From 0b682c54171d2923b6f3a6c0f4cae05b707ee711 Mon Sep 17 00:00:00 2001
From: Anurag Thakur <anurag105csec21@bpitindia.edu.in>
Date: Fri, 16 Sep 2022 00:45:17 +0530
Subject: [PATCH] Add SIMD

---
 .vscode/settings.json |  5 +++--
 src/dense/ftdense.c   | 47 ++++++++++++++++++++++++++++++-------------
 src/dense/rules.mk    |  5 +++--
 3 files changed, 39 insertions(+), 18 deletions(-)

diff --git a/.vscode/settings.json b/.vscode/settings.json
index 1a2f8af47..7f56b4bc5 100644
--- a/.vscode/settings.json
+++ b/.vscode/settings.json
@@ -2,6 +2,7 @@
     "files.associations": {
         "ftoutln.h": "c",
         "svprop.h": "c",
-        "ftdebug.h": "c"
+        "ftdebug.h": "c",
+        "tmmintrin.h": "c"
     }
-}
\ No newline at end of file
+}
diff --git a/src/dense/ftdense.c b/src/dense/ftdense.c
index dfd0f7417..4b0a20b28 100644
--- a/src/dense/ftdense.c
+++ b/src/dense/ftdense.c
@@ -11,6 +11,7 @@
 #include "ftdense.h"
 
 #include <math.h>
+#include <tmmintrin.h>
 #include "ftdenseerrs.h"
 
 #define PIXEL_BITS 8
@@ -372,22 +373,40 @@ dense_render_glyph( dense_worker* worker, const FT_Bitmap* target )
 
   unsigned char* dest     = target->buffer;
   unsigned char* dest_end = target->buffer + worker->m_w * worker->m_h;
-  float          value    = 0.0f;
-  while ( dest < dest_end )
-  {
-    value += *source++;
-    if ( value > 0.0f )
-    {
-      int n = (int)( fabs( value ) * 255.0f + 0.5f );
-      if ( n > 255 )
-        n = 255;
-      *dest = (unsigned char)n;
-    }
-    else
-      *dest = 0;
-    dest++;
+
+  __m128 offset = _mm_setzero_ps();
+  __m128i mask = _mm_set1_epi32(0x0c080400);
+  __m128 sign_mask = _mm_set1_ps(-0.f);
+  for (int i = 0; i < worker->m_h*worker->m_w; i += 4) {
+    __m128 x = _mm_load_ps(&source[i]);
+    x = _mm_add_ps(x, _mm_castsi128_ps(_mm_slli_si128(_mm_castps_si128(x), 4)));
+    x = _mm_add_ps(x, _mm_shuffle_ps(_mm_setzero_ps(), x, 0x40));
+    x = _mm_add_ps(x, offset);
+    __m128 y = _mm_andnot_ps(sign_mask, x);  // fabs(x)
+    y = _mm_min_ps(y, _mm_set1_ps(1.0f));
+    y = _mm_mul_ps(y, _mm_set1_ps(255.0f));
+    __m128i z = _mm_cvtps_epi32(y);
+    z = _mm_shuffle_epi8(z, mask);
+    _mm_store_ss((float *)&dest[i], (__m128)z);
+    offset = _mm_shuffle_ps(x, x, _MM_SHUFFLE(3, 3, 3, 3));
   }
 
+  // float          value    = 0.0f;
+  // while ( dest < dest_end )
+  // {
+  //   value += *source++;
+  //   if ( value > 0.0f )
+  //   {
+  //     int n = (int)( fabs( value ) * 255.0f + 0.5f );
+  //     if ( n > 255 )
+  //       n = 255;
+  //     *dest = (unsigned char)n;
+  //   }
+  //   else
+  //     *dest = 0;
+  //   dest++;
+  // }
+
   free(worker->m_a);
   return error;
 }
diff --git a/src/dense/rules.mk b/src/dense/rules.mk
index 005116873..38874f28e 100644
--- a/src/dense/rules.mk
+++ b/src/dense/rules.mk
@@ -22,8 +22,9 @@ DENSE_DIR := $(SRC_DIR)/dense
 #
 DENSE_COMPILE := $(CC) $(ANSIFLAGS)                               \
                         $I$(subst /,$(COMPILER_SEP),$(DENSE_DIR)) \
-                        $(INCLUDE_FLAGS)                           \
-                        $(FT_CFLAGS)
+                        $(INCLUDE_FLAGS)                          \
+                        $(FT_CFLAGS)                              \
+                        "-msse4.1"
 
 
 # DENSE driver sources (i.e., C files)