Remove excessive floating-point divides

Loft the loop-invariant divide outside the hot loop, and/or invert the variable to turn FDIV into FMUL.
tesseract-ocr · Sep 3, 2024 · 41faf69 · 41faf69
1 parent 027ad18
commit 41faf69
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 8 deletions.
diff --git a/src/lstm/networkio.cpp b/src/lstm/networkio.cpp
@@ -224,6 +224,7 @@ void NetworkIO::Copy2DImage(int batch, Image pix, float black, float contrast, T
   int target_width = stride_map_.Size(FD_WIDTH);
   int num_features = NumFeatures();
   bool color = num_features == 3;
+  float inv_contrast = 1.0/contrast;
   if (width > target_width) {
     width = target_width;
   }
@@ -236,11 +237,11 @@ void NetworkIO::Copy2DImage(int batch, Image pix, float black, float contrast, T
           int f = 0;
           for (int c = COLOR_RED; c <= COLOR_BLUE; ++c) {
             int pixel = GET_DATA_BYTE(line + x, c);
-            SetPixel(t, f++, pixel, black, contrast);
+            SetPixel(t, f++, pixel, black, inv_contrast);
           }
         } else {
           int pixel = GET_DATA_BYTE(line, x);
-          SetPixel(t, 0, pixel, black, contrast);
+          SetPixel(t, 0, pixel, black, inv_contrast);
         }
       }
     }
@@ -264,6 +265,7 @@ void NetworkIO::Copy1DGreyImage(int batch, Image pix, float black, float contras
   index.AddOffset(batch, FD_BATCH);
   int t = index.t();
   int target_width = stride_map_.Size(FD_WIDTH);
+  float inv_contrast = 1.0/contrast;
   if (width > target_width) {
     width = target_width;
   }
@@ -272,7 +274,7 @@ void NetworkIO::Copy1DGreyImage(int batch, Image pix, float black, float contras
     for (int y = 0; y < height; ++y) {
       uint32_t *line = pixGetData(pix) + wpl * y;
       int pixel = GET_DATA_BYTE(line, x);
-      SetPixel(t, y, pixel, black, contrast);
+      SetPixel(t, y, pixel, black, inv_contrast);
     }
   }
   for (; x < target_width; ++x) {
@@ -287,8 +289,9 @@ void NetworkIO::Copy1DGreyImage(int batch, Image pix, float black, float contras
 // pixel: the value of the pixel from the image (in one channel)
 // black: the pixel value to map to the lowest of the range of *this
 // contrast: the range of pixel values to stretch to half the range of *this.
-void NetworkIO::SetPixel(int t, int f, int pixel, float black, float contrast) {
-  float float_pixel = (pixel - black) / contrast - 1.0f;
+// inv_contrast: one over the contrast, to save a divide
+void NetworkIO::SetPixel(int t, int f, int pixel, float black, float inv_contrast) {
+  float float_pixel = (pixel - black) * inv_contrast - 1.0f;
   if (int_mode_) {
     i_[t][f] = ClipToRange<int>(IntCastRounded((INT8_MAX + 1) * float_pixel), -INT8_MAX, INT8_MAX);
   } else {

diff --git a/src/textord/pithsync.cpp b/src/textord/pithsync.cpp
@@ -120,6 +120,7 @@ void FPCUTPT::assign(       // constructor
                          // half of pitch
   int16_t half_pitch = pitch / 2 - 1;
   uint32_t lead_flag; // new flag
+  float inv_projection_scale = 1.0/projection_scale;
 
   if (half_pitch > 31) {
     half_pitch = 31;
@@ -166,7 +167,7 @@ void FPCUTPT::assign(       // constructor
             }
           }
           balance_count =
-              static_cast<int16_t>(balance_count * textord_balance_factor / projection_scale);
+              static_cast<int16_t>(balance_count * textord_balance_factor * inv_projection_scale);
         }
         r_index = segpt->region_index + 1;
         total = segpt->mean_sum + dist;
@@ -221,6 +222,7 @@ void FPCUTPT::assign_cheap( // constructor
                          // half of pitch
   int16_t half_pitch = pitch / 2 - 1;
   uint32_t lead_flag; // new flag
+  float inv_projection_scale = 1.0/projection_scale;
 
   if (half_pitch > 31) {
     half_pitch = 31;
@@ -260,7 +262,7 @@ void FPCUTPT::assign_cheap( // constructor
           lead_flag &= lead_flag - 1;
         }
         balance_count =
-            static_cast<int16_t>(balance_count * textord_balance_factor / projection_scale);
+            static_cast<int16_t>(balance_count * textord_balance_factor * projection_scale);
       }
       r_index = segpt->region_index + 1;
       total = segpt->mean_sum + dist;
@@ -511,6 +513,7 @@ double check_pitch_sync3(    // find segmentation
   int16_t best_fake;            // best fake level
   int16_t best_count;           // no of cuts
   FPSEGPT_IT seg_it = seg_list; // output iterator
+  float inv_projection_scale = 1.0/projection_scale;
 
   end = (end - start) % pitch;
   if (pitch < 3) {
@@ -597,7 +600,7 @@ double check_pitch_sync3(    // find segmentation
         offset = projection->pile_count(x);
         faking = true;
       } else {
-        projection_offset = static_cast<int16_t>(projection->pile_count(x) / projection_scale);
+        projection_offset = static_cast<int16_t>(projection->pile_count(x) * inv_projection_scale);
         if (projection_offset > offset) {
           offset = projection_offset;
         }