From 44a0ee3fa711be899da9f55ce857627d7d6e8dda Mon Sep 17 00:00:00 2001
From: Pascal Massimino <pascal.massimino@gmail.com>
Date: Thu, 16 Nov 2017 06:59:52 -0800
Subject: [PATCH] introduce WebPHasAlpha8b and WebPHasAlpha32b

Rewrote WebPPictureHasTransparency() to use them (even for argb).
This is 10% faster, for some reasons.

SSE2 version should be straightforward.
Removes a TODO.

Change-Id: I7ad5848fc5e355e2df505dbcd5a0f42fb6cbab41
---
 src/dsp/alpha_processing.c | 21 +++++++++++++++++++++
 src/dsp/dsp.h              |  5 +++++
 src/enc/picture_csp_enc.c  | 27 ++++++++++++---------------
 3 files changed, 38 insertions(+), 15 deletions(-)

diff --git a/src/dsp/alpha_processing.c b/src/dsp/alpha_processing.c
index dfead26d..590e3bc3 100644
--- a/src/dsp/alpha_processing.c
+++ b/src/dsp/alpha_processing.c
@@ -346,6 +346,19 @@ static void ExtractGreen_C(const uint32_t* argb, uint8_t* alpha, int size) {
 }
 #endif  // !WEBP_NEON_OMIT_C_CODE
 
+//------------------------------------------------------------------------------
+
+static int HasAlpha8b_C(const uint8_t* src, int length) {
+  while (length-- > 0) if (*src++ != 0xff) return 1;
+  return 0;
+}
+
+static int HasAlpha32b_C(const uint8_t* src, int length) {
+  int x;
+  for (x = 0; length-- > 0; x += 4) if (src[x] != 0xff) return 1;
+  return 0;
+}
+
 //------------------------------------------------------------------------------
 // Simple channel manipulations.
 
@@ -371,6 +384,9 @@ void (*WebPExtractGreen)(const uint32_t* argb, uint8_t* alpha, int size);
 void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
                     int len, int step, uint32_t* out);
 
+int (*WebPHasAlpha8b)(const uint8_t* src, int length);
+int (*WebPHasAlpha32b)(const uint8_t* src, int length);
+
 //------------------------------------------------------------------------------
 // Init function
 
@@ -398,6 +414,9 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessing(void) {
   WebPExtractGreen = ExtractGreen_C;
 #endif
 
+  WebPHasAlpha8b = HasAlpha8b_C;
+  WebPHasAlpha32b = HasAlpha32b_C;
+
   // If defined, use CPUInfo() to overwrite some pointers with faster versions.
   if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_SSE2)
@@ -433,6 +452,8 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitAlphaProcessing(void) {
   assert(WebPExtractAlpha != NULL);
   assert(WebPExtractGreen != NULL);
   assert(WebPPackRGB != NULL);
+  assert(WebPHasAlpha8b != NULL);
+  assert(WebPHasAlpha32b != NULL);
 
   alpha_processing_last_cpuinfo_used = VP8GetCPUInfo;
 }
diff --git a/src/dsp/dsp.h b/src/dsp/dsp.h
index d2001121..7a87595c 100644
--- a/src/dsp/dsp.h
+++ b/src/dsp/dsp.h
@@ -578,6 +578,11 @@ void WebPMultARGBRow_C(uint32_t* const ptr, int width, int inverse);
 extern void (*WebPPackRGB)(const uint8_t* r, const uint8_t* g, const uint8_t* b,
                            int len, int step, uint32_t* out);
 
+// This function returns true if src[i] contains a value different from 0xff.
+extern int (*WebPHasAlpha8b)(const uint8_t* src, int length);
+// This function returns true if src[4*i] contains a value different from 0xff.
+extern int (*WebPHasAlpha32b)(const uint8_t* src, int length);
+
 // To be called first before using the above.
 void WebPInitAlphaProcessing(void);
 
diff --git a/src/enc/picture_csp_enc.c b/src/enc/picture_csp_enc.c
index 62be097b..cf5170b3 100644
--- a/src/enc/picture_csp_enc.c
+++ b/src/enc/picture_csp_enc.c
@@ -41,12 +41,15 @@ static const union {
 static int CheckNonOpaque(const uint8_t* alpha, int width, int height,
                           int x_step, int y_step) {
   if (alpha == NULL) return 0;
-  while (height-- > 0) {
-    int x;
-    for (x = 0; x < width * x_step; x += x_step) {
-      if (alpha[x] != 0xff) return 1;  // TODO(skal): check 4/8 bytes at a time.
+  WebPInitAlphaProcessing();
+  if (x_step == 1) {
+    for (; height-- > 0; alpha += y_step) {
+      if (WebPHasAlpha8b(alpha, width)) return 1;
+    }
+  } else {
+    for (; height-- > 0; alpha += y_step) {
+      if (WebPHasAlpha32b(alpha, width)) return 1;
     }
-    alpha += y_step;
   }
   return 0;
 }
@@ -58,15 +61,10 @@ int WebPPictureHasTransparency(const WebPPicture* picture) {
     return CheckNonOpaque(picture->a, picture->width, picture->height,
                           1, picture->a_stride);
   } else {
-    int x, y;
-    const uint32_t* argb = picture->argb;
-    if (argb == NULL) return 0;
-    for (y = 0; y < picture->height; ++y) {
-      for (x = 0; x < picture->width; ++x) {
-        if (argb[x] < 0xff000000u) return 1;   // test any alpha values != 0xff
-      }
-      argb += picture->argb_stride;
-    }
+    const int alpha_offset = ALPHA_IS_LAST ? 3 : 0;
+    return CheckNonOpaque((const uint8_t*)picture->argb + alpha_offset,
+                          picture->width, picture->height,
+                          4, picture->argb_stride * sizeof(*picture->argb));
   }
   return 0;
 }
@@ -858,7 +856,6 @@ static int ImportYUVAFromRGBA(const uint8_t* r_ptr,
     return 0;
   }
   if (has_alpha) {
-    WebPInitAlphaProcessing();
     assert(step == 4);
 #if defined(USE_GAMMA_COMPRESSION) && defined(USE_INVERSE_ALPHA_TABLE)
     assert(kAlphaFix + kGammaFix <= 31);