introduce a separate WebPRescalerDspInit to initialize pointers

so that we keep the details of WebPRescaler in utils/rescaler.c
when possible.

Change-Id: Ib6c1029a09b84cbc7a7d2f70dafa4d4d9132cecc

src/dsp/dsp.h

@@ -284,6 +284,22 @@ WEBP_TSAN_IGNORE_FUNCTION void WebPInitSamplers(void);
 // Must be called before using WebPYUV444Converters[]
 WEBP_TSAN_IGNORE_FUNCTION void WebPInitYUV444Converters(void);
 
+//------------------------------------------------------------------------------
+// Rescaler
+
+struct WebPRescaler;
+
+// Import a row of data and save its contribution in the rescaler.
+// 'channel' denotes the channel number to be imported.
+extern void (*WebPRescalerImportRow)(struct WebPRescaler* const wrk,
+                                     const uint8_t* const src, int channel);
+
+// Export one row (starting at x_out position) from rescaler.
+extern void (*WebPRescalerExportRow)(struct WebPRescaler* const wrk, int x_out);
+
+// Must be called first before using the above.
+WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void);
+
 //------------------------------------------------------------------------------
 // Utilities for processing transparent channel.
 

src/dsp/rescaler.c

@@ -15,12 +15,8 @@
 //------------------------------------------------------------------------------
 // Implementations of critical functions ImportRow / ExportRow
 
-void (*WebPRescalerImportRow)(WebPRescaler* const wrk,
-                              const uint8_t* const src, int channel) = NULL;
-void (*WebPRescalerExportRow)(WebPRescaler* const wrk, int x_out) = NULL;
-
-#define RFIX 30
-#define MULT_FIX(x, y) (((int64_t)(x) * (y) + (1 << (RFIX - 1))) >> RFIX)
+#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) {
@@ -83,38 +79,24 @@ static void ExportRowC(WebPRescaler* const wrk, int x_out) {
   }
 }
 
+#undef MULT_FIX
+#undef ROUNDER
+
 //------------------------------------------------------------------------------
 
+void (*WebPRescalerImportRow)(struct WebPRescaler* const wrk,
+                              const uint8_t* const src, int channel);
+void (*WebPRescalerExportRow)(struct WebPRescaler* const wrk, int x_out);
+
 extern void WebPRescalerDspInitMIPS32(void);
 extern void WebPRescalerDspInitMIPSdspR2(void);
 
-void WebPRescalerInit(WebPRescaler* const wrk, int src_width, int src_height,
-                      uint8_t* const dst, int dst_width, int dst_height,
-                      int dst_stride, int num_channels, int x_add, int x_sub,
-                      int y_add, int y_sub, int32_t* const work) {
-  wrk->x_expand = (src_width < dst_width);
-  wrk->src_width = src_width;
-  wrk->src_height = src_height;
-  wrk->dst_width = dst_width;
-  wrk->dst_height = dst_height;
-  wrk->dst = dst;
-  wrk->dst_stride = dst_stride;
-  wrk->num_channels = num_channels;
-  // for 'x_expand', we use bilinear interpolation
-  wrk->x_add = wrk->x_expand ? (x_sub - 1) : x_add - x_sub;
-  wrk->x_sub = wrk->x_expand ? (x_add - 1) : x_sub;
-  wrk->y_accum = y_add;
-  wrk->y_add = y_add;
-  wrk->y_sub = y_sub;
-  wrk->fx_scale = (1 << RFIX) / x_sub;
-  wrk->fy_scale = (1 << RFIX) / y_sub;
-  wrk->fxy_scale = wrk->x_expand ?
-      ((int64_t)dst_height << RFIX) / (x_sub * src_height) :
-      ((int64_t)dst_height << RFIX) / (x_add * src_height);
-  wrk->irow = work;
-  wrk->frow = work + num_channels * dst_width;
+static volatile VP8CPUInfo rescaler_last_cpuinfo_used =
+    (VP8CPUInfo)&rescaler_last_cpuinfo_used;
+
+WEBP_TSAN_IGNORE_FUNCTION void WebPRescalerDspInit(void) {
+  if (rescaler_last_cpuinfo_used == VP8GetCPUInfo) return;
 
-  if (WebPRescalerImportRow == NULL) {
   WebPRescalerImportRow = ImportRowC;
   WebPRescalerExportRow = ExportRowC;
   if (VP8GetCPUInfo != NULL) {
@@ -129,8 +111,5 @@ void WebPRescalerInit(WebPRescaler* const wrk, int src_width, int src_height,
     }
 #endif
   }
+  rescaler_last_cpuinfo_used = VP8GetCPUInfo;
 }
-}
-
-#undef MULT_FIX
-#undef RFIX

src/utils/rescaler.c

@@ -13,8 +13,40 @@
 
 #include <assert.h>
 #include <stdlib.h>
+#include "../dsp/dsp.h"
 #include "./rescaler.h"
 
+//------------------------------------------------------------------------------
+
+void WebPRescalerInit(WebPRescaler* const wrk, int src_width, int src_height,
+                      uint8_t* const dst, int dst_width, int dst_height,
+                      int dst_stride, int num_channels, int x_add, int x_sub,
+                      int y_add, int y_sub, int32_t* const work) {
+  wrk->x_expand = (src_width < dst_width);
+  wrk->src_width = src_width;
+  wrk->src_height = src_height;
+  wrk->dst_width = dst_width;
+  wrk->dst_height = dst_height;
+  wrk->dst = dst;
+  wrk->dst_stride = dst_stride;
+  wrk->num_channels = num_channels;
+  // for 'x_expand', we use bilinear interpolation
+  wrk->x_add = wrk->x_expand ? (x_sub - 1) : x_add - x_sub;
+  wrk->x_sub = wrk->x_expand ? (x_add - 1) : x_sub;
+  wrk->y_accum = y_add;
+  wrk->y_add = y_add;
+  wrk->y_sub = y_sub;
+  wrk->fx_scale = (1 << WEBP_RESCALER_RFIX) / x_sub;
+  wrk->fy_scale = (1 << WEBP_RESCALER_RFIX) / y_sub;
+  wrk->fxy_scale = wrk->x_expand ?
+      ((int64_t)dst_height << WEBP_RESCALER_RFIX) / (x_sub * src_height) :
+      ((int64_t)dst_height << WEBP_RESCALER_RFIX) / (x_add * src_height);
+  wrk->irow = work;
+  wrk->frow = work + num_channels * dst_width;
+
+  WebPRescalerDspInit();
+}
+
 //------------------------------------------------------------------------------
 // all-in-one calls
 

src/utils/rescaler.h

@@ -20,8 +20,11 @@ extern "C" {
 
 #include "../webp/types.h"
 
+#define WEBP_RESCALER_RFIX 30   // fixed-point precision for multiplies
+
 // Structure used for on-the-fly rescaling
-typedef struct {
+typedef struct WebPRescaler WebPRescaler;
+struct WebPRescaler {
   int x_expand;               // true if we're expanding in the x direction
   int num_channels;           // bytes to jump between pixels
   int fy_scale, fx_scale;     // fixed-point scaling factor
@@ -35,7 +38,7 @@ typedef struct {
   uint8_t* dst;
   int dst_stride;
   int32_t* irow, *frow;       // work buffer
-} WebPRescaler;
+};
 
 // Initialize a rescaler given scratch area 'work' and dimensions of src & dst.
 void WebPRescalerInit(WebPRescaler* const rescaler,
@@ -57,13 +60,6 @@ int WebPRescaleNeededLines(const WebPRescaler* const rescaler,
 int WebPRescalerImport(WebPRescaler* const rescaler, int num_rows,
                        const uint8_t* src, int src_stride);
 
-// Import a row of data and save its contribution in the rescaler.
-// 'channel' denotes the channel number to be imported.
-extern void (*WebPRescalerImportRow)(WebPRescaler* const wrk,
-                                     const uint8_t* const src, int channel);
-// Export one row (starting at x_out position) from rescaler.
-extern void (*WebPRescalerExportRow)(WebPRescaler* const wrk, int x_out);
-
 // Return true if there is pending output rows ready.
 static WEBP_INLINE
 int WebPRescalerHasPendingOutput(const WebPRescaler* const rescaler) {