use mul_inv instead of div for tiling

Bug: skia:
Change-Id: I5f88e7923fe204faba8dc5d87454805a4d470d52
Reviewed-on: https://skia-review.googlesource.com/20688
Reviewed-by: Mike Klein <mtklein@chromium.org>
Reviewed-by: Florin Malita <fmalita@chromium.org>
Commit-Queue: Mike Reed <reed@google.com>

src/jumper/SkJumper.h

@@ -71,6 +71,11 @@ struct SkJumper_SamplerCtx {
     float scaley[SkJumper_kMaxStride];
 };
 
+struct SkJumper_TileCtx {
+    float scale;
+    float invScale; // cache of 1/scale
+};
+
 struct SkJumper_CallbackCtx {
     MAYBE_MSABI void (*fn)(SkJumper_CallbackCtx* self, int active_pixels/*<= SkJumper_kMaxStride*/);
 

src/jumper/SkJumper_stages.cpp

@@ -945,25 +945,27 @@ SI F ulp_before(F f) {
     return unaligned_load<F>(&bits);
 }
 
-SI F exclusive_clamp(F v, float limit) {
+SI F exclusive_clamp(F v, const SkJumper_TileCtx* ctx) {
     v = max(0,v);
-    return min(v, ulp_before(limit));
+    return min(v, ulp_before(ctx->scale));
 }
-SI F exclusive_repeat(F v, float limit) {
+SI F exclusive_repeat(F v, const SkJumper_TileCtx* ctx) {
-    v = v - floor_(v/limit)*limit;
+    v = v - floor_(v*ctx->invScale)*ctx->scale;
-    return min(v, ulp_before(limit));
+    return min(v, ulp_before(ctx->scale));
 }
-SI F exclusive_mirror(F v, float limit) {
+SI F exclusive_mirror(F v, const SkJumper_TileCtx* ctx) {
-    v = abs_( (v-limit) - (limit+limit)*floor_((v-limit)/(limit+limit)) - limit );
+    auto limit = ctx->scale;
+    auto invLimit = ctx->invScale;
+    v = abs_( (v-limit) - (limit+limit)*floor_((v-limit)*(invLimit*0.5f)) - limit );
     return min(v, ulp_before(limit));
 }
 // Clamp x or y to [0,limit) == [0,limit - 1 ulp] (think, sampling from images).
-STAGE(clamp_x)  { r = exclusive_clamp (r, *(const float*)ctx); }
+STAGE(clamp_x)  { r = exclusive_clamp (r, (const SkJumper_TileCtx*)ctx); }
-STAGE(clamp_y)  { g = exclusive_clamp (g, *(const float*)ctx); }
+STAGE(clamp_y)  { g = exclusive_clamp (g, (const SkJumper_TileCtx*)ctx); }
-STAGE(repeat_x) { r = exclusive_repeat(r, *(const float*)ctx); }
+STAGE(repeat_x) { r = exclusive_repeat(r, (const SkJumper_TileCtx*)ctx); }
-STAGE(repeat_y) { g = exclusive_repeat(g, *(const float*)ctx); }
+STAGE(repeat_y) { g = exclusive_repeat(g, (const SkJumper_TileCtx*)ctx); }
-STAGE(mirror_x) { r = exclusive_mirror(r, *(const float*)ctx); }
+STAGE(mirror_x) { r = exclusive_mirror(r, (const SkJumper_TileCtx*)ctx); }
-STAGE(mirror_y) { g = exclusive_mirror(g, *(const float*)ctx); }
+STAGE(mirror_y) { g = exclusive_mirror(g, (const SkJumper_TileCtx*)ctx); }
 
 // Clamp x to [0,1], both sides exclusive (think, gradients).
 STAGE( clamp_x_1) { r = min(max(0, r), 1.0f); }

src/shaders/SkImageShader.cpp

@@ -298,8 +298,12 @@ bool SkImageShader::onAppendStages(SkRasterPipeline* p, SkColorSpace* dstCS, SkA
     gather->ctable  = pm.ctable() ? pm.ctable()->readColors() : nullptr;
     gather->stride  = pm.rowBytesAsPixels();
 
-    auto limit_x = alloc->make<float>(pm. width()),
+    auto limit_x = alloc->make<SkJumper_TileCtx>(),
-         limit_y = alloc->make<float>(pm.height());
+         limit_y = alloc->make<SkJumper_TileCtx>();
+    limit_x->scale = pm.width();
+    limit_x->invScale = 1.0f / pm.width();
+    limit_y->scale = pm.height();
+    limit_y->invScale = 1.0f / pm.height();
 
     auto append_tiling_and_gather = [&] {
         switch (fTileModeX) {