add a callback stage to SkRasterPipeline

This lets us temporarily escape to piece of code outside
SkRasterPipeline.  We should be able to use this to replace
   - parametric_{r,g,b,a}
   - table_{r,g,b,a}
   - color_lookup_table
   - shader_adapter*

* We want to obsolete shader_adapter for other reasons anyway,
but we _could_ replace it with this if we want to.

Change-Id: I42b657b3c19c679796ed1876856cae0c8471307e
Reviewed-on: https://skia-review.googlesource.com/12102
Commit-Queue: Mike Klein <mtklein@chromium.org>
Reviewed-by: Herb Derby <herb@google.com>
Reviewed-by: Matt Sarett <msarett@google.com>

6 files changed, 406 insertions, 9 deletions

diff --git a/src/core/SkRasterPipeline.h b/src/core/SkRasterPipeline.h
index ee4479de06..29c560d7c7 100644
--- a/src/core/SkRasterPipeline.h
+++ b/src/core/SkRasterPipeline.h
@@ -56,6 +56,7 @@
 // the Stage*.  This mostly matters on 64-bit Windows where every register is precious.
 
 #define SK_RASTER_PIPELINE_STAGES(M)                             \
+    M(callback)                                                  \
     M(move_src_dst) M(move_dst_src) M(swap)                      \
     M(clamp_0) M(clamp_1) M(clamp_a)                             \
     M(unpremul) M(premul)                                        \
diff --git a/src/jumper/SkJumper.h b/src/jumper/SkJumper.h
index 7a3f4e85f5..0567e75eb0 100644
--- a/src/jumper/SkJumper.h
+++ b/src/jumper/SkJumper.h
@@ -12,6 +12,13 @@
 // and SkJumper_stages.cpp (compiled into Skia _and_ offline into SkJumper_generated.h).
 // Keep it simple!
 
+// Sometimes we need to make sure externally facing functions are called with MS' ABI, not System V.
+#if defined(JUMPER) && defined(WIN)
+    #define MAYBE_MSABI __attribute__((ms_abi))
+#else
+    #define MAYBE_MSABI
+#endif
+
 #if defined(JUMPER) && (defined(__aarch64__) || defined(__arm__))
     // To reduce SkJumper's dependency on the Android NDK,
     // we provide what we need from <string.h>, <stdint.h>, and <stddef.h> ourselves.
@@ -74,4 +81,9 @@ struct SkJumper_SamplerCtx {
     float scaley[SkJumper_kMaxStride];
 };
 
+struct SkJumper_CallbackCtx {
+    MAYBE_MSABI void (*fn)(void* arg, int active_pixels/*<= SkJumper_kMaxStride*/);
+    void* arg;
+};
+
 #endif//SkJumper_DEFINED
diff --git a/src/jumper/SkJumper_generated.S b/src/jumper/SkJumper_generated.S
index 54ee03807b..0de21d61e7 100644
--- a/src/jumper/SkJumper_generated.S
+++ b/src/jumper/SkJumper_generated.S
@@ -3028,6 +3028,38 @@ _sk_bicubic_p3y_aarch64:
   .long  0x4e21d641                          // fadd          v1.4s, v18.4s, v1.4s
   .long  0x91004021                          // add           x1, x1, #0x10
   .long  0xd61f0060                          // br            x3
+
+HIDDEN _sk_callback_aarch64
+.globl _sk_callback_aarch64
+FUNCTION(_sk_callback_aarch64)
+_sk_callback_aarch64:
+  .long  0xd10283ff                          // sub           sp, sp, #0xa0
+  .long  0xa90853f5                          // stp           x21, x20, [sp, #128]
+  .long  0xa9097bf3                          // stp           x19, x30, [sp, #144]
+  .long  0xad031fe6                          // stp           q6, q7, [sp, #96]
+  .long  0xad0217e4                          // stp           q4, q5, [sp, #64]
+  .long  0xad010fe2                          // stp           q2, q3, [sp, #32]
+  .long  0xad0007e0                          // stp           q0, q1, [sp]
+  .long  0xaa0103f4                          // mov           x20, x1
+  .long  0xf9400288                          // ldr           x8, [x20]
+  .long  0xaa0003f5                          // mov           x21, x0
+  .long  0x321e03e1                          // orr           w1, wzr, #0x4
+  .long  0xaa0203f3                          // mov           x19, x2
+  .long  0xa9402109                          // ldp           x9, x8, [x8]
+  .long  0xaa0803e0                          // mov           x0, x8
+  .long  0xd63f0120                          // blr           x9
+  .long  0xf9400683                          // ldr           x3, [x20, #8]
+  .long  0x91004281                          // add           x1, x20, #0x10
+  .long  0xaa1503e0                          // mov           x0, x21
+  .long  0xaa1303e2                          // mov           x2, x19
+  .long  0xad4007e0                          // ldp           q0, q1, [sp]
+  .long  0xad410fe2                          // ldp           q2, q3, [sp, #32]
+  .long  0xad4217e4                          // ldp           q4, q5, [sp, #64]
+  .long  0xad431fe6                          // ldp           q6, q7, [sp, #96]
+  .long  0xa9497bf3                          // ldp           x19, x30, [sp, #144]
+  .long  0xa94853f5                          // ldp           x21, x20, [sp, #128]
+  .long  0x910283ff                          // add           sp, sp, #0xa0
+  .long  0xd61f0060                          // br            x3
 #elif defined(__arm__)
 .balign 4
 
@@ -6286,6 +6318,44 @@ _sk_bicubic_p3y_vfp4:
   .long  0x3ec71c72                          // .word         0x3ec71c72
   .long  0xbeaaaaab                          // .word         0xbeaaaaab
   .long  0xbeaaaaab                          // .word         0xbeaaaaab
+
+HIDDEN _sk_callback_vfp4
+.globl _sk_callback_vfp4
+FUNCTION(_sk_callback_vfp4)
+_sk_callback_vfp4:
+  .long  0xe92d4070                          // push          {r4, r5, r6, lr}
+  .long  0xed2d8b10                          // vpush         {d8-d15}
+  .long  0xe1a05001                          // mov           r5, r1
+  .long  0xe1a06000                          // mov           r6, r0
+  .long  0xe5950000                          // ldr           r0, [r5]
+  .long  0xe1a04002                          // mov           r4, r2
+  .long  0xe3a01002                          // mov           r1, #2
+  .long  0xeeb08b47                          // vmov.f64      d8, d7
+  .long  0xe5902000                          // ldr           r2, [r0]
+  .long  0xe5900004                          // ldr           r0, [r0, #4]
+  .long  0xeeb09b46                          // vmov.f64      d9, d6
+  .long  0xeeb0ab45                          // vmov.f64      d10, d5
+  .long  0xeeb0bb44                          // vmov.f64      d11, d4
+  .long  0xeeb0cb43                          // vmov.f64      d12, d3
+  .long  0xeeb0db42                          // vmov.f64      d13, d2
+  .long  0xeeb0eb41                          // vmov.f64      d14, d1
+  .long  0xeeb0fb40                          // vmov.f64      d15, d0
+  .long  0xe12fff32                          // blx           r2
+  .long  0xe2851008                          // add           r1, r5, #8
+  .long  0xe5953004                          // ldr           r3, [r5, #4]
+  .long  0xe1a00006                          // mov           r0, r6
+  .long  0xe1a02004                          // mov           r2, r4
+  .long  0xeeb00b4f                          // vmov.f64      d0, d15
+  .long  0xeeb01b4e                          // vmov.f64      d1, d14
+  .long  0xeeb02b4d                          // vmov.f64      d2, d13
+  .long  0xeeb03b4c                          // vmov.f64      d3, d12
+  .long  0xeeb04b4b                          // vmov.f64      d4, d11
+  .long  0xeeb05b4a                          // vmov.f64      d5, d10
+  .long  0xeeb06b49                          // vmov.f64      d6, d9
+  .long  0xeeb07b48                          // vmov.f64      d7, d8
+  .long  0xecbd8b10                          // vpop          {d8-d15}
+  .long  0xe8bd4070                          // pop           {r4, r5, r6, lr}
+  .long  0xe12fff13                          // bx            r3
 #elif defined(__x86_64__)
 
 HIDDEN _sk_start_pipeline_hsw
@@ -8700,7 +8770,7 @@ _sk_load_4444_hsw:
   .byte  255                                 // (bad)
   .byte  255                                 // (bad)
   .byte  255                                 // (bad)
-  .byte  233,255,255,255,225                 // jmpq          ffffffffe2002284 <_sk_bicubic_p3y_hsw+0xffffffffe1ffecd5>
+  .byte  233,255,255,255,225                 // jmpq          ffffffffe2002284 <_sk_callback_hsw+0xffffffffe1ffec7e>
   .byte  255                                 // (bad)
   .byte  255                                 // (bad)
   .byte  255                                 // (bad)
@@ -10023,6 +10093,54 @@ _sk_bicubic_p3y_hsw:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
   .byte  255,224                             // jmpq          *%rax
 
+HIDDEN _sk_callback_hsw
+.globl _sk_callback_hsw
+FUNCTION(_sk_callback_hsw)
+_sk_callback_hsw:
+  .byte  65,87                               // push          %r15
+  .byte  65,86                               // push          %r14
+  .byte  65,84                               // push          %r12
+  .byte  83                                  // push          %rbx
+  .byte  72,129,236,24,1,0,0                 // sub           $0x118,%rsp
+  .byte  197,252,17,188,36,224,0,0,0         // vmovups       %ymm7,0xe0(%rsp)
+  .byte  197,252,17,180,36,192,0,0,0         // vmovups       %ymm6,0xc0(%rsp)
+  .byte  197,252,17,172,36,160,0,0,0         // vmovups       %ymm5,0xa0(%rsp)
+  .byte  197,252,17,164,36,128,0,0,0         // vmovups       %ymm4,0x80(%rsp)
+  .byte  197,252,17,92,36,96                 // vmovups       %ymm3,0x60(%rsp)
+  .byte  197,252,17,84,36,64                 // vmovups       %ymm2,0x40(%rsp)
+  .byte  197,252,17,76,36,32                 // vmovups       %ymm1,0x20(%rsp)
+  .byte  197,252,17,4,36                     // vmovups       %ymm0,(%rsp)
+  .byte  72,137,203                          // mov           %rcx,%rbx
+  .byte  73,137,214                          // mov           %rdx,%r14
+  .byte  73,137,255                          // mov           %rdi,%r15
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  73,137,244                          // mov           %rsi,%r12
+  .byte  72,139,120,8                        // mov           0x8(%rax),%rdi
+  .byte  72,133,219                          // test          %rbx,%rbx
+  .byte  190,8,0,0,0                         // mov           $0x8,%esi
+  .byte  15,69,243                           // cmovne        %ebx,%esi
+  .byte  197,248,119                         // vzeroupper
+  .byte  255,16                              // callq         *(%rax)
+  .byte  76,137,230                          // mov           %r12,%rsi
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  76,137,255                          // mov           %r15,%rdi
+  .byte  76,137,242                          // mov           %r14,%rdx
+  .byte  72,137,217                          // mov           %rbx,%rcx
+  .byte  197,252,16,4,36                     // vmovups       (%rsp),%ymm0
+  .byte  197,252,16,76,36,32                 // vmovups       0x20(%rsp),%ymm1
+  .byte  197,252,16,84,36,64                 // vmovups       0x40(%rsp),%ymm2
+  .byte  197,252,16,92,36,96                 // vmovups       0x60(%rsp),%ymm3
+  .byte  197,252,16,164,36,128,0,0,0         // vmovups       0x80(%rsp),%ymm4
+  .byte  197,252,16,172,36,160,0,0,0         // vmovups       0xa0(%rsp),%ymm5
+  .byte  197,252,16,180,36,192,0,0,0         // vmovups       0xc0(%rsp),%ymm6
+  .byte  197,252,16,188,36,224,0,0,0         // vmovups       0xe0(%rsp),%ymm7
+  .byte  72,129,196,24,1,0,0                 // add           $0x118,%rsp
+  .byte  91                                  // pop           %rbx
+  .byte  65,92                               // pop           %r12
+  .byte  65,94                               // pop           %r14
+  .byte  65,95                               // pop           %r15
+  .byte  255,224                             // jmpq          *%rax
+
 HIDDEN _sk_start_pipeline_avx
 .globl _sk_start_pipeline_avx
 FUNCTION(_sk_start_pipeline_avx)
@@ -14570,6 +14688,54 @@ _sk_bicubic_p3y_avx:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
   .byte  255,224                             // jmpq          *%rax
 
+HIDDEN _sk_callback_avx
+.globl _sk_callback_avx
+FUNCTION(_sk_callback_avx)
+_sk_callback_avx:
+  .byte  65,87                               // push          %r15
+  .byte  65,86                               // push          %r14
+  .byte  65,84                               // push          %r12
+  .byte  83                                  // push          %rbx
+  .byte  72,129,236,24,1,0,0                 // sub           $0x118,%rsp
+  .byte  197,252,17,188,36,224,0,0,0         // vmovups       %ymm7,0xe0(%rsp)
+  .byte  197,252,17,180,36,192,0,0,0         // vmovups       %ymm6,0xc0(%rsp)
+  .byte  197,252,17,172,36,160,0,0,0         // vmovups       %ymm5,0xa0(%rsp)
+  .byte  197,252,17,164,36,128,0,0,0         // vmovups       %ymm4,0x80(%rsp)
+  .byte  197,252,17,92,36,96                 // vmovups       %ymm3,0x60(%rsp)
+  .byte  197,252,17,84,36,64                 // vmovups       %ymm2,0x40(%rsp)
+  .byte  197,252,17,76,36,32                 // vmovups       %ymm1,0x20(%rsp)
+  .byte  197,252,17,4,36                     // vmovups       %ymm0,(%rsp)
+  .byte  72,137,203                          // mov           %rcx,%rbx
+  .byte  73,137,214                          // mov           %rdx,%r14
+  .byte  73,137,255                          // mov           %rdi,%r15
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  73,137,244                          // mov           %rsi,%r12
+  .byte  72,139,120,8                        // mov           0x8(%rax),%rdi
+  .byte  72,133,219                          // test          %rbx,%rbx
+  .byte  190,8,0,0,0                         // mov           $0x8,%esi
+  .byte  15,69,243                           // cmovne        %ebx,%esi
+  .byte  197,248,119                         // vzeroupper
+  .byte  255,16                              // callq         *(%rax)
+  .byte  76,137,230                          // mov           %r12,%rsi
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  76,137,255                          // mov           %r15,%rdi
+  .byte  76,137,242                          // mov           %r14,%rdx
+  .byte  72,137,217                          // mov           %rbx,%rcx
+  .byte  197,252,16,4,36                     // vmovups       (%rsp),%ymm0
+  .byte  197,252,16,76,36,32                 // vmovups       0x20(%rsp),%ymm1
+  .byte  197,252,16,84,36,64                 // vmovups       0x40(%rsp),%ymm2
+  .byte  197,252,16,92,36,96                 // vmovups       0x60(%rsp),%ymm3
+  .byte  197,252,16,164,36,128,0,0,0         // vmovups       0x80(%rsp),%ymm4
+  .byte  197,252,16,172,36,160,0,0,0         // vmovups       0xa0(%rsp),%ymm5
+  .byte  197,252,16,180,36,192,0,0,0         // vmovups       0xc0(%rsp),%ymm6
+  .byte  197,252,16,188,36,224,0,0,0         // vmovups       0xe0(%rsp),%ymm7
+  .byte  72,129,196,24,1,0,0                 // add           $0x118,%rsp
+  .byte  91                                  // pop           %rbx
+  .byte  65,92                               // pop           %r12
+  .byte  65,94                               // pop           %r14
+  .byte  65,95                               // pop           %r15
+  .byte  255,224                             // jmpq          *%rax
+
 HIDDEN _sk_start_pipeline_sse41
 .globl _sk_start_pipeline_sse41
 FUNCTION(_sk_start_pipeline_sse41)
@@ -18250,6 +18416,47 @@ _sk_bicubic_p3y_sse41:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
   .byte  255,224                             // jmpq          *%rax
 
+HIDDEN _sk_callback_sse41
+.globl _sk_callback_sse41
+FUNCTION(_sk_callback_sse41)
+_sk_callback_sse41:
+  .byte  65,87                               // push          %r15
+  .byte  65,86                               // push          %r14
+  .byte  83                                  // push          %rbx
+  .byte  72,129,236,128,0,0,0                // sub           $0x80,%rsp
+  .byte  15,41,124,36,112                    // movaps        %xmm7,0x70(%rsp)
+  .byte  15,41,116,36,96                     // movaps        %xmm6,0x60(%rsp)
+  .byte  15,41,108,36,80                     // movaps        %xmm5,0x50(%rsp)
+  .byte  15,41,100,36,64                     // movaps        %xmm4,0x40(%rsp)
+  .byte  15,41,92,36,48                      // movaps        %xmm3,0x30(%rsp)
+  .byte  15,41,84,36,32                      // movaps        %xmm2,0x20(%rsp)
+  .byte  15,41,76,36,16                      // movaps        %xmm1,0x10(%rsp)
+  .byte  15,41,4,36                          // movaps        %xmm0,(%rsp)
+  .byte  73,137,214                          // mov           %rdx,%r14
+  .byte  73,137,255                          // mov           %rdi,%r15
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,137,243                          // mov           %rsi,%rbx
+  .byte  72,139,120,8                        // mov           0x8(%rax),%rdi
+  .byte  190,4,0,0,0                         // mov           $0x4,%esi
+  .byte  255,16                              // callq         *(%rax)
+  .byte  72,137,222                          // mov           %rbx,%rsi
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  76,137,255                          // mov           %r15,%rdi
+  .byte  76,137,242                          // mov           %r14,%rdx
+  .byte  15,40,4,36                          // movaps        (%rsp),%xmm0
+  .byte  15,40,76,36,16                      // movaps        0x10(%rsp),%xmm1
+  .byte  15,40,84,36,32                      // movaps        0x20(%rsp),%xmm2
+  .byte  15,40,92,36,48                      // movaps        0x30(%rsp),%xmm3
+  .byte  15,40,100,36,64                     // movaps        0x40(%rsp),%xmm4
+  .byte  15,40,108,36,80                     // movaps        0x50(%rsp),%xmm5
+  .byte  15,40,116,36,96                     // movaps        0x60(%rsp),%xmm6
+  .byte  15,40,124,36,112                    // movaps        0x70(%rsp),%xmm7
+  .byte  72,129,196,128,0,0,0                // add           $0x80,%rsp
+  .byte  91                                  // pop           %rbx
+  .byte  65,94                               // pop           %r14
+  .byte  65,95                               // pop           %r15
+  .byte  255,224                             // jmpq          *%rax
+
 HIDDEN _sk_start_pipeline_sse2
 .globl _sk_start_pipeline_sse2
 FUNCTION(_sk_start_pipeline_sse2)
@@ -22177,4 +22384,45 @@ _sk_bicubic_p3y_sse2:
   .byte  68,15,17,128,160,0,0,0              // movups        %xmm8,0xa0(%rax)
   .byte  72,173                              // lods          %ds:(%rsi),%rax
   .byte  255,224                             // jmpq          *%rax
+
+HIDDEN _sk_callback_sse2
+.globl _sk_callback_sse2
+FUNCTION(_sk_callback_sse2)
+_sk_callback_sse2:
+  .byte  65,87                               // push          %r15
+  .byte  65,86                               // push          %r14
+  .byte  83                                  // push          %rbx
+  .byte  72,129,236,128,0,0,0                // sub           $0x80,%rsp
+  .byte  15,41,124,36,112                    // movaps        %xmm7,0x70(%rsp)
+  .byte  15,41,116,36,96                     // movaps        %xmm6,0x60(%rsp)
+  .byte  15,41,108,36,80                     // movaps        %xmm5,0x50(%rsp)
+  .byte  15,41,100,36,64                     // movaps        %xmm4,0x40(%rsp)
+  .byte  15,41,92,36,48                      // movaps        %xmm3,0x30(%rsp)
+  .byte  15,41,84,36,32                      // movaps        %xmm2,0x20(%rsp)
+  .byte  15,41,76,36,16                      // movaps        %xmm1,0x10(%rsp)
+  .byte  15,41,4,36                          // movaps        %xmm0,(%rsp)
+  .byte  73,137,214                          // mov           %rdx,%r14
+  .byte  73,137,255                          // mov           %rdi,%r15
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,137,243                          // mov           %rsi,%rbx
+  .byte  72,139,120,8                        // mov           0x8(%rax),%rdi
+  .byte  190,4,0,0,0                         // mov           $0x4,%esi
+  .byte  255,16                              // callq         *(%rax)
+  .byte  72,137,222                          // mov           %rbx,%rsi
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  76,137,255                          // mov           %r15,%rdi
+  .byte  76,137,242                          // mov           %r14,%rdx
+  .byte  15,40,4,36                          // movaps        (%rsp),%xmm0
+  .byte  15,40,76,36,16                      // movaps        0x10(%rsp),%xmm1
+  .byte  15,40,84,36,32                      // movaps        0x20(%rsp),%xmm2
+  .byte  15,40,92,36,48                      // movaps        0x30(%rsp),%xmm3
+  .byte  15,40,100,36,64                     // movaps        0x40(%rsp),%xmm4
+  .byte  15,40,108,36,80                     // movaps        0x50(%rsp),%xmm5
+  .byte  15,40,116,36,96                     // movaps        0x60(%rsp),%xmm6
+  .byte  15,40,124,36,112                    // movaps        0x70(%rsp),%xmm7
+  .byte  72,129,196,128,0,0,0                // add           $0x80,%rsp
+  .byte  91                                  // pop           %rbx
+  .byte  65,94                               // pop           %r14
+  .byte  65,95                               // pop           %r15
+  .byte  255,224                             // jmpq          *%rax
 #endif
diff --git a/src/jumper/SkJumper_generated_win.S b/src/jumper/SkJumper_generated_win.S
index b305f23943..8762cbe61d 100644
--- a/src/jumper/SkJumper_generated_win.S
+++ b/src/jumper/SkJumper_generated_win.S
@@ -1357,7 +1357,7 @@ _sk_lerp_565_hsw LABEL PROC
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
-  DB  233,255,255,255,225                 ; jmpq          ffffffffe2001478 <_sk_bicubic_p3y_hsw+0xffffffffe1ffde19>
+  DB  233,255,255,255,225                 ; jmpq          ffffffffe2001478 <_sk_callback_hsw+0xffffffffe1ffddc2>
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
@@ -2328,7 +2328,7 @@ _sk_load_4444_hsw LABEL PROC
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
-  DB  233,255,255,255,225                 ; jmpq          ffffffffe2002334 <_sk_bicubic_p3y_hsw+0xffffffffe1ffecd5>
+  DB  233,255,255,255,225                 ; jmpq          ffffffffe2002334 <_sk_callback_hsw+0xffffffffe1ffec7e>
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
   DB  255                                 ; (bad)
@@ -3573,6 +3573,44 @@ _sk_bicubic_p3y_hsw LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_callback_hsw
+_sk_callback_hsw LABEL PROC
+  DB  65,86                               ; push          %r14
+  DB  83                                  ; push          %rbx
+  DB  72,129,236,40,1,0,0                 ; sub           $0x128,%rsp
+  DB  197,252,17,188,36,0,1,0,0           ; vmovups       %ymm7,0x100(%rsp)
+  DB  197,252,17,180,36,224,0,0,0         ; vmovups       %ymm6,0xe0(%rsp)
+  DB  197,252,17,172,36,192,0,0,0         ; vmovups       %ymm5,0xc0(%rsp)
+  DB  197,252,17,164,36,160,0,0,0         ; vmovups       %ymm4,0xa0(%rsp)
+  DB  197,252,17,156,36,128,0,0,0         ; vmovups       %ymm3,0x80(%rsp)
+  DB  197,252,17,84,36,96                 ; vmovups       %ymm2,0x60(%rsp)
+  DB  197,252,17,76,36,64                 ; vmovups       %ymm1,0x40(%rsp)
+  DB  197,252,17,68,36,32                 ; vmovups       %ymm0,0x20(%rsp)
+  DB  72,137,203                          ; mov           %rcx,%rbx
+  DB  73,137,214                          ; mov           %rdx,%r14
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,72,8                         ; mov           0x8(%rax),%rcx
+  DB  72,133,219                          ; test          %rbx,%rbx
+  DB  186,8,0,0,0                         ; mov           $0x8,%edx
+  DB  15,69,211                           ; cmovne        %ebx,%edx
+  DB  197,248,119                         ; vzeroupper
+  DB  255,16                              ; callq         *(%rax)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  76,137,242                          ; mov           %r14,%rdx
+  DB  72,137,217                          ; mov           %rbx,%rcx
+  DB  197,252,16,68,36,32                 ; vmovups       0x20(%rsp),%ymm0
+  DB  197,252,16,76,36,64                 ; vmovups       0x40(%rsp),%ymm1
+  DB  197,252,16,84,36,96                 ; vmovups       0x60(%rsp),%ymm2
+  DB  197,252,16,156,36,128,0,0,0         ; vmovups       0x80(%rsp),%ymm3
+  DB  197,252,16,164,36,160,0,0,0         ; vmovups       0xa0(%rsp),%ymm4
+  DB  197,252,16,172,36,192,0,0,0         ; vmovups       0xc0(%rsp),%ymm5
+  DB  197,252,16,180,36,224,0,0,0         ; vmovups       0xe0(%rsp),%ymm6
+  DB  197,252,16,188,36,0,1,0,0           ; vmovups       0x100(%rsp),%ymm7
+  DB  72,129,196,40,1,0,0                 ; add           $0x128,%rsp
+  DB  91                                  ; pop           %rbx
+  DB  65,94                               ; pop           %r14
+  DB  255,224                             ; jmpq          *%rax
+
 PUBLIC _sk_start_pipeline_avx
 _sk_start_pipeline_avx LABEL PROC
   DB  65,87                               ; push          %r15
@@ -7949,6 +7987,44 @@ _sk_bicubic_p3y_avx LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_callback_avx
+_sk_callback_avx LABEL PROC
+  DB  65,86                               ; push          %r14
+  DB  83                                  ; push          %rbx
+  DB  72,129,236,40,1,0,0                 ; sub           $0x128,%rsp
+  DB  197,252,17,188,36,0,1,0,0           ; vmovups       %ymm7,0x100(%rsp)
+  DB  197,252,17,180,36,224,0,0,0         ; vmovups       %ymm6,0xe0(%rsp)
+  DB  197,252,17,172,36,192,0,0,0         ; vmovups       %ymm5,0xc0(%rsp)
+  DB  197,252,17,164,36,160,0,0,0         ; vmovups       %ymm4,0xa0(%rsp)
+  DB  197,252,17,156,36,128,0,0,0         ; vmovups       %ymm3,0x80(%rsp)
+  DB  197,252,17,84,36,96                 ; vmovups       %ymm2,0x60(%rsp)
+  DB  197,252,17,76,36,64                 ; vmovups       %ymm1,0x40(%rsp)
+  DB  197,252,17,68,36,32                 ; vmovups       %ymm0,0x20(%rsp)
+  DB  72,137,203                          ; mov           %rcx,%rbx
+  DB  73,137,214                          ; mov           %rdx,%r14
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,72,8                         ; mov           0x8(%rax),%rcx
+  DB  72,133,219                          ; test          %rbx,%rbx
+  DB  186,8,0,0,0                         ; mov           $0x8,%edx
+  DB  15,69,211                           ; cmovne        %ebx,%edx
+  DB  197,248,119                         ; vzeroupper
+  DB  255,16                              ; callq         *(%rax)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  76,137,242                          ; mov           %r14,%rdx
+  DB  72,137,217                          ; mov           %rbx,%rcx
+  DB  197,252,16,68,36,32                 ; vmovups       0x20(%rsp),%ymm0
+  DB  197,252,16,76,36,64                 ; vmovups       0x40(%rsp),%ymm1
+  DB  197,252,16,84,36,96                 ; vmovups       0x60(%rsp),%ymm2
+  DB  197,252,16,156,36,128,0,0,0         ; vmovups       0x80(%rsp),%ymm3
+  DB  197,252,16,164,36,160,0,0,0         ; vmovups       0xa0(%rsp),%ymm4
+  DB  197,252,16,172,36,192,0,0,0         ; vmovups       0xc0(%rsp),%ymm5
+  DB  197,252,16,180,36,224,0,0,0         ; vmovups       0xe0(%rsp),%ymm6
+  DB  197,252,16,188,36,0,1,0,0           ; vmovups       0x100(%rsp),%ymm7
+  DB  72,129,196,40,1,0,0                 ; add           $0x128,%rsp
+  DB  91                                  ; pop           %rbx
+  DB  65,94                               ; pop           %r14
+  DB  255,224                             ; jmpq          *%rax
+
 PUBLIC _sk_start_pipeline_sse41
 _sk_start_pipeline_sse41 LABEL PROC
   DB  65,87                               ; push          %r15
@@ -11466,6 +11542,33 @@ _sk_bicubic_p3y_sse41 LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_callback_sse41
+_sk_callback_sse41 LABEL PROC
+  DB  83                                  ; push          %rbx
+  DB  72,131,236,32                       ; sub           $0x20,%rsp
+  DB  68,15,40,197                        ; movaps        %xmm5,%xmm8
+  DB  68,15,40,204                        ; movaps        %xmm4,%xmm9
+  DB  68,15,40,211                        ; movaps        %xmm3,%xmm10
+  DB  68,15,40,218                        ; movaps        %xmm2,%xmm11
+  DB  68,15,40,225                        ; movaps        %xmm1,%xmm12
+  DB  68,15,40,232                        ; movaps        %xmm0,%xmm13
+  DB  72,137,211                          ; mov           %rdx,%rbx
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,72,8                         ; mov           0x8(%rax),%rcx
+  DB  186,4,0,0,0                         ; mov           $0x4,%edx
+  DB  255,16                              ; callq         *(%rax)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,137,218                          ; mov           %rbx,%rdx
+  DB  65,15,40,197                        ; movaps        %xmm13,%xmm0
+  DB  65,15,40,204                        ; movaps        %xmm12,%xmm1
+  DB  65,15,40,211                        ; movaps        %xmm11,%xmm2
+  DB  65,15,40,218                        ; movaps        %xmm10,%xmm3
+  DB  65,15,40,225                        ; movaps        %xmm9,%xmm4
+  DB  65,15,40,232                        ; movaps        %xmm8,%xmm5
+  DB  72,131,196,32                       ; add           $0x20,%rsp
+  DB  91                                  ; pop           %rbx
+  DB  255,224                             ; jmpq          *%rax
+
 PUBLIC _sk_start_pipeline_sse2
 _sk_start_pipeline_sse2 LABEL PROC
   DB  65,87                               ; push          %r15
@@ -15226,5 +15329,32 @@ _sk_bicubic_p3y_sse2 LABEL PROC
   DB  68,15,17,128,160,0,0,0              ; movups        %xmm8,0xa0(%rax)
   DB  72,173                              ; lods          %ds:(%rsi),%rax
   DB  255,224                             ; jmpq          *%rax
+
+PUBLIC _sk_callback_sse2
+_sk_callback_sse2 LABEL PROC
+  DB  83                                  ; push          %rbx
+  DB  72,131,236,32                       ; sub           $0x20,%rsp
+  DB  68,15,40,197                        ; movaps        %xmm5,%xmm8
+  DB  68,15,40,204                        ; movaps        %xmm4,%xmm9
+  DB  68,15,40,211                        ; movaps        %xmm3,%xmm10
+  DB  68,15,40,218                        ; movaps        %xmm2,%xmm11
+  DB  68,15,40,225                        ; movaps        %xmm1,%xmm12
+  DB  68,15,40,232                        ; movaps        %xmm0,%xmm13
+  DB  72,137,211                          ; mov           %rdx,%rbx
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,72,8                         ; mov           0x8(%rax),%rcx
+  DB  186,4,0,0,0                         ; mov           $0x4,%edx
+  DB  255,16                              ; callq         *(%rax)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,137,218                          ; mov           %rbx,%rdx
+  DB  65,15,40,197                        ; movaps        %xmm13,%xmm0
+  DB  65,15,40,204                        ; movaps        %xmm12,%xmm1
+  DB  65,15,40,211                        ; movaps        %xmm11,%xmm2
+  DB  65,15,40,218                        ; movaps        %xmm10,%xmm3
+  DB  65,15,40,225                        ; movaps        %xmm9,%xmm4
+  DB  65,15,40,232                        ; movaps        %xmm8,%xmm5
+  DB  72,131,196,32                       ; add           $0x20,%rsp
+  DB  91                                  ; pop           %rbx
+  DB  255,224                             ; jmpq          *%rax
 ENDIF
 END
diff --git a/src/jumper/SkJumper_stages.cpp b/src/jumper/SkJumper_stages.cpp
index 2e6746c338..7c97235d2f 100644
--- a/src/jumper/SkJumper_stages.cpp
+++ b/src/jumper/SkJumper_stages.cpp
@@ -87,9 +87,7 @@ struct LazyCtx {
     // tail is always < kStride.
     using Stage = void(size_t x, void** program, K* k, size_t tail, F,F,F,F, F,F,F,F);
 
-    #if defined(JUMPER) && defined(WIN)
-    __attribute__((ms_abi))
-    #endif
+    MAYBE_MSABI
     extern "C" size_t WRAP(start_pipeline)(size_t x, void** program, K* k, size_t limit) {
         F v{};
         auto start = (Stage*)load_and_inc(program);
@@ -125,9 +123,7 @@ struct LazyCtx {
     using Stage = void(size_t x, void** program, K* k, F,F,F,F, F,F,F,F);
 
     // On Windows, start_pipeline() has a normal Windows ABI, and then the rest is System V.
-    #if defined(JUMPER) && defined(WIN)
-    __attribute__((ms_abi))
-    #endif
+    MAYBE_MSABI
     extern "C" size_t WRAP(start_pipeline)(size_t x, void** program, K* k, size_t limit) {
         F v{};
         auto start = (Stage*)load_and_inc(program);
@@ -1022,3 +1018,8 @@ STAGE(bicubic_n3y) { bicubic_y<-3>(ctx, &g); }
 STAGE(bicubic_n1y) { bicubic_y<-1>(ctx, &g); }
 STAGE(bicubic_p1y) { bicubic_y<+1>(ctx, &g); }
 STAGE(bicubic_p3y) { bicubic_y<+3>(ctx, &g); }
+
+STAGE(callback) {
+    auto c = (const SkJumper_CallbackCtx*)ctx;
+    c->fn(c->arg, tail ? tail : kStride);
+}
diff --git a/src/opts/SkRasterPipeline_opts.h b/src/opts/SkRasterPipeline_opts.h
index 1146b3d7fb..b15ebf36d8 100644
--- a/src/opts/SkRasterPipeline_opts.h
+++ b/src/opts/SkRasterPipeline_opts.h
@@ -1098,6 +1098,11 @@ STAGE_CTX(shader_adapter, SkShader::Context*) {
     SkNf::Load4(buf, &r, &g, &b, &a);
 }
 
+STAGE_CTX(callback, const void*) {
+    auto c = (const SkJumper_CallbackCtx*)ctx;
+    c->fn(c->arg, tail ? tail : N);
+}
+
 SI Fn enum_to_Fn(SkRasterPipeline::StockStage st) {
     switch (st) {
     #define M(stage) case SkRasterPipeline::stage: return stage;