From d581ba40ba8129809fed759f6b240b3d20da2892 Mon Sep 17 00:00:00 2001
From: Pascal Massimino <pascal.massimino@gmail.com>
Date: Sat, 10 Jan 2015 06:46:00 -0800
Subject: [PATCH] follow-up: clean up WebPRescalerXXX dsp function

by removing redundant RFIX macros and using a plain-C fallback.

Change-Id: I52436c672bf20780b6fe3bcf43fe73e1abac10ff
---
 src/dsp/dsp.h                  |  3 +++
 src/dsp/rescaler.c             | 10 +++++-----
 src/dsp/rescaler_mips32.c      | 17 +++--------------
 src/dsp/rescaler_mips_dsp_r2.c | 27 ++++++++-------------------
 4 files changed, 19 insertions(+), 38 deletions(-)

diff --git a/src/dsp/dsp.h b/src/dsp/dsp.h
index 10467431..c65248bb 100644
--- a/src/dsp/dsp.h
+++ b/src/dsp/dsp.h
@@ -297,6 +297,9 @@ extern void (*WebPRescalerImportRow)(struct WebPRescaler* const wrk,
 // Export one row (starting at x_out position) from rescaler.
 extern void (*WebPRescalerExportRow)(struct WebPRescaler* const wrk, int x_out);
 
+// Plain-C implementation, as fall-back.
+extern void WebPRescalerExportRowC(struct WebPRescaler* const wrk, int x_out);
+
 // Must be called first before using the above.
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void);
 
diff --git a/src/dsp/rescaler.c b/src/dsp/rescaler.c
index 60e6480f..640b462c 100644
--- a/src/dsp/rescaler.c
+++ b/src/dsp/rescaler.c
@@ -18,8 +18,8 @@
 #define ROUNDER (1 << (WEBP_RESCALER_RFIX - 1))
 #define MULT_FIX(x, y) (((int64_t)(x) * (y) + ROUNDER) >> WEBP_RESCALER_RFIX)
 
-static void ImportRowC(WebPRescaler* const wrk,
-                       const uint8_t* const src, int channel) {
+static void RescalerImportRowC(WebPRescaler* const wrk,
+                               const uint8_t* const src, int channel) {
   const int x_stride = wrk->num_channels;
   const int x_out_max = wrk->dst_width * wrk->num_channels;
   int x_in = channel;
@@ -61,7 +61,7 @@ static void ImportRowC(WebPRescaler* const wrk,
   }
 }
 
-static void ExportRowC(WebPRescaler* const wrk, int x_out) {
+void WebPRescalerExportRowC(WebPRescaler* const wrk, int x_out) {
   if (wrk->y_accum <= 0) {
     uint8_t* const dst = wrk->dst;
     int32_t* const irow = wrk->irow;
@@ -97,8 +97,8 @@ static volatile VP8CPUInfo rescaler_last_cpuinfo_used =
 WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void) {
   if (rescaler_last_cpuinfo_used == VP8GetCPUInfo) return;
 
-  WebPRescalerImportRow = ImportRowC;
-  WebPRescalerExportRow = ExportRowC;
+  WebPRescalerImportRow = RescalerImportRowC;
+  WebPRescalerExportRow = WebPRescalerExportRowC;
   if (VP8GetCPUInfo != NULL) {
 #if defined(WEBP_USE_MIPS32)
     if (VP8GetCPUInfo(kMIPS32)) {
diff --git a/src/dsp/rescaler_mips32.c b/src/dsp/rescaler_mips32.c
index 132d57f8..6d6d4378 100644
--- a/src/dsp/rescaler_mips32.c
+++ b/src/dsp/rescaler_mips32.c
@@ -114,9 +114,6 @@ static void ImportRow(WebPRescaler* const wrk,
   }
 }
 
-#define RFIX 30
-#define MULT_FIX(x, y) (((int64_t)(x) * (y) + (1 << (RFIX - 1))) >> RFIX)
-
 static void ExportRow(WebPRescaler* const wrk, int x_out) {
   if (wrk->y_accum <= 0) {
     uint8_t* const dst = wrk->dst;
@@ -172,22 +169,14 @@ static void ExportRow(WebPRescaler* const wrk, int x_out) {
         : [temp2]"r"(temp2), [yscale]"r"(yscale), [temp8]"r"(temp8)
         : "memory", "hi", "lo"
       );
+      wrk->y_accum += wrk->y_add;
+      wrk->dst += wrk->dst_stride;
     } else {
-      for (; x_out < x_out_max; ++x_out) {
-        const int frac = (int)MULT_FIX(frow[x_out], yscale);
-        const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
-        dst[x_out] = (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
-        irow[x_out] = frac;   // new fractional start
-      }
+      WebPRescalerExportRowC(wrk, x_out);
     }
-    wrk->y_accum += wrk->y_add;
-    wrk->dst += wrk->dst_stride;
   }
 }
 
-#undef MULT_FIX
-#undef RFIX
-
 #endif  // WEBP_USE_MIPS32
 
 //------------------------------------------------------------------------------
diff --git a/src/dsp/rescaler_mips_dsp_r2.c b/src/dsp/rescaler_mips_dsp_r2.c
index f09d2440..8dd64e53 100644
--- a/src/dsp/rescaler_mips_dsp_r2.c
+++ b/src/dsp/rescaler_mips_dsp_r2.c
@@ -110,19 +110,16 @@ static void ImportRow(WebPRescaler* const wrk,
   }
 }
 
-#define RFIX 30
-#define MULT_FIX(x, y) (((int64_t)(x) * (y) + (1 << (RFIX - 1))) >> RFIX)
-
 static void ExportRow(WebPRescaler* const wrk, int x_out) {
   if (wrk->y_accum <= 0) {
-    uint8_t* dst = wrk->dst;
-    int32_t* irow = wrk->irow;
-    const int32_t* frow = wrk->frow;
-    const int yscale = wrk->fy_scale * (-wrk->y_accum);
-    const int x_out_max = wrk->dst_width * wrk->num_channels;
     // if wrk->fxy_scale can fit into 32 bits use optimized code,
     // otherwise use C code
     if ((wrk->fxy_scale >> 32) == 0) {
+      uint8_t* dst = wrk->dst;
+      int32_t* irow = wrk->irow;
+      const int32_t* frow = wrk->frow;
+      const int yscale = wrk->fy_scale * (-wrk->y_accum);
+      const int x_out_max = wrk->dst_width * wrk->num_channels;
       int temp0, temp1, temp3, temp4, temp5, temp6, temp7;
       const int temp2 = (int)wrk->fxy_scale;
       const int rest = (x_out_max - x_out) & 1;
@@ -190,22 +187,14 @@ static void ExportRow(WebPRescaler* const wrk, int x_out) {
           [rest]"r"(rest)
         : "memory", "hi", "lo"
       );
+      wrk->y_accum += wrk->y_add;
+      wrk->dst += wrk->dst_stride;
     } else {
-      for (; x_out < x_out_max; ++x_out) {
-        const int frac = (int)MULT_FIX(frow[x_out], yscale);
-        const int v = (int)MULT_FIX(irow[x_out] - frac, wrk->fxy_scale);
-        dst[x_out] = (!(v & ~0xff)) ? v : (v < 0) ? 0 : 255;
-        irow[x_out] = frac;   // new fractional start
-      }
+      WebPRescalerExportRowC(wrk, x_out);
     }
-    wrk->y_accum += wrk->y_add;
-    wrk->dst += wrk->dst_stride;
   }
 }
 
-#undef MULT_FIX
-#undef RFIX
-
 #endif  // WEBP_USE_MIPS_DSP_R2
 
 //------------------------------------------------------------------------------