lowp: add move_src_dst and srcover

This is enough to run the bench SkRasterPipeline_compile.

    $ ninja -C out monobench; and out/monobench SkRasterPipeline_compile 300

    Before: 300  SkRasterPipeline_compile 48.4858ns
    After:  300  SkRasterPipeline_compile 37.5801ns

Change-Id: Icb80348908dfb016826700a44566222c9f7a853c
Reviewed-on: https://skia-review.googlesource.com/18595
Reviewed-by: Herb Derby <herb@google.com>
Commit-Queue: Mike Klein <mtklein@chromium.org>

4 files changed, 110 insertions, 23 deletions

diff --git a/src/jumper/SkJumper.cpp b/src/jumper/SkJumper.cpp
index 7059822647..535e4a883b 100644
--- a/src/jumper/SkJumper.cpp
+++ b/src/jumper/SkJumper.cpp
@@ -83,9 +83,11 @@ extern "C" {
         SK_RASTER_PIPELINE_STAGES(M)
     #undef M
 
-    StageFn ASM(load_8888,  ssse3_lowp),
-            ASM(store_8888, ssse3_lowp),
-            ASM(swap_rb,    ssse3_lowp);
+    StageFn ASM(load_8888,    ssse3_lowp),
+            ASM(store_8888,   ssse3_lowp),
+            ASM(swap_rb,      ssse3_lowp),
+            ASM(move_src_dst, ssse3_lowp),
+            ASM(srcover,      ssse3_lowp);
 #endif
 
     // Portable, single-pixel stages.
@@ -186,9 +188,11 @@ StartPipelineFn* SkRasterPipeline::build_pipeline(void** ip) const {
         for (const StageList* st = fStages; st; st = st->prev) {
             StageFn* fn = nullptr;
             switch (st->stage) {
-                case SkRasterPipeline::load_8888:  fn = ASM(load_8888, ssse3_lowp); break;
-                case SkRasterPipeline::store_8888: fn = ASM(store_8888,ssse3_lowp); break;
-                case SkRasterPipeline::swap_rb:    fn = ASM(swap_rb,   ssse3_lowp); break;
+                case SkRasterPipeline::load_8888:    fn = ASM(load_8888,    ssse3_lowp); break;
+                case SkRasterPipeline::store_8888:   fn = ASM(store_8888,   ssse3_lowp); break;
+                case SkRasterPipeline::swap_rb:      fn = ASM(swap_rb,      ssse3_lowp); break;
+                case SkRasterPipeline::move_src_dst: fn = ASM(move_src_dst, ssse3_lowp); break;
+                case SkRasterPipeline::srcover:      fn = ASM(srcover,      ssse3_lowp); break;
                 default:
                     //SkDebugf("can't %d\n", st->stage);
                     ip = reset_point;
diff --git a/src/jumper/SkJumper_generated.S b/src/jumper/SkJumper_generated.S
index 52bf029be2..de0fb731a0 100644
--- a/src/jumper/SkJumper_generated.S
+++ b/src/jumper/SkJumper_generated.S
@@ -36953,7 +36953,7 @@ _sk_load_8888_ssse3_lowp:
   .byte  117,113                             // jne           127 <_sk_load_8888_ssse3_lowp+0x7b>
   .byte  69,15,16,76,147,16                  // movups        0x10(%r11,%rdx,4),%xmm9
   .byte  69,15,16,4,147                      // movups        (%r11,%rdx,4),%xmm8
-  .byte  102,15,111,5,55,2,0,0               // movdqa        0x237(%rip),%xmm0        # 300 <_sk_swap_rb_ssse3_lowp+0x14>
+  .byte  102,15,111,5,167,2,0,0              // movdqa        0x2a7(%rip),%xmm0        # 370 <_sk_srcover_ssse3_lowp+0x65>
   .byte  102,68,15,56,0,192                  // pshufb        %xmm0,%xmm8
   .byte  102,68,15,56,0,200                  // pshufb        %xmm0,%xmm9
   .byte  102,65,15,111,208                   // movdqa        %xmm8,%xmm2
@@ -36967,7 +36967,7 @@ _sk_load_8888_ssse3_lowp:
   .byte  102,15,239,210                      // pxor          %xmm2,%xmm2
   .byte  102,65,15,96,208                    // punpcklbw     %xmm8,%xmm2
   .byte  102,65,15,104,216                   // punpckhbw     %xmm8,%xmm3
-  .byte  102,68,15,111,5,1,2,0,0             // movdqa        0x201(%rip),%xmm8        # 310 <_sk_swap_rb_ssse3_lowp+0x24>
+  .byte  102,68,15,111,5,113,2,0,0           // movdqa        0x271(%rip),%xmm8        # 380 <_sk_srcover_ssse3_lowp+0x75>
   .byte  102,65,15,228,192                   // pmulhuw       %xmm8,%xmm0
   .byte  102,65,15,228,200                   // pmulhuw       %xmm8,%xmm1
   .byte  102,65,15,228,208                   // pmulhuw       %xmm8,%xmm2
@@ -37109,6 +37109,41 @@ _sk_swap_rb_ssse3_lowp:
   .byte  65,15,40,208                        // movaps        %xmm8,%xmm2
   .byte  255,224                             // jmpq          *%rax
 
+HIDDEN _sk_move_src_dst_ssse3_lowp
+.globl _sk_move_src_dst_ssse3_lowp
+FUNCTION(_sk_move_src_dst_ssse3_lowp)
+_sk_move_src_dst_ssse3_lowp:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  15,40,224                           // movaps        %xmm0,%xmm4
+  .byte  15,40,233                           // movaps        %xmm1,%xmm5
+  .byte  15,40,242                           // movaps        %xmm2,%xmm6
+  .byte  15,40,251                           // movaps        %xmm3,%xmm7
+  .byte  255,224                             // jmpq          *%rax
+
+HIDDEN _sk_srcover_ssse3_lowp
+.globl _sk_srcover_ssse3_lowp
+FUNCTION(_sk_srcover_ssse3_lowp)
+_sk_srcover_ssse3_lowp:
+  .byte  102,68,15,111,5,124,0,0,0           // movdqa        0x7c(%rip),%xmm8        # 390 <_sk_srcover_ssse3_lowp+0x85>
+  .byte  102,68,15,249,195                   // psubw         %xmm3,%xmm8
+  .byte  102,68,15,111,204                   // movdqa        %xmm4,%xmm9
+  .byte  102,69,15,56,11,200                 // pmulhrsw      %xmm8,%xmm9
+  .byte  102,69,15,56,29,201                 // pabsw         %xmm9,%xmm9
+  .byte  102,65,15,253,193                   // paddw         %xmm9,%xmm0
+  .byte  102,68,15,111,205                   // movdqa        %xmm5,%xmm9
+  .byte  102,69,15,56,11,200                 // pmulhrsw      %xmm8,%xmm9
+  .byte  102,69,15,56,29,201                 // pabsw         %xmm9,%xmm9
+  .byte  102,65,15,253,201                   // paddw         %xmm9,%xmm1
+  .byte  102,68,15,111,206                   // movdqa        %xmm6,%xmm9
+  .byte  102,69,15,56,11,200                 // pmulhrsw      %xmm8,%xmm9
+  .byte  102,69,15,56,29,201                 // pabsw         %xmm9,%xmm9
+  .byte  102,65,15,253,209                   // paddw         %xmm9,%xmm2
+  .byte  102,68,15,56,11,199                 // pmulhrsw      %xmm7,%xmm8
+  .byte  102,69,15,56,29,192                 // pabsw         %xmm8,%xmm8
+  .byte  102,65,15,253,216                   // paddw         %xmm8,%xmm3
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  255,224                             // jmpq          *%rax
+
 BALIGN16
   .byte  0,4,8                               // add           %al,(%rax,%rcx,1)
   .byte  12,1                                // or            $0x1,%al
@@ -37117,10 +37152,7 @@ BALIGN16
   .byte  3,7                                 // add           (%rdi),%eax
   .byte  11,15                               // or            (%rdi),%ecx
   .byte  129,128,129,128,129,128,129,128,129,128// addl          $0x80818081,-0x7f7e7f7f(%rax)
-  .byte  129                                 // .byte         0x81
-  .byte  128                                 // .byte         0x80
-  .byte  129                                 // .byte         0x81
-  .byte  128                                 // .byte         0x80
-  .byte  129                                 // .byte         0x81
-  .byte  128                                 // .byte         0x80
+  .byte  129,128,129,128,129,128,0,128,0,128 // addl          $0x80008000,-0x7f7e7f7f(%rax)
+  .byte  0,128,0,128,0,128                   // add           %al,-0x7fff8000(%rax)
+  .byte  0,128,0,128,0,128                   // add           %al,-0x7fff8000(%rax)
 #endif
diff --git a/src/jumper/SkJumper_generated_win.S b/src/jumper/SkJumper_generated_win.S
index ecbffaa201..67c9b86a2b 100644
--- a/src/jumper/SkJumper_generated_win.S
+++ b/src/jumper/SkJumper_generated_win.S
@@ -26366,7 +26366,7 @@ _sk_load_8888_ssse3_lowp LABEL PROC
   DB  117,113                             ; jne           1b7 <_sk_load_8888_ssse3_lowp+0x7b>
   DB  69,15,16,76,147,16                  ; movups        0x10(%r11,%rdx,4),%xmm9
   DB  69,15,16,4,147                      ; movups        (%r11,%rdx,4),%xmm8
-  DB  102,15,111,5,55,2,0,0               ; movdqa        0x237(%rip),%xmm0        # 390 <_sk_swap_rb_ssse3_lowp+0x14>
+  DB  102,15,111,5,167,2,0,0              ; movdqa        0x2a7(%rip),%xmm0        # 400 <_sk_srcover_ssse3_lowp+0x65>
   DB  102,68,15,56,0,192                  ; pshufb        %xmm0,%xmm8
   DB  102,68,15,56,0,200                  ; pshufb        %xmm0,%xmm9
   DB  102,65,15,111,208                   ; movdqa        %xmm8,%xmm2
@@ -26380,7 +26380,7 @@ _sk_load_8888_ssse3_lowp LABEL PROC
   DB  102,15,239,210                      ; pxor          %xmm2,%xmm2
   DB  102,65,15,96,208                    ; punpcklbw     %xmm8,%xmm2
   DB  102,65,15,104,216                   ; punpckhbw     %xmm8,%xmm3
-  DB  102,68,15,111,5,1,2,0,0             ; movdqa        0x201(%rip),%xmm8        # 3a0 <_sk_swap_rb_ssse3_lowp+0x24>
+  DB  102,68,15,111,5,113,2,0,0           ; movdqa        0x271(%rip),%xmm8        # 410 <_sk_srcover_ssse3_lowp+0x75>
   DB  102,65,15,228,192                   ; pmulhuw       %xmm8,%xmm0
   DB  102,65,15,228,200                   ; pmulhuw       %xmm8,%xmm1
   DB  102,65,15,228,208                   ; pmulhuw       %xmm8,%xmm2
@@ -26518,6 +26518,37 @@ _sk_swap_rb_ssse3_lowp LABEL PROC
   DB  65,15,40,208                        ; movaps        %xmm8,%xmm2
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_move_src_dst_ssse3_lowp
+_sk_move_src_dst_ssse3_lowp LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  15,40,224                           ; movaps        %xmm0,%xmm4
+  DB  15,40,233                           ; movaps        %xmm1,%xmm5
+  DB  15,40,242                           ; movaps        %xmm2,%xmm6
+  DB  15,40,251                           ; movaps        %xmm3,%xmm7
+  DB  255,224                             ; jmpq          *%rax
+
+PUBLIC _sk_srcover_ssse3_lowp
+_sk_srcover_ssse3_lowp LABEL PROC
+  DB  102,68,15,111,5,124,0,0,0           ; movdqa        0x7c(%rip),%xmm8        # 420 <_sk_srcover_ssse3_lowp+0x85>
+  DB  102,68,15,249,195                   ; psubw         %xmm3,%xmm8
+  DB  102,68,15,111,204                   ; movdqa        %xmm4,%xmm9
+  DB  102,69,15,56,11,200                 ; pmulhrsw      %xmm8,%xmm9
+  DB  102,69,15,56,29,201                 ; pabsw         %xmm9,%xmm9
+  DB  102,65,15,253,193                   ; paddw         %xmm9,%xmm0
+  DB  102,68,15,111,205                   ; movdqa        %xmm5,%xmm9
+  DB  102,69,15,56,11,200                 ; pmulhrsw      %xmm8,%xmm9
+  DB  102,69,15,56,29,201                 ; pabsw         %xmm9,%xmm9
+  DB  102,65,15,253,201                   ; paddw         %xmm9,%xmm1
+  DB  102,68,15,111,206                   ; movdqa        %xmm6,%xmm9
+  DB  102,69,15,56,11,200                 ; pmulhrsw      %xmm8,%xmm9
+  DB  102,69,15,56,29,201                 ; pabsw         %xmm9,%xmm9
+  DB  102,65,15,253,209                   ; paddw         %xmm9,%xmm2
+  DB  102,68,15,56,11,199                 ; pmulhrsw      %xmm7,%xmm8
+  DB  102,69,15,56,29,192                 ; pabsw         %xmm8,%xmm8
+  DB  102,65,15,253,216                   ; paddw         %xmm8,%xmm3
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  255,224                             ; jmpq          *%rax
+
 ALIGN 16
   DB  0,4,8                               ; add           %al,(%rax,%rcx,1)
   DB  12,1                                ; or            $0x1,%al
@@ -26526,11 +26557,8 @@ ALIGN 16
   DB  3,7                                 ; add           (%rdi),%eax
   DB  11,15                               ; or            (%rdi),%ecx
   DB  129,128,129,128,129,128,129,128,129,128; addl          $0x80818081,-0x7f7e7f7f(%rax)
-  DB  129                                 ; .byte         0x81
-  DB  128                                 ; .byte         0x80
-  DB  129                                 ; .byte         0x81
-  DB  128                                 ; .byte         0x80
-  DB  129                                 ; .byte         0x81
-  DB  128                                 ; .byte         0x80
+  DB  129,128,129,128,129,128,0,128,0,128 ; addl          $0x80008000,-0x7f7e7f7f(%rax)
+  DB  0,128,0,128,0,128                   ; add           %al,-0x7fff8000(%rax)
+  DB  0,128,0,128,0,128                   ; add           %al,-0x7fff8000(%rax)
 ENDIF
 END
diff --git a/src/jumper/SkJumper_stages_lowp.cpp b/src/jumper/SkJumper_stages_lowp.cpp
index 4d0df94af6..139382377d 100644
--- a/src/jumper/SkJumper_stages_lowp.cpp
+++ b/src/jumper/SkJumper_stages_lowp.cpp
@@ -28,7 +28,7 @@ struct F {
     U16 vec;
 
     F() = default;
-    F(uint16_t bits) : vec(bits) {}
+    F(float f) : vec((uint16_t)(f * 0x8000)) {}
 
     F(U16 v) : vec(v) {}
     operator U16() const { return vec; }
@@ -38,6 +38,7 @@ SI F operator+(F x, F y) { return x.vec + y.vec; }
 SI F operator-(F x, F y) { return x.vec - y.vec; }
 SI F operator*(F x, F y) { return _mm_abs_epi16(_mm_mulhrs_epi16(x.vec, y.vec)); }
 SI F mad(F f, F m, F a) { return f*m+a; }
+SI F inv(F v) { return 1.0f - v; }
 
 SI F operator<<(F x, int bits) { return x.vec << bits; }
 SI F operator>>(F x, int bits) { return x.vec >> bits; }
@@ -178,3 +179,25 @@ STAGE(swap_rb) {
     r = b;
     b = tmp;
 }
+
+STAGE(move_src_dst) {
+    dr = r;
+    dg = g;
+    db = b;
+    da = a;
+}
+
+// Most blend modes apply the same logic to each channel.
+#define BLEND_MODE(name)                       \
+    SI F name##_channel(F s, F d, F sa, F da); \
+    STAGE(name) {                              \
+        r = name##_channel(r,dr,a,da);         \
+        g = name##_channel(g,dg,a,da);         \
+        b = name##_channel(b,db,a,da);         \
+        a = name##_channel(a,da,a,da);         \
+    }                                          \
+    SI F name##_channel(F s, F d, F sa, F da)
+
+BLEND_MODE(srcover) { return mad(d, inv(sa), s); }
+
+#undef BLEND_MODE