@ static void jpeg_compress_data_ready(JPEG_HandleTypeDef *hjpeg, uint8_t *pDataOu
|
|||||||
// image in randomly aligned chunks. We only want to invalidate the cache of the output
|
// image in randomly aligned chunks. We only want to invalidate the cache of the output
|
||||||
// buffer for the initial DMA chunks. So, this code below will do that and then only
|
// buffer for the initial DMA chunks. So, this code below will do that and then only
|
||||||
// invalidate aligned regions when the processor is moving the final parts of the image.
|
// invalidate aligned regions when the processor is moving the final parts of the image.
|
||||||
if (!(((uint32_t) new_pDataOut) % __SCB_DCACHE_LINE_SIZE)) {
|
if ((!(((uint32_t) new_pDataOut) % __SCB_DCACHE_LINE_SIZE)) && (OutDataLength == JPEG_OUTPUT_CHUNK_SIZE)) {
|
||||||
SCB_InvalidateDCache_by_Addr((uint32_t *) new_pDataOut, JPEG_OUTPUT_CHUNK_SIZE);
|
SCB_InvalidateDCache_by_Addr((uint32_t *) new_pDataOut, JPEG_OUTPUT_CHUNK_SIZE);
|
||||||
}
|
}
|
||||||
|
|
||||||
@ -506,7 +511,7 @@ void jpeg_decompress(image_t *dst, image_t *src) {
|
|||||||
HAL_DMA2D_Init(&DMA2D_Handle);
|
HAL_DMA2D_Init(&DMA2D_Handle);
|
||||||
HAL_DMA2D_ConfigLayer(&DMA2D_Handle, 1);
|
HAL_DMA2D_ConfigLayer(&DMA2D_Handle, 1);
|
||||||
|
|
||||||
// Invalidate the dst image for DMA2D.
|
// Ensure any cached writes are dropped.
|
||||||
SCB_InvalidateDCache_by_Addr((uint32_t *) dst->data, image_size(dst));
|
SCB_InvalidateDCache_by_Addr((uint32_t *) dst->data, image_size(dst));
|
||||||
}
|
}
|
||||||
} else if (JPEG_state.jpeg_descr.Conf.ColorSpace == JPEG_CMYK_COLORSPACE) {
|
} else if (JPEG_state.jpeg_descr.Conf.ColorSpace == JPEG_CMYK_COLORSPACE) {
|
||||||
@ -565,6 +570,9 @@ void jpeg_decompress(image_t *dst, image_t *src) {
|
|||||||
HAL_JPEG_Resume(&JPEG_state.jpeg_descr, JPEG_PAUSE_RESUME_OUTPUT);
|
HAL_JPEG_Resume(&JPEG_state.jpeg_descr, JPEG_PAUSE_RESUME_OUTPUT);
|
||||||
}
|
}
|
||||||
|
|
||||||
|
// Ensure any cached reads are dropped.
|
||||||
|
SCB_InvalidateDCache_by_Addr((uint32_t *) this_mcu_row_buffer_ptr, dst_w_mcus_bytes);
|
||||||
|
|
||||||
if (JPEG_state.jpeg_descr.Conf.ColorSpace == JPEG_GRAYSCALE_COLORSPACE) {
|
if (JPEG_state.jpeg_descr.Conf.ColorSpace == JPEG_GRAYSCALE_COLORSPACE) {
|
||||||
for (int x_offset = 0; x_offset < src->w; x_offset += JPEG_MCU_W) {
|
for (int x_offset = 0; x_offset < src->w; x_offset += JPEG_MCU_W) {
|
||||||
uint8_t *Y0 = this_mcu_row_buffer_ptr + (x_offset * JPEG_MCU_H);
|
uint8_t *Y0 = this_mcu_row_buffer_ptr + (x_offset * JPEG_MCU_H);
|
||||||
@ -709,9 +717,21 @@ void jpeg_decompress(image_t *dst, image_t *src) {
|
|||||||
}
|
}
|
||||||
case PIXFORMAT_RGB565: {
|
case PIXFORMAT_RGB565: {
|
||||||
uint16_t *rp = IMAGE_COMPUTE_RGB565_PIXEL_ROW_PTR(dst, y_offset);
|
uint16_t *rp = IMAGE_COMPUTE_RGB565_PIXEL_ROW_PTR(dst, y_offset);
|
||||||
SCB_InvalidateDCache_by_Addr((uint32_t *) rp, dst->w * dy * sizeof(uint16_t));
|
|
||||||
HAL_DMA2D_Start(&DMA2D_Handle, (uint32_t) this_mcu_row_buffer_ptr, (uint32_t) rp, dst->w, dy);
|
HAL_DMA2D_Start(&DMA2D_Handle, (uint32_t) this_mcu_row_buffer_ptr, (uint32_t) rp, dst->w, dy);
|
||||||
|
|
||||||
|
// Invalidate any cached reads for the previous line that was just written.
|
||||||
|
if ((y_offset - mcu_h) >= 0) {
|
||||||
|
uint16_t *previous_rp = IMAGE_COMPUTE_RGB565_PIXEL_ROW_PTR(dst, (y_offset - mcu_h));
|
||||||
|
SCB_InvalidateDCache_by_Addr((uint32_t *) previous_rp, dst->w * mcu_h * sizeof(uint16_t));
|
||||||
|
}
|
||||||
|
|
||||||
HAL_DMA2D_PollForTransfer(&DMA2D_Handle, JPEG_CODEC_TIMEOUT);
|
HAL_DMA2D_PollForTransfer(&DMA2D_Handle, JPEG_CODEC_TIMEOUT);
|
||||||
|
|
||||||
|
// For the last row invalidate any cached reads for the line that was just written.
|
||||||
|
if ((y_offset + mcu_h) >= src->h) {
|
||||||
|
SCB_InvalidateDCache_by_Addr((uint32_t *) rp, dst->w * mcu_h * sizeof(uint16_t));
|
||||||
|
}
|
||||||
|
|
||||||
break;
|
break;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
|||||||
Loading…
Reference in New Issue
Block a user