Remove some debugging.

[qscale] / qscale.c

diff --git a/qscale.c b/qscale.c
index 8ceb53eb0ea8565d31cf72df64c8814154c9b991..1fcfd67716d645391b9123bb631cdfae8702b954 100644 (file)
--- a/qscale.c
+++ b/qscale.c
@@ -86,7 +86,7 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw
                                size_coeffs <<= 1;
                                coeffs = (float *)realloc(coeffs, size_coeffs * sizeof(float));
                        }
-               
+
                        coeffs[num_coeffs++] = f;
                        sum += f;
                }
@@ -95,36 +95,41 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw
                        coeffs[pd[x].startcoeff + sx - start] /= sum;
                }
        }
-       
+
        for (y = 0; y < h; ++y) {
                float *sptr = pix + y*sstride;
                unsigned char *dptr = npix + y*dstride;
                unsigned char ch;
                for (x = 0; x < nw; ++x) {
+                       int result;
                        float acc;
+                       int tmp;
                        static const float low = 0.0, high = 255.0;
                        asm (
-                               "pxor %0, %0               \n"
-                               "xor %%eax, %%eax          \n"
+                               "pxor %1, %1               \n"
+                               "xor %2, %2                \n"
                                ".lbl2:                    \n"
-                               "movups (%2,%%eax),%%xmm1  \n"
-                               "movups (%1,%%eax),%%xmm2  \n"
+                               "movups (%4,%2),%%xmm1     \n"
+                               "movups (%3,%2),%%xmm2     \n"
                                "mulps %%xmm2,%%xmm1       \n"
-                               "addps %%xmm1,%0           \n"
-                               "addl $16,%%eax            \n"
-                               "dec %3                    \n"
+                               "addps %%xmm1,%1           \n"
+                               "addl $16,%2               \n"
+                               "dec %5                    \n"
                                "jnz .lbl2                 \n"
-                               "haddps %0,%0              \n"
-                               "haddps %0,%0              \n"
-                               "maxss %4,%0               \n"
-                               "minss %5,%0               \n"
-                               : "=x" (acc)
+                               "haddps %1,%1              \n"
+                               "haddps %1,%1              \n"
+                               "maxss %6,%1               \n"
+                               "minss %7,%1               \n"
+                               "cvtss2si %1,%0            \n"
+                               : "=r" (result),
+                                 "=&x" (acc),
+                                 "=&r" (tmp)
                                : "r" (&coeffs[pd[x].startcoeff]),
                                  "r" (&sptr[pd[x].start]),
                                  "r" ((pd[x].end - pd[x].start + 1)/4),
                                  "m" (low),
                                  "m" (high)
-                               : "eax", "xmm1", "xmm2"
+                               : "xmm1", "xmm2"
                        );
 
 #if 0
@@ -136,7 +141,7 @@ void hscale(float *pix, unsigned char *npix, unsigned w, unsigned h, unsigned nw
                                ch = (unsigned char)acc;
                        *dptr++ = ch;
 #endif
-                       *dptr++ = (unsigned char)acc;
+                       *dptr++ = (unsigned char)result;
                }
                ch = dptr[-1];
                for ( ; x < dstride; ++x) {
@@ -352,9 +357,14 @@ int main(int argc, char **argv)
        dinfo.raw_data_out = TRUE;
        jpeg_start_decompress(&dinfo);
 
-       unsigned w0 = dinfo.image_width * samp_h0 / max_samp_h, h0 = dinfo.image_height * samp_v0 / max_samp_v;
-       unsigned w1 = dinfo.image_width * samp_h1 / max_samp_h, h1 = dinfo.image_height * samp_v1 / max_samp_v;
-       unsigned w2 = dinfo.image_width * samp_h2 / max_samp_h, h2 = dinfo.image_height * samp_v2 / max_samp_v;
+       unsigned w0 = dinfo.image_width * dinfo.comp_info[0].h_samp_factor / dinfo.max_h_samp_factor;
+       unsigned h0 = dinfo.image_height * dinfo.comp_info[0].v_samp_factor / dinfo.max_v_samp_factor;
+
+       unsigned w1 = dinfo.image_width * dinfo.comp_info[1].h_samp_factor / dinfo.max_h_samp_factor;
+       unsigned h1 = dinfo.image_height * dinfo.comp_info[1].v_samp_factor / dinfo.max_v_samp_factor;
+
+       unsigned w2 = dinfo.image_width * dinfo.comp_info[2].h_samp_factor / dinfo.max_h_samp_factor;
+       unsigned h2 = dinfo.image_height * dinfo.comp_info[2].v_samp_factor / dinfo.max_v_samp_factor;
 
        fprintf(stderr, "Scaling using Lanczos filter:\n");
        fprintf(stderr, "  Y component: %ux%u -> %ux%u\n", dinfo.comp_info[0].width_in_blocks * DCTSIZE, dinfo.comp_info[0].height_in_blocks * DCTSIZE, nw0, nh0);