SkJumper: a8

Change-Id: I123caaee0bb8e3967c0a1f2acf1d80bcf0f41758
Reviewed-on: https://skia-review.googlesource.com/8944
Reviewed-by: Mike Klein <mtklein@chromium.org>
Commit-Queue: Mike Klein <mtklein@chromium.org>

4 files changed, 361 insertions, 1 deletions

diff --git a/src/jumper/SkJumper.cpp b/src/jumper/SkJumper.cpp
index 90db9a94a8..d38090d180 100644
--- a/src/jumper/SkJumper.cpp
+++ b/src/jumper/SkJumper.cpp
@@ -70,6 +70,8 @@ static K kConstants = {
     M(lerp_u8)        \
     M(lerp_565)       \
     M(load_tables)    \
+    M(load_a8)        \
+    M(store_a8)       \
     M(load_565)       \
     M(store_565)      \
     M(load_8888)      \
diff --git a/src/jumper/SkJumper_generated.S b/src/jumper/SkJumper_generated.S
index 8b4106c050..973534a65b 100644
--- a/src/jumper/SkJumper_generated.S
+++ b/src/jumper/SkJumper_generated.S
@@ -526,6 +526,50 @@ _sk_load_tables_aarch64:
   .long  0x6e1c0602                          // mov           v2.s[3], v16.s[0]
   .long  0xd61f0060                          // br            x3
 
+.globl _sk_load_a8_aarch64
+_sk_load_a8_aarch64:
+  .long  0xa8c10c28                          // ldp           x8, x3, [x1],#16
+  .long  0xbd400c43                          // ldr           s3, [x2,#12]
+  .long  0x6f00e400                          // movi          v0.2d, #0x0
+  .long  0x6f00e401                          // movi          v1.2d, #0x0
+  .long  0xf9400108                          // ldr           x8, [x8]
+  .long  0x6f00e402                          // movi          v2.2d, #0x0
+  .long  0x8b000108                          // add           x8, x8, x0
+  .long  0x39400109                          // ldrb          w9, [x8]
+  .long  0x3940050a                          // ldrb          w10, [x8,#1]
+  .long  0x3940090b                          // ldrb          w11, [x8,#2]
+  .long  0x39400d08                          // ldrb          w8, [x8,#3]
+  .long  0x4e021d30                          // mov           v16.h[0], w9
+  .long  0x4e061d50                          // mov           v16.h[1], w10
+  .long  0x4e0a1d70                          // mov           v16.h[2], w11
+  .long  0x4e0e1d10                          // mov           v16.h[3], w8
+  .long  0x2f07b7f0                          // bic           v16.4h, #0xff, lsl #8
+  .long  0x2f10a610                          // uxtl          v16.4s, v16.4h
+  .long  0x6e21da10                          // ucvtf         v16.4s, v16.4s
+  .long  0x4f839203                          // fmul          v3.4s, v16.4s, v3.s[0]
+  .long  0xd61f0060                          // br            x3
+
+.globl _sk_store_a8_aarch64
+_sk_store_a8_aarch64:
+  .long  0xf9400028                          // ldr           x8, [x1]
+  .long  0xbd400850                          // ldr           s16, [x2,#8]
+  .long  0xf9400108                          // ldr           x8, [x8]
+  .long  0x4f909070                          // fmul          v16.4s, v3.4s, v16.s[0]
+  .long  0x6e21aa10                          // fcvtnu        v16.4s, v16.4s
+  .long  0x0e612a10                          // xtn           v16.4h, v16.4s
+  .long  0x0e0e3e09                          // umov          w9, v16.h[3]
+  .long  0x8b000108                          // add           x8, x8, x0
+  .long  0x39000d09                          // strb          w9, [x8,#3]
+  .long  0x0e0a3e09                          // umov          w9, v16.h[2]
+  .long  0x39000909                          // strb          w9, [x8,#2]
+  .long  0x0e063e09                          // umov          w9, v16.h[1]
+  .long  0x39000509                          // strb          w9, [x8,#1]
+  .long  0x0e023e09                          // umov          w9, v16.h[0]
+  .long  0x39000109                          // strb          w9, [x8]
+  .long  0xf9400423                          // ldr           x3, [x1,#8]
+  .long  0x91004021                          // add           x1, x1, #0x10
+  .long  0xd61f0060                          // br            x3
+
 .globl _sk_load_565_aarch64
 _sk_load_565_aarch64:
   .long  0xa8c10c28                          // ldp           x8, x3, [x1],#16
@@ -1273,6 +1317,48 @@ _sk_load_tables_vfp4:
   .long  0xe8bd48f0                          // pop           {r4, r5, r6, r7, fp, lr}
   .long  0xe12fff1c                          // bx            ip
 
+.globl _sk_load_a8_vfp4
+_sk_load_a8_vfp4:
+  .long  0xe24dd004                          // sub           sp, sp, #4
+  .long  0xe5913000                          // ldr           r3, [r1]
+  .long  0xf2801010                          // vmov.i32      d1, #0
+  .long  0xe591c004                          // ldr           ip, [r1, #4]
+  .long  0xe2811008                          // add           r1, r1, #8
+  .long  0xf2802010                          // vmov.i32      d2, #0
+  .long  0xe5933000                          // ldr           r3, [r3]
+  .long  0xe0833000                          // add           r3, r3, r0
+  .long  0xe1d330b0                          // ldrh          r3, [r3]
+  .long  0xe1cd30b0                          // strh          r3, [sp]
+  .long  0xe1a0300d                          // mov           r3, sp
+  .long  0xf4e3041f                          // vld1.16       {d16[0]}, [r3 :16]
+  .long  0xed920a03                          // vldr          s0, [r2, #12]
+  .long  0xf3c80a30                          // vmovl.u8      q8, d16
+  .long  0xf3d00a30                          // vmovl.u16     q8, d16
+  .long  0xf3fb06a0                          // vcvt.f32.u32  d16, d16
+  .long  0xf2a039c0                          // vmul.f32      d3, d16, d0[0]
+  .long  0xf2800010                          // vmov.i32      d0, #0
+  .long  0xe28dd004                          // add           sp, sp, #4
+  .long  0xe12fff1c                          // bx            ip
+
+.globl _sk_store_a8_vfp4
+_sk_store_a8_vfp4:
+  .long  0xe92d4800                          // push          {fp, lr}
+  .long  0xe2823008                          // add           r3, r2, #8
+  .long  0xf2c3061f                          // vmov.i32      d16, #1056964608
+  .long  0xf4e31c9f                          // vld1.32       {d17[]}, [r3 :32]
+  .long  0xe5913000                          // ldr           r3, [r1]
+  .long  0xf2430c31                          // vfma.f32      d16, d3, d17
+  .long  0xe5933000                          // ldr           r3, [r3]
+  .long  0xf3fb07a0                          // vcvt.u32.f32  d16, d16
+  .long  0xee10eb90                          // vmov.32       lr, d16[0]
+  .long  0xee30cb90                          // vmov.32       ip, d16[1]
+  .long  0xe7e3e000                          // strb          lr, [r3, r0]!
+  .long  0xe5c3c001                          // strb          ip, [r3, #1]
+  .long  0xe5913004                          // ldr           r3, [r1, #4]
+  .long  0xe2811008                          // add           r1, r1, #8
+  .long  0xe8bd4800                          // pop           {fp, lr}
+  .long  0xe12fff13                          // bx            r3
+
 .globl _sk_load_565_vfp4
 _sk_load_565_vfp4:
   .long  0xe24dd004                          // sub           sp, sp, #4
@@ -1976,6 +2062,34 @@ _sk_load_tables_hsw:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
   .byte  255,224                             // jmpq          *%rax
 
+.globl _sk_load_a8_hsw
+_sk_load_a8_hsw:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  196,226,125,49,4,56                 // vpmovzxbd     (%rax,%rdi,1),%ymm0
+  .byte  197,252,91,192                      // vcvtdq2ps     %ymm0,%ymm0
+  .byte  196,226,125,24,74,12                // vbroadcastss  0xc(%rdx),%ymm1
+  .byte  197,252,89,217                      // vmulps        %ymm1,%ymm0,%ymm3
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  197,252,87,192                      // vxorps        %ymm0,%ymm0,%ymm0
+  .byte  197,244,87,201                      // vxorps        %ymm1,%ymm1,%ymm1
+  .byte  197,236,87,210                      // vxorps        %ymm2,%ymm2,%ymm2
+  .byte  255,224                             // jmpq          *%rax
+
+.globl _sk_store_a8_hsw
+_sk_store_a8_hsw:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  196,98,125,24,66,8                  // vbroadcastss  0x8(%rdx),%ymm8
+  .byte  197,60,89,195                       // vmulps        %ymm3,%ymm8,%ymm8
+  .byte  196,65,125,91,192                   // vcvtps2dq     %ymm8,%ymm8
+  .byte  196,67,125,25,193,1                 // vextractf128  $0x1,%ymm8,%xmm9
+  .byte  196,66,57,43,193                    // vpackusdw     %xmm9,%xmm8,%xmm8
+  .byte  196,65,57,103,192                   // vpackuswb     %xmm8,%xmm8,%xmm8
+  .byte  197,121,214,4,56                    // vmovq         %xmm8,(%rax,%rdi,1)
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  255,224                             // jmpq          *%rax
+
 .globl _sk_load_565_hsw
 _sk_load_565_hsw:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
@@ -2760,6 +2874,36 @@ _sk_load_tables_avx:
   .byte  65,95                               // pop           %r15
   .byte  255,224                             // jmpq          *%rax
 
+.globl _sk_load_a8_avx
+_sk_load_a8_avx:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  196,226,121,49,68,56,4              // vpmovzxbd     0x4(%rax,%rdi,1),%xmm0
+  .byte  196,226,121,49,12,56                // vpmovzxbd     (%rax,%rdi,1),%xmm1
+  .byte  196,227,117,24,192,1                // vinsertf128   $0x1,%xmm0,%ymm1,%ymm0
+  .byte  197,252,91,192                      // vcvtdq2ps     %ymm0,%ymm0
+  .byte  196,226,125,24,74,12                // vbroadcastss  0xc(%rdx),%ymm1
+  .byte  197,252,89,217                      // vmulps        %ymm1,%ymm0,%ymm3
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  197,252,87,192                      // vxorps        %ymm0,%ymm0,%ymm0
+  .byte  197,244,87,201                      // vxorps        %ymm1,%ymm1,%ymm1
+  .byte  197,236,87,210                      // vxorps        %ymm2,%ymm2,%ymm2
+  .byte  255,224                             // jmpq          *%rax
+
+.globl _sk_store_a8_avx
+_sk_store_a8_avx:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  196,98,125,24,66,8                  // vbroadcastss  0x8(%rdx),%ymm8
+  .byte  197,60,89,195                       // vmulps        %ymm3,%ymm8,%ymm8
+  .byte  196,65,125,91,192                   // vcvtps2dq     %ymm8,%ymm8
+  .byte  196,67,125,25,193,1                 // vextractf128  $0x1,%ymm8,%xmm9
+  .byte  196,66,57,43,193                    // vpackusdw     %xmm9,%xmm8,%xmm8
+  .byte  196,65,57,103,192                   // vpackuswb     %xmm8,%xmm8,%xmm8
+  .byte  197,121,214,4,56                    // vmovq         %xmm8,(%rax,%rdi,1)
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  255,224                             // jmpq          *%rax
+
 .globl _sk_load_565_avx
 _sk_load_565_avx:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
@@ -3655,6 +3799,35 @@ _sk_load_tables_sse41:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
   .byte  255,224                             // jmpq          *%rax
 
+.globl _sk_load_a8_sse41
+_sk_load_a8_sse41:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  102,15,56,49,4,56                   // pmovzxbd      (%rax,%rdi,1),%xmm0
+  .byte  15,91,192                           // cvtdq2ps      %xmm0,%xmm0
+  .byte  243,15,16,90,12                     // movss         0xc(%rdx),%xmm3
+  .byte  15,198,219,0                        // shufps        $0x0,%xmm3,%xmm3
+  .byte  15,89,216                           // mulps         %xmm0,%xmm3
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  15,87,192                           // xorps         %xmm0,%xmm0
+  .byte  15,87,201                           // xorps         %xmm1,%xmm1
+  .byte  15,87,210                           // xorps         %xmm2,%xmm2
+  .byte  255,224                             // jmpq          *%rax
+
+.globl _sk_store_a8_sse41
+_sk_store_a8_sse41:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  243,68,15,16,66,8                   // movss         0x8(%rdx),%xmm8
+  .byte  69,15,198,192,0                     // shufps        $0x0,%xmm8,%xmm8
+  .byte  68,15,89,195                        // mulps         %xmm3,%xmm8
+  .byte  102,69,15,91,192                    // cvtps2dq      %xmm8,%xmm8
+  .byte  102,69,15,56,43,192                 // packusdw      %xmm8,%xmm8
+  .byte  102,69,15,103,192                   // packuswb      %xmm8,%xmm8
+  .byte  102,68,15,126,4,56                  // movd          %xmm8,(%rax,%rdi,1)
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  255,224                             // jmpq          *%rax
+
 .globl _sk_load_565_sse41
 _sk_load_565_sse41:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
@@ -4542,6 +4715,40 @@ _sk_load_tables_sse2:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
   .byte  255,224                             // jmpq          *%rax
 
+.globl _sk_load_a8_sse2
+_sk_load_a8_sse2:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  102,15,110,4,56                     // movd          (%rax,%rdi,1),%xmm0
+  .byte  102,15,239,201                      // pxor          %xmm1,%xmm1
+  .byte  102,15,96,193                       // punpcklbw     %xmm1,%xmm0
+  .byte  102,15,97,193                       // punpcklwd     %xmm1,%xmm0
+  .byte  15,91,192                           // cvtdq2ps      %xmm0,%xmm0
+  .byte  243,15,16,90,12                     // movss         0xc(%rdx),%xmm3
+  .byte  15,198,219,0                        // shufps        $0x0,%xmm3,%xmm3
+  .byte  15,89,216                           // mulps         %xmm0,%xmm3
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  15,87,192                           // xorps         %xmm0,%xmm0
+  .byte  102,15,239,201                      // pxor          %xmm1,%xmm1
+  .byte  15,87,210                           // xorps         %xmm2,%xmm2
+  .byte  255,224                             // jmpq          *%rax
+
+.globl _sk_store_a8_sse2
+_sk_store_a8_sse2:
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  72,139,0                            // mov           (%rax),%rax
+  .byte  243,68,15,16,66,8                   // movss         0x8(%rdx),%xmm8
+  .byte  69,15,198,192,0                     // shufps        $0x0,%xmm8,%xmm8
+  .byte  68,15,89,195                        // mulps         %xmm3,%xmm8
+  .byte  102,69,15,91,192                    // cvtps2dq      %xmm8,%xmm8
+  .byte  102,65,15,114,240,16                // pslld         $0x10,%xmm8
+  .byte  102,65,15,114,224,16                // psrad         $0x10,%xmm8
+  .byte  102,69,15,107,192                   // packssdw      %xmm8,%xmm8
+  .byte  102,69,15,103,192                   // packuswb      %xmm8,%xmm8
+  .byte  102,68,15,126,4,56                  // movd          %xmm8,(%rax,%rdi,1)
+  .byte  72,173                              // lods          %ds:(%rsi),%rax
+  .byte  255,224                             // jmpq          *%rax
+
 .globl _sk_load_565_sse2
 _sk_load_565_sse2:
   .byte  72,173                              // lods          %ds:(%rsi),%rax
diff --git a/src/jumper/SkJumper_generated_win.S b/src/jumper/SkJumper_generated_win.S
index 8d80694771..e22e4cdfe3 100644
--- a/src/jumper/SkJumper_generated_win.S
+++ b/src/jumper/SkJumper_generated_win.S
@@ -457,6 +457,34 @@ _sk_load_tables_hsw LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_load_a8_hsw
+_sk_load_a8_hsw LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  196,226,125,49,4,56                 ; vpmovzxbd     (%rax,%rdi,1),%ymm0
+  DB  197,252,91,192                      ; vcvtdq2ps     %ymm0,%ymm0
+  DB  196,226,125,24,74,12                ; vbroadcastss  0xc(%rdx),%ymm1
+  DB  197,252,89,217                      ; vmulps        %ymm1,%ymm0,%ymm3
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  197,252,87,192                      ; vxorps        %ymm0,%ymm0,%ymm0
+  DB  197,244,87,201                      ; vxorps        %ymm1,%ymm1,%ymm1
+  DB  197,236,87,210                      ; vxorps        %ymm2,%ymm2,%ymm2
+  DB  255,224                             ; jmpq          *%rax
+
+PUBLIC _sk_store_a8_hsw
+_sk_store_a8_hsw LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  196,98,125,24,66,8                  ; vbroadcastss  0x8(%rdx),%ymm8
+  DB  197,60,89,195                       ; vmulps        %ymm3,%ymm8,%ymm8
+  DB  196,65,125,91,192                   ; vcvtps2dq     %ymm8,%ymm8
+  DB  196,67,125,25,193,1                 ; vextractf128  $0x1,%ymm8,%xmm9
+  DB  196,66,57,43,193                    ; vpackusdw     %xmm9,%xmm8,%xmm8
+  DB  196,65,57,103,192                   ; vpackuswb     %xmm8,%xmm8,%xmm8
+  DB  197,121,214,4,56                    ; vmovq         %xmm8,(%rax,%rdi,1)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  255,224                             ; jmpq          *%rax
+
 PUBLIC _sk_load_565_hsw
 _sk_load_565_hsw LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
@@ -1268,6 +1296,36 @@ _sk_load_tables_avx LABEL PROC
   DB  65,95                               ; pop           %r15
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_load_a8_avx
+_sk_load_a8_avx LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  196,226,121,49,68,56,4              ; vpmovzxbd     0x4(%rax,%rdi,1),%xmm0
+  DB  196,226,121,49,12,56                ; vpmovzxbd     (%rax,%rdi,1),%xmm1
+  DB  196,227,117,24,192,1                ; vinsertf128   $0x1,%xmm0,%ymm1,%ymm0
+  DB  197,252,91,192                      ; vcvtdq2ps     %ymm0,%ymm0
+  DB  196,226,125,24,74,12                ; vbroadcastss  0xc(%rdx),%ymm1
+  DB  197,252,89,217                      ; vmulps        %ymm1,%ymm0,%ymm3
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  197,252,87,192                      ; vxorps        %ymm0,%ymm0,%ymm0
+  DB  197,244,87,201                      ; vxorps        %ymm1,%ymm1,%ymm1
+  DB  197,236,87,210                      ; vxorps        %ymm2,%ymm2,%ymm2
+  DB  255,224                             ; jmpq          *%rax
+
+PUBLIC _sk_store_a8_avx
+_sk_store_a8_avx LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  196,98,125,24,66,8                  ; vbroadcastss  0x8(%rdx),%ymm8
+  DB  197,60,89,195                       ; vmulps        %ymm3,%ymm8,%ymm8
+  DB  196,65,125,91,192                   ; vcvtps2dq     %ymm8,%ymm8
+  DB  196,67,125,25,193,1                 ; vextractf128  $0x1,%ymm8,%xmm9
+  DB  196,66,57,43,193                    ; vpackusdw     %xmm9,%xmm8,%xmm8
+  DB  196,65,57,103,192                   ; vpackuswb     %xmm8,%xmm8,%xmm8
+  DB  197,121,214,4,56                    ; vmovq         %xmm8,(%rax,%rdi,1)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  255,224                             ; jmpq          *%rax
+
 PUBLIC _sk_load_565_avx
 _sk_load_565_avx LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
@@ -2190,6 +2248,35 @@ _sk_load_tables_sse41 LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_load_a8_sse41
+_sk_load_a8_sse41 LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  102,15,56,49,4,56                   ; pmovzxbd      (%rax,%rdi,1),%xmm0
+  DB  15,91,192                           ; cvtdq2ps      %xmm0,%xmm0
+  DB  243,15,16,90,12                     ; movss         0xc(%rdx),%xmm3
+  DB  15,198,219,0                        ; shufps        $0x0,%xmm3,%xmm3
+  DB  15,89,216                           ; mulps         %xmm0,%xmm3
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  15,87,192                           ; xorps         %xmm0,%xmm0
+  DB  15,87,201                           ; xorps         %xmm1,%xmm1
+  DB  15,87,210                           ; xorps         %xmm2,%xmm2
+  DB  255,224                             ; jmpq          *%rax
+
+PUBLIC _sk_store_a8_sse41
+_sk_store_a8_sse41 LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  243,68,15,16,66,8                   ; movss         0x8(%rdx),%xmm8
+  DB  69,15,198,192,0                     ; shufps        $0x0,%xmm8,%xmm8
+  DB  68,15,89,195                        ; mulps         %xmm3,%xmm8
+  DB  102,69,15,91,192                    ; cvtps2dq      %xmm8,%xmm8
+  DB  102,69,15,56,43,192                 ; packusdw      %xmm8,%xmm8
+  DB  102,69,15,103,192                   ; packuswb      %xmm8,%xmm8
+  DB  102,68,15,126,4,56                  ; movd          %xmm8,(%rax,%rdi,1)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  255,224                             ; jmpq          *%rax
+
 PUBLIC _sk_load_565_sse41
 _sk_load_565_sse41 LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
@@ -3104,6 +3191,40 @@ _sk_load_tables_sse2 LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
   DB  255,224                             ; jmpq          *%rax
 
+PUBLIC _sk_load_a8_sse2
+_sk_load_a8_sse2 LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  102,15,110,4,56                     ; movd          (%rax,%rdi,1),%xmm0
+  DB  102,15,239,201                      ; pxor          %xmm1,%xmm1
+  DB  102,15,96,193                       ; punpcklbw     %xmm1,%xmm0
+  DB  102,15,97,193                       ; punpcklwd     %xmm1,%xmm0
+  DB  15,91,192                           ; cvtdq2ps      %xmm0,%xmm0
+  DB  243,15,16,90,12                     ; movss         0xc(%rdx),%xmm3
+  DB  15,198,219,0                        ; shufps        $0x0,%xmm3,%xmm3
+  DB  15,89,216                           ; mulps         %xmm0,%xmm3
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  15,87,192                           ; xorps         %xmm0,%xmm0
+  DB  102,15,239,201                      ; pxor          %xmm1,%xmm1
+  DB  15,87,210                           ; xorps         %xmm2,%xmm2
+  DB  255,224                             ; jmpq          *%rax
+
+PUBLIC _sk_store_a8_sse2
+_sk_store_a8_sse2 LABEL PROC
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  72,139,0                            ; mov           (%rax),%rax
+  DB  243,68,15,16,66,8                   ; movss         0x8(%rdx),%xmm8
+  DB  69,15,198,192,0                     ; shufps        $0x0,%xmm8,%xmm8
+  DB  68,15,89,195                        ; mulps         %xmm3,%xmm8
+  DB  102,69,15,91,192                    ; cvtps2dq      %xmm8,%xmm8
+  DB  102,65,15,114,240,16                ; pslld         $0x10,%xmm8
+  DB  102,65,15,114,224,16                ; psrad         $0x10,%xmm8
+  DB  102,69,15,107,192                   ; packssdw      %xmm8,%xmm8
+  DB  102,69,15,103,192                   ; packuswb      %xmm8,%xmm8
+  DB  102,68,15,126,4,56                  ; movd          %xmm8,(%rax,%rdi,1)
+  DB  72,173                              ; lods          %ds:(%rsi),%rax
+  DB  255,224                             ; jmpq          *%rax
+
 PUBLIC _sk_load_565_sse2
 _sk_load_565_sse2 LABEL PROC
   DB  72,173                              ; lods          %ds:(%rsi),%rax
diff --git a/src/jumper/SkJumper_stages.cpp b/src/jumper/SkJumper_stages.cpp
index 580432c60d..899323b0db 100644
--- a/src/jumper/SkJumper_stages.cpp
+++ b/src/jumper/SkJumper_stages.cpp
@@ -42,6 +42,7 @@ static Dst bit_cast(const Src& src) {
     static F   rsqrt(F v)          { return 1.0f / sqrtf(v); }
     static U32 round(F v, F scale) { return (uint32_t)lrintf(v*scale); }
     static U16 pack(U32 v)         { return (U16)v; }
+    static U8  pack(U16 v)         { return  (U8)v; }
 
     static F if_then_else(I32 c, F t, F e) { return c ? t : e; }
 
@@ -67,6 +68,7 @@ static Dst bit_cast(const Src& src) {
     static F   rsqrt(F v) { auto e = vrsqrteq_f32(v); return vrsqrtsq_f32(v,e*e) * e; }
     static U32 round(F v, F scale)                  { return vcvtnq_u32_f32(v*scale); }
     static U16 pack(U32 v)                          { return __builtin_convertvector(v, U16); }
+    static U8  pack(U16 v)                          { return __builtin_convertvector(v,  U8); }
 
     static F if_then_else(I32 c, F t, F e) { return vbslq_f32((U32)c,t,e); }
 
@@ -94,6 +96,7 @@ static Dst bit_cast(const Src& src) {
     static F   rsqrt(F v) { auto e = vrsqrte_f32(v); return vrsqrts_f32(v,e*e) * e; }
     static U32 round(F v, F scale)                 { return vcvt_u32_f32(mad(v,scale,0.5f)); }
     static U16 pack(U32 v)                         { return __builtin_convertvector(v, U16); }
+    static U8  pack(U16 v)                         { return __builtin_convertvector(v,  U8); }
 
     static F if_then_else(I32 c, F t, F e) { return vbsl_f32((U32)c,t,e); }
 
@@ -123,6 +126,10 @@ static Dst bit_cast(const Src& src) {
                 hi = _mm256_extractf128_si256(v, 1);
         return _mm_packus_epi32(lo, hi);
     }
+    static U8 pack(U16 v) {
+        __m128i r = _mm_packus_epi16(v,v);
+        return unaligned_load<U8>(&r);
+    }
 
     static F if_then_else(I32 c, F t, F e) { return _mm256_blendv_ps(e,t,c); }
 
@@ -151,6 +158,10 @@ static Dst bit_cast(const Src& src) {
                 hi = _mm256_extractf128_si256(v, 1);
         return _mm_packus_epi32(lo, hi);
     }
+    static U8 pack(U16 v) {
+        __m128i r = _mm_packus_epi16(v,v);
+        return unaligned_load<U8>(&r);
+    }
 
     static F if_then_else(I32 c, F t, F e) { return _mm256_blendv_ps(e,t,c); }
 
@@ -187,6 +198,12 @@ static Dst bit_cast(const Src& src) {
     #endif
         return unaligned_load<U16>(&p);  // We have two copies.  Return (the lower) one.
     }
+    static U8 pack(U16 v) {
+        __m128i r;
+        memcpy(&r, &v, sizeof(v));
+        r = _mm_packus_epi16(r,r);
+        return unaligned_load<U8>(&r);
+    }
 
     static F if_then_else(I32 c, F t, F e) {
     #if defined(__SSE4_1__)
@@ -213,7 +230,7 @@ static Dst bit_cast(const Src& src) {
     static U32 expand(U16 v) { return __builtin_convertvector(     v, U32); }
     static U32 expand(U8  v) { return __builtin_convertvector(     v, U32); }
 #else
-    static F   cast  (U32 v) { return (F)v; }
+    static F   cast  (U32 v) { return   (F)v; }
     static U32 expand(U16 v) { return (U32)v; }
     static U32 expand(U8  v) { return (U32)v; }
 #endif
@@ -525,6 +542,19 @@ STAGE(load_tables) {
     a = cast(        (px >> 24)) * k->_1_255;
 }
 
+STAGE(load_a8) {
+    auto ptr = *(const uint8_t**)ctx + x;
+
+    r = g = b = 0.0f;
+    a = cast(expand(unaligned_load<U8>(ptr))) * k->_1_255;
+}
+STAGE(store_a8) {
+    auto ptr = *(uint8_t**)ctx + x;
+
+    U8 packed = pack(pack(round(a, k->_255)));
+    memcpy(ptr, &packed, sizeof(packed));
+}
+
 STAGE(load_565) {
     auto ptr = *(const uint16_t**)ctx + x;