]> git.sesse.net Git - fjl/blobdiff - driver.c
Fix a Makefile typo.
[fjl] / driver.c
index 20d64f2cd0c9c062e775fb0c66aad4dd6be4b797..6dcc8b3da676161ecd60e05898a27ebe3d19bf3f 100644 (file)
--- a/driver.c
+++ b/driver.c
@@ -14,8 +14,13 @@ struct jpeg_image {
        unsigned width, height;
        unsigned num_components;
        unsigned hsample[256], vsample[256], qtable[256];
+       unsigned max_hsample, max_vsample;
+       unsigned stride[256];
+       unsigned num_blocks_horizontal, num_blocks_vertical;
        uint32_t qvalues[256][DCTSIZE2];
        void* idct_data[256];
+       uint8_t* pixel_data[256];
+       uint8_t* pixel_write_pointer[256];
 };
 
 ssize_t stdio_read(void* userdata, uint8_t* buf, size_t count) 
@@ -62,8 +67,8 @@ void read_sof(struct byte_source* source, struct jpeg_image* image)
        assert(len >= 8);
        image->precision = read_uint8(byte_source_input_func, source);
        assert(image->precision == 8);
-       image->width = read_uint16(byte_source_input_func, source);
        image->height = read_uint16(byte_source_input_func, source);
+       image->width = read_uint16(byte_source_input_func, source);
        image->num_components = read_uint8(byte_source_input_func, source);
        len -= 8;
 
@@ -79,9 +84,78 @@ void read_sof(struct byte_source* source, struct jpeg_image* image)
                image->qtable[c] = read_uint8(byte_source_input_func, source);
                len -= 3;
 
+               if (image->hsample[c] > image->max_hsample) {
+                       image->max_hsample = image->hsample[c];
+               }
+               if (image->vsample[c] > image->max_vsample) {
+                       image->max_vsample = image->vsample[c];
+               }
+
                fprintf(stderr, "Component %u: sampling factors %u x %x, quantization table %u\n",
                        c, image->hsample[c], image->vsample[c], image->qtable[c]);
        }
+       
+       image->num_blocks_horizontal = (image->width + image->max_hsample * DCTSIZE - 1) / (image->max_hsample * DCTSIZE);
+       image->num_blocks_vertical = (image->height + image->max_vsample * DCTSIZE - 1) / (image->max_vsample * DCTSIZE);
+
+       for (unsigned c = 0; c < 256; ++c) {
+               if (image->hsample[c] == 0) {
+                       continue;
+               }
+
+               unsigned width = image->num_blocks_horizontal * image->hsample[c] * DCTSIZE;
+               unsigned height = image->num_blocks_vertical * image->vsample[c] * DCTSIZE;
+               image->stride[c] = width;
+               image->pixel_data[c] = (uint8_t*)malloc(width * height);
+               assert(image->pixel_data[c] != NULL);
+               image->pixel_write_pointer[c] = image->pixel_data[c];
+
+               fprintf(stderr, "Component %u: allocating %d x %d\n", c, width, height);
+       }
+}
+
+void decode_ac_coefficients(const struct huffman_table* tbl, struct bit_source* bits, int16_t* coeff)
+{
+       for (unsigned i = 1; i < DCTSIZE2; ++i) {
+               possibly_refill(bits, DEHUF_AC_TABLE_BITS);
+               unsigned lookup = peek_bits(bits, DEHUF_AC_TABLE_BITS);
+               int code = tbl->ac_table_codes[lookup];
+
+               if (__builtin_expect(code == AC_DEHUF_SLOW_PATH, 0)) {
+                       unsigned rs = read_huffman_symbol_no_refill(tbl, bits);
+                       unsigned r = rs >> 4;
+                       unsigned s = rs & 0xf;
+                       i += r;
+                       possibly_refill(bits, s);
+
+                       if (rs == 0x00) {
+                               assert(code == AC_DEHUF_SLOW_PATH || code == AC_END_OF_BLOCK);
+                               /* end of block */
+                               break;
+                       }
+                       if (rs == 0xf0) {
+                               assert(code == AC_DEHUF_SLOW_PATH || code == AC_SIXTEEN_ZEROS);
+                               /* 16 zero coefficients */
+                               continue;
+                       }
+
+                       coeff[unzigzag[i]] = extend(read_bits(bits, s), s);
+               } else {
+                       unsigned length = tbl->ac_table_length[lookup];
+                       int r = tbl->ac_table_skip[lookup];
+                       assert(r >= 0);
+                       i += r;
+                       assert(bits->bits_available >= length);
+                       read_bits(bits, length);
+                       if (code == AC_END_OF_BLOCK) {
+                               break;
+                       }
+                       if (code == AC_SIXTEEN_ZEROS) {
+                               continue;
+                       }
+                       coeff[unzigzag[i]] = code;
+               }
+       }
 }
 
 void read_scan(struct byte_source* source, struct jpeg_image* image, huffman_tables_t* tables)
@@ -121,22 +195,19 @@ void read_scan(struct byte_source* source, struct jpeg_image* image, huffman_tab
        }
 
        struct bit_source bits;
-       init_bit_source(&bits, byte_source_input_func, source);
+       init_bit_source(&bits, byte_source_input_func, 8, source);
                
-       // Some debugging code.
-       assert(image->width % 8 == 0);
-       int y_stride = image->width;
-       uint8_t* y_row_data = (uint8_t*)malloc(y_stride * DCTSIZE * image->vsample[1]);
-       assert(y_row_data != NULL);
-       int xb = 0;
+       unsigned mcu_x = 0, mcu_y = 0;
 
-       for ( ;; ) {
+       while (!bits.source_eof) {
                for (unsigned c = 0; c < num_components; ++c) {
                        unsigned cn = component_num[c];
                        assert(image->idct_data[image->qtable[cn]] != NULL);
-                       
-                       for (unsigned local_yb = 0; local_yb < image->vsample[cn]; ++local_yb) {
-                               for (unsigned local_xb = 0; local_xb < image->hsample[cn]; ++local_xb) {
+
+                       uint8_t* pixel_write_pointer_y = image->pixel_write_pointer[cn];
+                       for (unsigned local_yb = 0; local_yb < image->vsample[cn]; ++local_yb, pixel_write_pointer_y += image->stride[cn] * DCTSIZE) {
+                               uint8_t* pixel_write_pointer = pixel_write_pointer_y;
+                               for (unsigned local_xb = 0; local_xb < image->hsample[cn]; ++local_xb, pixel_write_pointer += DCTSIZE) {
                                        const struct huffman_table* dc_table = &((*tables)[DC_CLASS][dc_huffman_table[c]]);
                                        const struct huffman_table* ac_table = &((*tables)[AC_CLASS][ac_huffman_table[c]]);
 
@@ -147,52 +218,36 @@ void read_scan(struct byte_source* source, struct jpeg_image* image, huffman_tab
                                        
                                        int16_t coeff[DCTSIZE2] = { 0 };
                                        coeff[0] = last_dc[c];
+                                       decode_ac_coefficients(ac_table, &bits, coeff);
 
-                                       // decode AC components
-                                       for (unsigned i = 1; i < DCTSIZE2; ++i) {
-                                               unsigned rs = read_huffman_symbol(ac_table, &bits);
-                                               unsigned r = rs >> 4;
-                                               unsigned s = rs & 0xf;
-
-                                               if (rs == 0x00) {
-                                                       /* end of block */
-                                                       break;
-                                               }
-                                               if (rs == 0xf0) {
-                                                       /* 16 zero coefficients */
-                                                       i += 15;
-                                                       continue;
-                                               }
-
-                                               possibly_refill(&bits, s);
-
-                                               i += r;
-                                               coeff[unzigzag[i]] = extend(read_bits(&bits, s), s);
-                                       }
-                       
                                        uint8_t pixdata[DCTSIZE2];      
                                        idct_choice(coeff, image->idct_data[image->qtable[cn]], pixdata);
 
-                                       if (cn != 1) {
-                                               continue;
-                                       }
-
-                                       for (int y = 0; y < DCTSIZE; ++y) {
-                                               memcpy(y_row_data + (local_yb * DCTSIZE + y) * y_stride + (xb + local_xb) * DCTSIZE,
-                                                      pixdata + y * DCTSIZE,
-                                                      DCTSIZE);
+                                       uint8_t* dest_pixdata = pixel_write_pointer;
+                                       for (unsigned y = 0; y < DCTSIZE; ++y, dest_pixdata += image->stride[cn]) {
+                                               memcpy(dest_pixdata, pixdata + y * DCTSIZE, DCTSIZE);
                                        }
                                }
                        }
-
-                       if (cn != 1) {
-                               continue;
+                       image->pixel_write_pointer[cn] += DCTSIZE * image->hsample[cn];
+               }
+       
+               if (++mcu_x == image->num_blocks_horizontal) {
+                       ++mcu_y;
+                       mcu_x = 0;
+               
+                       for (unsigned c = 0; c < num_components; ++c) {
+                               unsigned cn = component_num[c];
+                               image->pixel_write_pointer[cn] += (image->vsample[cn] * DCTSIZE - 1) * image->stride[cn];
                        }
 
-                       xb += image->hsample[cn];
-                       if (xb * DCTSIZE == y_stride) {
-                               fwrite(y_row_data, y_stride * DCTSIZE * image->vsample[cn], 1, stdout);
-                               xb = 0;
+                       // Some debug code.
+                       const int c = 1;
+                       if (mcu_y == image->num_blocks_vertical) {
+                               unsigned stride = image->num_blocks_horizontal * image->hsample[c] * DCTSIZE;
+                               unsigned height = image->num_blocks_vertical * image->vsample[c] * DCTSIZE;
+                               printf("P5\n%u %u\n255\n", stride, height);
+                               fwrite(image->pixel_data[c], stride * height, 1, stdout);
                        }
                }
        }
@@ -270,6 +325,9 @@ int main(void)
                case 0xd8:
                        /* SOI */
                        break;
+               case 0xd9:
+                       /* EOI */
+                       exit(0);
                case 0xc4:
                        /* DHT (define Huffman tables) */
                        read_huffman_tables(&tables, byte_source_input_func, &source);