AuroraMiddleware/skia2
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

SkJumper: 565

Browse Source 
Change-Id: Icbd41e3dde9b39a61ccbe8e7622334ae53e5212a
Reviewed-on: https://skia-review.googlesource.com/8922
Reviewed-by: Mike Klein <mtklein@chromium.org>
Commit-Queue: Mike Klein <mtklein@chromium.org>
This commit is contained in:
Mike Klein 2017-02-23 13:03:57 -05:00 committed by Skia Commit-Bot
parent e992d41e91
commit 3f81f3703a
5 changed files with 637 additions and 13 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
src/jumper/SkJumper.cpp
View File

@ -40,6 +40,9 @@ static K kConstants = {
0.0025f, 0.6975f, 0.3000f, 1/12.92f, 0.055f, // from_srgb
12.46f, 0.411192f, 0.689206f, -0.0988f, 0.0043f, // to_srgb
0x77800000, 0x07800000, 0x04000400, // fp16 <-> fp32
0x0000f800, 0x000007e0, 0x0000001f, // 565
1.0f/0x0000f800, 1.0f/0x000007e0, 1.0f/0x0000001f,
31.0f, 63.0f,
};
#define STAGES(M) \
@ -64,6 +67,8 @@ static K kConstants = {
M(scale_u8) \
M(lerp_u8) \
M(load_tables) \
M(load_565) \
M(store_565) \
M(load_8888) \
M(store_8888) \
M(load_f16) \

10
src/jumper/SkJumper.h
View File

@ -47,6 +47,16 @@ struct SkJumper_constants {
uint32_t _0x77800000;
uint32_t _0x07800000;
uint32_t _0x04000400;
// 565
uint32_t r_565_mask;
uint32_t g_565_mask;
uint32_t b_565_mask;
float r_565_scale;
float g_565_scale;
float b_565_scale;
float _31;
float _63;
};
#endif//SkJumper_DEFINED

331
src/jumper/SkJumper_generated.S
View File

@ -462,6 +462,55 @@ _sk_load_tables_aarch64:
.long 0x6e1c0602 // mov v2.s[3], v16.s[0]
.long 0xd61f0060 // br x3
.globl _sk_load_565_aarch64
_sk_load_565_aarch64:
.long 0xa8c10c28 // ldp x8, x3, [x1],#16
.long 0xd37ff809 // lsl x9, x0, #1
.long 0xf9400108 // ldr x8, [x8]
.long 0xfc696900 // ldr d0, [x8,x9]
.long 0x9101a048 // add x8, x2, #0x68
.long 0x4d40c901 // ld1r {v1.4s}, [x8]
.long 0x9101b048 // add x8, x2, #0x6c
.long 0x4d40c902 // ld1r {v2.4s}, [x8]
.long 0x9101c048 // add x8, x2, #0x70
.long 0x4d40c903 // ld1r {v3.4s}, [x8]
.long 0x2f10a400 // uxtl v0.4s, v0.4h
.long 0x4e201c21 // and v1.16b, v1.16b, v0.16b
.long 0x4e201c42 // and v2.16b, v2.16b, v0.16b
.long 0x4e201c71 // and v17.16b, v3.16b, v0.16b
.long 0x2d4e8c50 // ldp s16, s3, [x2,#116]
.long 0x4e21d820 // scvtf v0.4s, v1.4s
.long 0x4e21d841 // scvtf v1.4s, v2.4s
.long 0x4e21da22 // scvtf v2.4s, v17.4s
.long 0x4f909000 // fmul v0.4s, v0.4s, v16.s[0]
.long 0xbd407c50 // ldr s16, [x2,#124]
.long 0x4f839021 // fmul v1.4s, v1.4s, v3.s[0]
.long 0x4d40c843 // ld1r {v3.4s}, [x2]
.long 0x4f909042 // fmul v2.4s, v2.4s, v16.s[0]
.long 0xd61f0060 // br x3
.globl _sk_store_565_aarch64
_sk_store_565_aarch64:
.long 0xf9400028 // ldr x8, [x1]
.long 0x2d504450 // ldp s16, s17, [x2,#128]
.long 0xd37ff809 // lsl x9, x0, #1
.long 0xf9400108 // ldr x8, [x8]
.long 0x4f909012 // fmul v18.4s, v0.4s, v16.s[0]
.long 0x4f919031 // fmul v17.4s, v1.4s, v17.s[0]
.long 0x6e21aa52 // fcvtnu v18.4s, v18.4s
.long 0x6e21aa31 // fcvtnu v17.4s, v17.4s
.long 0x4f909050 // fmul v16.4s, v2.4s, v16.s[0]
.long 0x4f2b5652 // shl v18.4s, v18.4s, #11
.long 0x4f255631 // shl v17.4s, v17.4s, #5
.long 0x4eb21e31 // orr v17.16b, v17.16b, v18.16b
.long 0x6e21aa10 // fcvtnu v16.4s, v16.4s
.long 0x4eb01e30 // orr v16.16b, v17.16b, v16.16b
.long 0x0e612a10 // xtn v16.4h, v16.4s
.long 0xfc296910 // str d16, [x8,x9]
.long 0xf9400423 // ldr x3, [x1,#8]
.long 0x91004021 // add x1, x1, #0x10
.long 0xd61f0060 // br x3
.globl _sk_load_8888_aarch64
_sk_load_8888_aarch64:
.long 0xa8c10c28 // ldp x8, x3, [x1],#16
@ -1081,6 +1130,68 @@ _sk_load_tables_vfp4:
.long 0xe8bd48f0 // pop {r4, r5, r6, r7, fp, lr}
.long 0xe12fff1c // bx ip
.globl _sk_load_565_vfp4
_sk_load_565_vfp4:
.long 0xe24dd004 // sub sp, sp, #4
.long 0xe5913000 // ldr r3, [r1]
.long 0xe591c004 // ldr ip, [r1, #4]
.long 0xe2811008 // add r1, r1, #8
.long 0xe5933000 // ldr r3, [r3]
.long 0xe7933080 // ldr r3, [r3, r0, lsl #1]
.long 0xe58d3000 // str r3, [sp]
.long 0xe1a0300d // mov r3, sp
.long 0xf4e3083f // vld1.32 {d16[0]}, [r3 :32]
.long 0xe282306c // add r3, r2, #108
.long 0xf4e31c9f // vld1.32 {d17[]}, [r3 :32]
.long 0xe2823068 // add r3, r2, #104
.long 0xf3d04a30 // vmovl.u16 q10, d16
.long 0xf4e32c9f // vld1.32 {d18[]}, [r3 :32]
.long 0xe2823070 // add r3, r2, #112
.long 0xf24201b4 // vand d16, d18, d20
.long 0xf4e32c9f // vld1.32 {d18[]}, [r3 :32]
.long 0xf24111b4 // vand d17, d17, d20
.long 0xf24221b4 // vand d18, d18, d20
.long 0xf4a23c9f // vld1.32 {d3[]}, [r2 :32]
.long 0xf3fb0620 // vcvt.f32.s32 d16, d16
.long 0xf3fb1621 // vcvt.f32.s32 d17, d17
.long 0xf3fb2622 // vcvt.f32.s32 d18, d18
.long 0xed920a1d // vldr s0, [r2, #116]
.long 0xed921a1e // vldr s2, [r2, #120]
.long 0xed922a1f // vldr s4, [r2, #124]
.long 0xf2a009c0 // vmul.f32 d0, d16, d0[0]
.long 0xf2a119c1 // vmul.f32 d1, d17, d1[0]
.long 0xf2a229c2 // vmul.f32 d2, d18, d2[0]
.long 0xe28dd004 // add sp, sp, #4
.long 0xe12fff1c // bx ip
.globl _sk_store_565_vfp4
_sk_store_565_vfp4:
.long 0xe2823080 // add r3, r2, #128
.long 0xf2c3361f // vmov.i32 d19, #1056964608
.long 0xf2c3461f // vmov.i32 d20, #1056964608
.long 0xf4e31c9f // vld1.32 {d17[]}, [r3 :32]
.long 0xe2823084 // add r3, r2, #132
.long 0xf2403c31 // vfma.f32 d19, d0, d17
.long 0xf4e32c9f // vld1.32 {d18[]}, [r3 :32]
.long 0xf2c3061f // vmov.i32 d16, #1056964608
.long 0xf2414c32 // vfma.f32 d20, d1, d18
.long 0xf2420c31 // vfma.f32 d16, d2, d17
.long 0xe5913000 // ldr r3, [r1]
.long 0xe5933000 // ldr r3, [r3]
.long 0xf3fb17a3 // vcvt.u32.f32 d17, d19
.long 0xe0833080 // add r3, r3, r0, lsl #1
.long 0xf3fb27a4 // vcvt.u32.f32 d18, d20
.long 0xf3fb07a0 // vcvt.u32.f32 d16, d16
.long 0xf2eb1531 // vshl.s32 d17, d17, #11
.long 0xf2e52532 // vshl.s32 d18, d18, #5
.long 0xf26101b0 // vorr d16, d17, d16
.long 0xf26001b2 // vorr d16, d16, d18
.long 0xf3f60121 // vuzp.16 d16, d17
.long 0xf4c3080f // vst1.32 {d16[0]}, [r3]
.long 0xe5913004 // ldr r3, [r1, #4]
.long 0xe2811008 // add r1, r1, #8
.long 0xe12fff13 // bx r3
.globl _sk_load_8888_vfp4
_sk_load_8888_vfp4:
.long 0xe92d4800 // push {fp, lr}
@ -1666,6 +1777,52 @@ _sk_load_tables_hsw:
.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
.byte 72,139,0 // mov (%rax),%rax
.byte 196,226,125,51,20,120 // vpmovzxwd (%rax,%rdi,2),%ymm2
.byte 196,226,125,88,66,104 // vpbroadcastd 0x68(%rdx),%ymm0
.byte 197,253,219,194 // vpand %ymm2,%ymm0,%ymm0
.byte 197,252,91,192 // vcvtdq2ps %ymm0,%ymm0
.byte 196,226,125,24,74,116 // vbroadcastss 0x74(%rdx),%ymm1
.byte 197,244,89,192 // vmulps %ymm0,%ymm1,%ymm0
.byte 196,226,125,88,74,108 // vpbroadcastd 0x6c(%rdx),%ymm1
.byte 197,245,219,202 // vpand %ymm2,%ymm1,%ymm1
.byte 197,252,91,201 // vcvtdq2ps %ymm1,%ymm1
.byte 196,226,125,24,90,120 // vbroadcastss 0x78(%rdx),%ymm3
.byte 197,228,89,201 // vmulps %ymm1,%ymm3,%ymm1
.byte 196,226,125,88,90,112 // vpbroadcastd 0x70(%rdx),%ymm3
.byte 197,229,219,210 // vpand %ymm2,%ymm3,%ymm2
.byte 197,252,91,210 // vcvtdq2ps %ymm2,%ymm2
.byte 196,226,125,24,90,124 // vbroadcastss 0x7c(%rdx),%ymm3
.byte 197,228,89,210 // vmulps %ymm2,%ymm3,%ymm2
.byte 196,226,125,24,26 // vbroadcastss (%rdx),%ymm3
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_store_565_hsw
_sk_store_565_hsw:
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 72,139,0 // mov (%rax),%rax
.byte 196,98,125,24,130,128,0,0,0 // vbroadcastss 0x80(%rdx),%ymm8
.byte 197,60,89,200 // vmulps %ymm0,%ymm8,%ymm9
.byte 196,65,125,91,201 // vcvtps2dq %ymm9,%ymm9
.byte 196,193,53,114,241,11 // vpslld $0xb,%ymm9,%ymm9
.byte 196,98,125,24,146,132,0,0,0 // vbroadcastss 0x84(%rdx),%ymm10
.byte 197,44,89,209 // vmulps %ymm1,%ymm10,%ymm10
.byte 196,65,125,91,210 // vcvtps2dq %ymm10,%ymm10
.byte 196,193,45,114,242,5 // vpslld $0x5,%ymm10,%ymm10
.byte 196,65,45,235,201 // vpor %ymm9,%ymm10,%ymm9
.byte 197,60,89,194 // vmulps %ymm2,%ymm8,%ymm8
.byte 196,65,125,91,192 // vcvtps2dq %ymm8,%ymm8
.byte 196,65,53,235,192 // vpor %ymm8,%ymm9,%ymm8
.byte 196,67,125,57,193,1 // vextracti128 $0x1,%ymm8,%xmm9
.byte 196,66,57,43,193 // vpackusdw %xmm9,%xmm8,%xmm8
.byte 197,122,127,4,120 // vmovdqu %xmm8,(%rax,%rdi,2)
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_load_8888_hsw
_sk_load_8888_hsw:
.byte 72,173 // lods %ds:(%rsi),%rax
@ -2333,6 +2490,66 @@ _sk_load_tables_avx:
.byte 65,95 // pop %r15
.byte 255,224 // jmpq *%rax
.globl _sk_load_565_avx
_sk_load_565_avx:
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 72,139,0 // mov (%rax),%rax
.byte 196,226,121,51,68,120,8 // vpmovzxwd 0x8(%rax,%rdi,2),%xmm0
.byte 196,226,121,51,12,120 // vpmovzxwd (%rax,%rdi,2),%xmm1
.byte 196,227,117,24,208,1 // vinsertf128 $0x1,%xmm0,%ymm1,%ymm2
.byte 197,249,110,66,104 // vmovd 0x68(%rdx),%xmm0
.byte 196,227,121,4,192,0 // vpermilps $0x0,%xmm0,%xmm0
.byte 196,227,125,24,192,1 // vinsertf128 $0x1,%xmm0,%ymm0,%ymm0
.byte 197,252,84,194 // vandps %ymm2,%ymm0,%ymm0
.byte 197,252,91,192 // vcvtdq2ps %ymm0,%ymm0
.byte 196,226,125,24,74,116 // vbroadcastss 0x74(%rdx),%ymm1
.byte 197,244,89,192 // vmulps %ymm0,%ymm1,%ymm0
.byte 197,249,110,74,108 // vmovd 0x6c(%rdx),%xmm1
.byte 196,227,121,4,201,0 // vpermilps $0x0,%xmm1,%xmm1
.byte 196,227,117,24,201,1 // vinsertf128 $0x1,%xmm1,%ymm1,%ymm1
.byte 197,244,84,202 // vandps %ymm2,%ymm1,%ymm1
.byte 197,252,91,201 // vcvtdq2ps %ymm1,%ymm1
.byte 196,226,125,24,90,120 // vbroadcastss 0x78(%rdx),%ymm3
.byte 197,228,89,201 // vmulps %ymm1,%ymm3,%ymm1
.byte 197,249,110,90,112 // vmovd 0x70(%rdx),%xmm3
.byte 196,227,121,4,219,0 // vpermilps $0x0,%xmm3,%xmm3
.byte 196,227,101,24,219,1 // vinsertf128 $0x1,%xmm3,%ymm3,%ymm3
.byte 197,228,84,210 // vandps %ymm2,%ymm3,%ymm2
.byte 197,252,91,210 // vcvtdq2ps %ymm2,%ymm2
.byte 196,226,125,24,90,124 // vbroadcastss 0x7c(%rdx),%ymm3
.byte 197,228,89,210 // vmulps %ymm2,%ymm3,%ymm2
.byte 196,226,125,24,26 // vbroadcastss (%rdx),%ymm3
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_store_565_avx
_sk_store_565_avx:
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 72,139,0 // mov (%rax),%rax
.byte 196,98,125,24,130,128,0,0,0 // vbroadcastss 0x80(%rdx),%ymm8
.byte 197,60,89,200 // vmulps %ymm0,%ymm8,%ymm9
.byte 196,65,125,91,201 // vcvtps2dq %ymm9,%ymm9
.byte 196,193,41,114,241,11 // vpslld $0xb,%xmm9,%xmm10
.byte 196,67,125,25,201,1 // vextractf128 $0x1,%ymm9,%xmm9
.byte 196,193,49,114,241,11 // vpslld $0xb,%xmm9,%xmm9
.byte 196,67,45,24,201,1 // vinsertf128 $0x1,%xmm9,%ymm10,%ymm9
.byte 196,98,125,24,146,132,0,0,0 // vbroadcastss 0x84(%rdx),%ymm10
.byte 197,44,89,209 // vmulps %ymm1,%ymm10,%ymm10
.byte 196,65,125,91,210 // vcvtps2dq %ymm10,%ymm10
.byte 196,193,33,114,242,5 // vpslld $0x5,%xmm10,%xmm11
.byte 196,67,125,25,210,1 // vextractf128 $0x1,%ymm10,%xmm10
.byte 196,193,41,114,242,5 // vpslld $0x5,%xmm10,%xmm10
.byte 196,67,37,24,210,1 // vinsertf128 $0x1,%xmm10,%ymm11,%ymm10
.byte 196,65,45,86,201 // vorpd %ymm9,%ymm10,%ymm9
.byte 197,60,89,194 // vmulps %ymm2,%ymm8,%ymm8
.byte 196,65,125,91,192 // vcvtps2dq %ymm8,%ymm8
.byte 196,65,53,86,192 // vorpd %ymm8,%ymm9,%ymm8
.byte 196,67,125,25,193,1 // vextractf128 $0x1,%ymm8,%xmm9
.byte 196,66,57,43,193 // vpackusdw %xmm9,%xmm8,%xmm8
.byte 197,122,127,4,120 // vmovdqu %xmm8,(%rax,%rdi,2)
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_load_8888_avx
_sk_load_8888_avx:
.byte 72,173 // lods %ds:(%rsi),%rax
@ -3096,6 +3313,61 @@ _sk_load_tables_sse41:
.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
.byte 72,139,0 // mov (%rax),%rax
.byte 102,68,15,56,51,12,120 // pmovzxwd (%rax,%rdi,2),%xmm9
.byte 102,15,110,66,104 // movd 0x68(%rdx),%xmm0
.byte 102,15,112,192,0 // pshufd $0x0,%xmm0,%xmm0
.byte 102,65,15,219,193 // pand %xmm9,%xmm0
.byte 15,91,200 // cvtdq2ps %xmm0,%xmm1
.byte 243,15,16,26 // movss (%rdx),%xmm3
.byte 243,15,16,66,116 // movss 0x74(%rdx),%xmm0
.byte 15,198,192,0 // shufps $0x0,%xmm0,%xmm0
.byte 15,89,193 // mulps %xmm1,%xmm0
.byte 102,15,110,74,108 // movd 0x6c(%rdx),%xmm1
.byte 102,15,112,201,0 // pshufd $0x0,%xmm1,%xmm1
.byte 102,65,15,219,201 // pand %xmm9,%xmm1
.byte 68,15,91,193 // cvtdq2ps %xmm1,%xmm8
.byte 243,15,16,74,120 // movss 0x78(%rdx),%xmm1
.byte 15,198,201,0 // shufps $0x0,%xmm1,%xmm1
.byte 65,15,89,200 // mulps %xmm8,%xmm1
.byte 102,15,110,82,112 // movd 0x70(%rdx),%xmm2
.byte 102,15,112,210,0 // pshufd $0x0,%xmm2,%xmm2
.byte 102,65,15,219,209 // pand %xmm9,%xmm2
.byte 68,15,91,194 // cvtdq2ps %xmm2,%xmm8
.byte 243,15,16,82,124 // movss 0x7c(%rdx),%xmm2
.byte 15,198,210,0 // shufps $0x0,%xmm2,%xmm2
.byte 65,15,89,208 // mulps %xmm8,%xmm2
.byte 15,198,219,0 // shufps $0x0,%xmm3,%xmm3
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_store_565_sse41
_sk_store_565_sse41:
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 72,139,0 // mov (%rax),%rax
.byte 243,68,15,16,130,128,0,0,0 // movss 0x80(%rdx),%xmm8
.byte 243,68,15,16,138,132,0,0,0 // movss 0x84(%rdx),%xmm9
.byte 69,15,198,192,0 // shufps $0x0,%xmm8,%xmm8
.byte 69,15,40,208 // movaps %xmm8,%xmm10
.byte 68,15,89,208 // mulps %xmm0,%xmm10
.byte 102,69,15,91,210 // cvtps2dq %xmm10,%xmm10
.byte 102,65,15,114,242,11 // pslld $0xb,%xmm10
.byte 69,15,198,201,0 // shufps $0x0,%xmm9,%xmm9
.byte 68,15,89,201 // mulps %xmm1,%xmm9
.byte 102,69,15,91,201 // cvtps2dq %xmm9,%xmm9
.byte 102,65,15,114,241,5 // pslld $0x5,%xmm9
.byte 102,69,15,235,202 // por %xmm10,%xmm9
.byte 68,15,89,194 // mulps %xmm2,%xmm8
.byte 102,69,15,91,192 // cvtps2dq %xmm8,%xmm8
.byte 102,69,15,86,193 // orpd %xmm9,%xmm8
.byte 102,69,15,56,43,192 // packusdw %xmm8,%xmm8
.byte 102,68,15,214,4,120 // movq %xmm8,(%rax,%rdi,2)
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_load_8888_sse41
_sk_load_8888_sse41:
.byte 72,173 // lods %ds:(%rsi),%rax
@ -3854,6 +4126,65 @@ _sk_load_tables_sse2:
.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
.byte 72,139,0 // mov (%rax),%rax
.byte 243,68,15,126,12,120 // movq (%rax,%rdi,2),%xmm9
.byte 102,15,239,192 // pxor %xmm0,%xmm0
.byte 102,68,15,97,200 // punpcklwd %xmm0,%xmm9
.byte 102,15,110,66,104 // movd 0x68(%rdx),%xmm0
.byte 102,15,112,192,0 // pshufd $0x0,%xmm0,%xmm0
.byte 102,65,15,219,193 // pand %xmm9,%xmm0
.byte 15,91,200 // cvtdq2ps %xmm0,%xmm1
.byte 243,15,16,26 // movss (%rdx),%xmm3
.byte 243,15,16,66,116 // movss 0x74(%rdx),%xmm0
.byte 15,198,192,0 // shufps $0x0,%xmm0,%xmm0
.byte 15,89,193 // mulps %xmm1,%xmm0
.byte 102,15,110,74,108 // movd 0x6c(%rdx),%xmm1
.byte 102,15,112,201,0 // pshufd $0x0,%xmm1,%xmm1
.byte 102,65,15,219,201 // pand %xmm9,%xmm1
.byte 68,15,91,193 // cvtdq2ps %xmm1,%xmm8
.byte 243,15,16,74,120 // movss 0x78(%rdx),%xmm1
.byte 15,198,201,0 // shufps $0x0,%xmm1,%xmm1
.byte 65,15,89,200 // mulps %xmm8,%xmm1
.byte 102,15,110,82,112 // movd 0x70(%rdx),%xmm2
.byte 102,15,112,210,0 // pshufd $0x0,%xmm2,%xmm2
.byte 102,65,15,219,209 // pand %xmm9,%xmm2
.byte 68,15,91,194 // cvtdq2ps %xmm2,%xmm8
.byte 243,15,16,82,124 // movss 0x7c(%rdx),%xmm2
.byte 15,198,210,0 // shufps $0x0,%xmm2,%xmm2
.byte 65,15,89,208 // mulps %xmm8,%xmm2
.byte 15,198,219,0 // shufps $0x0,%xmm3,%xmm3
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_store_565_sse2
_sk_store_565_sse2:
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 72,139,0 // mov (%rax),%rax
.byte 243,68,15,16,130,128,0,0,0 // movss 0x80(%rdx),%xmm8
.byte 243,68,15,16,138,132,0,0,0 // movss 0x84(%rdx),%xmm9
.byte 69,15,198,192,0 // shufps $0x0,%xmm8,%xmm8
.byte 69,15,40,208 // movaps %xmm8,%xmm10
.byte 68,15,89,208 // mulps %xmm0,%xmm10
.byte 102,69,15,91,210 // cvtps2dq %xmm10,%xmm10
.byte 102,65,15,114,242,11 // pslld $0xb,%xmm10
.byte 69,15,198,201,0 // shufps $0x0,%xmm9,%xmm9
.byte 68,15,89,201 // mulps %xmm1,%xmm9
.byte 102,69,15,91,201 // cvtps2dq %xmm9,%xmm9
.byte 102,65,15,114,241,5 // pslld $0x5,%xmm9
.byte 102,69,15,235,202 // por %xmm10,%xmm9
.byte 68,15,89,194 // mulps %xmm2,%xmm8
.byte 102,69,15,91,192 // cvtps2dq %xmm8,%xmm8
.byte 102,69,15,86,193 // orpd %xmm9,%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 102,68,15,214,4,120 // movq %xmm8,(%rax,%rdi,2)
.byte 72,173 // lods %ds:(%rsi),%rax
.byte 255,224 // jmpq *%rax
.globl _sk_load_8888_sse2
_sk_load_8888_sse2:
.byte 72,173 // lods %ds:(%rsi),%rax

220
src/jumper/SkJumper_generated_win.S
View File

@ -401,6 +401,52 @@ _sk_load_tables_hsw LABEL PROC
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
DB 72,139,0 ; mov (%rax),%rax
DB 196,226,125,51,20,120 ; vpmovzxwd (%rax,%rdi,2),%ymm2
DB 196,226,125,88,66,104 ; vpbroadcastd 0x68(%rdx),%ymm0
DB 197,253,219,194 ; vpand %ymm2,%ymm0,%ymm0
DB 197,252,91,192 ; vcvtdq2ps %ymm0,%ymm0
DB 196,226,125,24,74,116 ; vbroadcastss 0x74(%rdx),%ymm1
DB 197,244,89,192 ; vmulps %ymm0,%ymm1,%ymm0
DB 196,226,125,88,74,108 ; vpbroadcastd 0x6c(%rdx),%ymm1
DB 197,245,219,202 ; vpand %ymm2,%ymm1,%ymm1
DB 197,252,91,201 ; vcvtdq2ps %ymm1,%ymm1
DB 196,226,125,24,90,120 ; vbroadcastss 0x78(%rdx),%ymm3
DB 197,228,89,201 ; vmulps %ymm1,%ymm3,%ymm1
DB 196,226,125,88,90,112 ; vpbroadcastd 0x70(%rdx),%ymm3
DB 197,229,219,210 ; vpand %ymm2,%ymm3,%ymm2
DB 197,252,91,210 ; vcvtdq2ps %ymm2,%ymm2
DB 196,226,125,24,90,124 ; vbroadcastss 0x7c(%rdx),%ymm3
DB 197,228,89,210 ; vmulps %ymm2,%ymm3,%ymm2
DB 196,226,125,24,26 ; vbroadcastss (%rdx),%ymm3
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_store_565_hsw
_sk_store_565_hsw LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
DB 72,139,0 ; mov (%rax),%rax
DB 196,98,125,24,130,128,0,0,0 ; vbroadcastss 0x80(%rdx),%ymm8
DB 197,60,89,200 ; vmulps %ymm0,%ymm8,%ymm9
DB 196,65,125,91,201 ; vcvtps2dq %ymm9,%ymm9
DB 196,193,53,114,241,11 ; vpslld $0xb,%ymm9,%ymm9
DB 196,98,125,24,146,132,0,0,0 ; vbroadcastss 0x84(%rdx),%ymm10
DB 197,44,89,209 ; vmulps %ymm1,%ymm10,%ymm10
DB 196,65,125,91,210 ; vcvtps2dq %ymm10,%ymm10
DB 196,193,45,114,242,5 ; vpslld $0x5,%ymm10,%ymm10
DB 196,65,45,235,201 ; vpor %ymm9,%ymm10,%ymm9
DB 197,60,89,194 ; vmulps %ymm2,%ymm8,%ymm8
DB 196,65,125,91,192 ; vcvtps2dq %ymm8,%ymm8
DB 196,65,53,235,192 ; vpor %ymm8,%ymm9,%ymm8
DB 196,67,125,57,193,1 ; vextracti128 $0x1,%ymm8,%xmm9
DB 196,66,57,43,193 ; vpackusdw %xmm9,%xmm8,%xmm8
DB 197,122,127,4,120 ; vmovdqu %xmm8,(%rax,%rdi,2)
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_load_8888_hsw
_sk_load_8888_hsw LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
@ -1095,6 +1141,66 @@ _sk_load_tables_avx LABEL PROC
DB 65,95 ; pop %r15
DB 255,224 ; jmpq *%rax
PUBLIC _sk_load_565_avx
_sk_load_565_avx LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
DB 72,139,0 ; mov (%rax),%rax
DB 196,226,121,51,68,120,8 ; vpmovzxwd 0x8(%rax,%rdi,2),%xmm0
DB 196,226,121,51,12,120 ; vpmovzxwd (%rax,%rdi,2),%xmm1
DB 196,227,117,24,208,1 ; vinsertf128 $0x1,%xmm0,%ymm1,%ymm2
DB 197,249,110,66,104 ; vmovd 0x68(%rdx),%xmm0
DB 196,227,121,4,192,0 ; vpermilps $0x0,%xmm0,%xmm0
DB 196,227,125,24,192,1 ; vinsertf128 $0x1,%xmm0,%ymm0,%ymm0
DB 197,252,84,194 ; vandps %ymm2,%ymm0,%ymm0
DB 197,252,91,192 ; vcvtdq2ps %ymm0,%ymm0
DB 196,226,125,24,74,116 ; vbroadcastss 0x74(%rdx),%ymm1
DB 197,244,89,192 ; vmulps %ymm0,%ymm1,%ymm0
DB 197,249,110,74,108 ; vmovd 0x6c(%rdx),%xmm1
DB 196,227,121,4,201,0 ; vpermilps $0x0,%xmm1,%xmm1
DB 196,227,117,24,201,1 ; vinsertf128 $0x1,%xmm1,%ymm1,%ymm1
DB 197,244,84,202 ; vandps %ymm2,%ymm1,%ymm1
DB 197,252,91,201 ; vcvtdq2ps %ymm1,%ymm1
DB 196,226,125,24,90,120 ; vbroadcastss 0x78(%rdx),%ymm3
DB 197,228,89,201 ; vmulps %ymm1,%ymm3,%ymm1
DB 197,249,110,90,112 ; vmovd 0x70(%rdx),%xmm3
DB 196,227,121,4,219,0 ; vpermilps $0x0,%xmm3,%xmm3
DB 196,227,101,24,219,1 ; vinsertf128 $0x1,%xmm3,%ymm3,%ymm3
DB 197,228,84,210 ; vandps %ymm2,%ymm3,%ymm2
DB 197,252,91,210 ; vcvtdq2ps %ymm2,%ymm2
DB 196,226,125,24,90,124 ; vbroadcastss 0x7c(%rdx),%ymm3
DB 197,228,89,210 ; vmulps %ymm2,%ymm3,%ymm2
DB 196,226,125,24,26 ; vbroadcastss (%rdx),%ymm3
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_store_565_avx
_sk_store_565_avx LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
DB 72,139,0 ; mov (%rax),%rax
DB 196,98,125,24,130,128,0,0,0 ; vbroadcastss 0x80(%rdx),%ymm8
DB 197,60,89,200 ; vmulps %ymm0,%ymm8,%ymm9
DB 196,65,125,91,201 ; vcvtps2dq %ymm9,%ymm9
DB 196,193,41,114,241,11 ; vpslld $0xb,%xmm9,%xmm10
DB 196,67,125,25,201,1 ; vextractf128 $0x1,%ymm9,%xmm9
DB 196,193,49,114,241,11 ; vpslld $0xb,%xmm9,%xmm9
DB 196,67,45,24,201,1 ; vinsertf128 $0x1,%xmm9,%ymm10,%ymm9
DB 196,98,125,24,146,132,0,0,0 ; vbroadcastss 0x84(%rdx),%ymm10
DB 197,44,89,209 ; vmulps %ymm1,%ymm10,%ymm10
DB 196,65,125,91,210 ; vcvtps2dq %ymm10,%ymm10
DB 196,193,33,114,242,5 ; vpslld $0x5,%xmm10,%xmm11
DB 196,67,125,25,210,1 ; vextractf128 $0x1,%ymm10,%xmm10
DB 196,193,41,114,242,5 ; vpslld $0x5,%xmm10,%xmm10
DB 196,67,37,24,210,1 ; vinsertf128 $0x1,%xmm10,%ymm11,%ymm10
DB 196,65,45,86,201 ; vorpd %ymm9,%ymm10,%ymm9
DB 197,60,89,194 ; vmulps %ymm2,%ymm8,%ymm8
DB 196,65,125,91,192 ; vcvtps2dq %ymm8,%ymm8
DB 196,65,53,86,192 ; vorpd %ymm8,%ymm9,%ymm8
DB 196,67,125,25,193,1 ; vextractf128 $0x1,%ymm8,%xmm9
DB 196,66,57,43,193 ; vpackusdw %xmm9,%xmm8,%xmm8
DB 197,122,127,4,120 ; vmovdqu %xmm8,(%rax,%rdi,2)
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_load_8888_avx
_sk_load_8888_avx LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
@ -1885,6 +1991,61 @@ _sk_load_tables_sse41 LABEL PROC
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
DB 72,139,0 ; mov (%rax),%rax
DB 102,68,15,56,51,12,120 ; pmovzxwd (%rax,%rdi,2),%xmm9
DB 102,15,110,66,104 ; movd 0x68(%rdx),%xmm0
DB 102,15,112,192,0 ; pshufd $0x0,%xmm0,%xmm0
DB 102,65,15,219,193 ; pand %xmm9,%xmm0
DB 15,91,200 ; cvtdq2ps %xmm0,%xmm1
DB 243,15,16,26 ; movss (%rdx),%xmm3
DB 243,15,16,66,116 ; movss 0x74(%rdx),%xmm0
DB 15,198,192,0 ; shufps $0x0,%xmm0,%xmm0
DB 15,89,193 ; mulps %xmm1,%xmm0
DB 102,15,110,74,108 ; movd 0x6c(%rdx),%xmm1
DB 102,15,112,201,0 ; pshufd $0x0,%xmm1,%xmm1
DB 102,65,15,219,201 ; pand %xmm9,%xmm1
DB 68,15,91,193 ; cvtdq2ps %xmm1,%xmm8
DB 243,15,16,74,120 ; movss 0x78(%rdx),%xmm1
DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1
DB 65,15,89,200 ; mulps %xmm8,%xmm1
DB 102,15,110,82,112 ; movd 0x70(%rdx),%xmm2
DB 102,15,112,210,0 ; pshufd $0x0,%xmm2,%xmm2
DB 102,65,15,219,209 ; pand %xmm9,%xmm2
DB 68,15,91,194 ; cvtdq2ps %xmm2,%xmm8
DB 243,15,16,82,124 ; movss 0x7c(%rdx),%xmm2
DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2
DB 65,15,89,208 ; mulps %xmm8,%xmm2
DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_store_565_sse41
_sk_store_565_sse41 LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
DB 72,139,0 ; mov (%rax),%rax
DB 243,68,15,16,130,128,0,0,0 ; movss 0x80(%rdx),%xmm8
DB 243,68,15,16,138,132,0,0,0 ; movss 0x84(%rdx),%xmm9
DB 69,15,198,192,0 ; shufps $0x0,%xmm8,%xmm8
DB 69,15,40,208 ; movaps %xmm8,%xmm10
DB 68,15,89,208 ; mulps %xmm0,%xmm10
DB 102,69,15,91,210 ; cvtps2dq %xmm10,%xmm10
DB 102,65,15,114,242,11 ; pslld $0xb,%xmm10
DB 69,15,198,201,0 ; shufps $0x0,%xmm9,%xmm9
DB 68,15,89,201 ; mulps %xmm1,%xmm9
DB 102,69,15,91,201 ; cvtps2dq %xmm9,%xmm9
DB 102,65,15,114,241,5 ; pslld $0x5,%xmm9
DB 102,69,15,235,202 ; por %xmm10,%xmm9
DB 68,15,89,194 ; mulps %xmm2,%xmm8
DB 102,69,15,91,192 ; cvtps2dq %xmm8,%xmm8
DB 102,69,15,86,193 ; orpd %xmm9,%xmm8
DB 102,69,15,56,43,192 ; packusdw %xmm8,%xmm8
DB 102,68,15,214,4,120 ; movq %xmm8,(%rax,%rdi,2)
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_load_8888_sse41
_sk_load_8888_sse41 LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
@ -2670,6 +2831,65 @@ _sk_load_tables_sse2 LABEL PROC
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
DB 72,139,0 ; mov (%rax),%rax
DB 243,68,15,126,12,120 ; movq (%rax,%rdi,2),%xmm9
DB 102,15,239,192 ; pxor %xmm0,%xmm0
DB 102,68,15,97,200 ; punpcklwd %xmm0,%xmm9
DB 102,15,110,66,104 ; movd 0x68(%rdx),%xmm0
DB 102,15,112,192,0 ; pshufd $0x0,%xmm0,%xmm0
DB 102,65,15,219,193 ; pand %xmm9,%xmm0
DB 15,91,200 ; cvtdq2ps %xmm0,%xmm1
DB 243,15,16,26 ; movss (%rdx),%xmm3
DB 243,15,16,66,116 ; movss 0x74(%rdx),%xmm0
DB 15,198,192,0 ; shufps $0x0,%xmm0,%xmm0
DB 15,89,193 ; mulps %xmm1,%xmm0
DB 102,15,110,74,108 ; movd 0x6c(%rdx),%xmm1
DB 102,15,112,201,0 ; pshufd $0x0,%xmm1,%xmm1
DB 102,65,15,219,201 ; pand %xmm9,%xmm1
DB 68,15,91,193 ; cvtdq2ps %xmm1,%xmm8
DB 243,15,16,74,120 ; movss 0x78(%rdx),%xmm1
DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1
DB 65,15,89,200 ; mulps %xmm8,%xmm1
DB 102,15,110,82,112 ; movd 0x70(%rdx),%xmm2
DB 102,15,112,210,0 ; pshufd $0x0,%xmm2,%xmm2
DB 102,65,15,219,209 ; pand %xmm9,%xmm2
DB 68,15,91,194 ; cvtdq2ps %xmm2,%xmm8
DB 243,15,16,82,124 ; movss 0x7c(%rdx),%xmm2
DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2
DB 65,15,89,208 ; mulps %xmm8,%xmm2
DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_store_565_sse2
_sk_store_565_sse2 LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax
DB 72,139,0 ; mov (%rax),%rax
DB 243,68,15,16,130,128,0,0,0 ; movss 0x80(%rdx),%xmm8
DB 243,68,15,16,138,132,0,0,0 ; movss 0x84(%rdx),%xmm9
DB 69,15,198,192,0 ; shufps $0x0,%xmm8,%xmm8
DB 69,15,40,208 ; movaps %xmm8,%xmm10
DB 68,15,89,208 ; mulps %xmm0,%xmm10
DB 102,69,15,91,210 ; cvtps2dq %xmm10,%xmm10
DB 102,65,15,114,242,11 ; pslld $0xb,%xmm10
DB 69,15,198,201,0 ; shufps $0x0,%xmm9,%xmm9
DB 68,15,89,201 ; mulps %xmm1,%xmm9
DB 102,69,15,91,201 ; cvtps2dq %xmm9,%xmm9
DB 102,65,15,114,241,5 ; pslld $0x5,%xmm9
DB 102,69,15,235,202 ; por %xmm10,%xmm9
DB 68,15,89,194 ; mulps %xmm2,%xmm8
DB 102,69,15,91,192 ; cvtps2dq %xmm8,%xmm8
DB 102,69,15,86,193 ; orpd %xmm9,%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 102,68,15,214,4,120 ; movq %xmm8,(%rax,%rdi,2)
DB 72,173 ; lods %ds:(%rsi),%rax
DB 255,224 ; jmpq *%rax
PUBLIC _sk_load_8888_sse2
_sk_load_8888_sse2 LABEL PROC
DB 72,173 ; lods %ds:(%rsi),%rax

84
src/jumper/SkJumper_stages.cpp
View File

@ -11,6 +11,19 @@
// It's tricky to relocate code referencing ordinary constants, so we read them from this struct.
using K = const SkJumper_constants;
template <typename T, typename P>
static T unaligned_load(const P* p) {
T v;
memcpy(&v, p, sizeof(v));
return v;
}
template <typename Dst, typename Src>
static Dst bit_cast(const Src& src) {
static_assert(sizeof(Dst) == sizeof(Src), "");
return unaligned_load<Dst>(&src);
}
#if !defined(JUMPER)
// This path should lead to portable code that can be compiled directly into Skia.
// (All other paths are compiled offline by Clang into SkJumper_generated.h.)
@ -19,6 +32,7 @@ using K = const SkJumper_constants;
using F = float;
using I32 = int32_t;
using U32 = uint32_t;
using U16 = uint16_t;
using U8 = uint8_t;
static F mad(F f, F m, F a) { return f*m+a; }
@ -27,6 +41,7 @@ using K = const SkJumper_constants;
static F rcp (F v) { return 1.0f / v; }
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 F if_then_else(I32 c, F t, F e) { return c ? t : e; }
@ -41,6 +56,7 @@ using K = const SkJumper_constants;
using F = float __attribute__((ext_vector_type(4)));
using I32 = int32_t __attribute__((ext_vector_type(4)));
using U32 = uint32_t __attribute__((ext_vector_type(4)));
using U16 = uint16_t __attribute__((ext_vector_type(4)));
using U8 = uint8_t __attribute__((ext_vector_type(4)));
// We polyfill a few routines that Clang doesn't build into ext_vector_types.
@ -50,6 +66,7 @@ using K = const SkJumper_constants;
static F rcp (F v) { auto e = vrecpeq_f32 (v); return vrecpsq_f32 (v,e ) * e; }
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 F if_then_else(I32 c, F t, F e) { return vbslq_f32((U32)c,t,e); }
@ -67,6 +84,7 @@ using K = const SkJumper_constants;
using F = float __attribute__((ext_vector_type(2)));
using I32 = int32_t __attribute__((ext_vector_type(2)));
using U32 = uint32_t __attribute__((ext_vector_type(2)));
using U16 = uint16_t __attribute__((ext_vector_type(2)));
using U8 = uint8_t __attribute__((ext_vector_type(2)));
static F mad(F f, F m, F a) { return vfma_f32(a,f,m); }
@ -75,6 +93,7 @@ using K = const SkJumper_constants;
static F rcp (F v) { auto e = vrecpe_f32 (v); return vrecps_f32 (v,e ) * e; }
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 F if_then_else(I32 c, F t, F e) { return vbsl_f32((U32)c,t,e); }
@ -89,6 +108,7 @@ using K = const SkJumper_constants;
using F = float __attribute__((ext_vector_type(8)));
using I32 = int32_t __attribute__((ext_vector_type(8)));
using U32 = uint32_t __attribute__((ext_vector_type(8)));
using U16 = uint16_t __attribute__((ext_vector_type(8)));
using U8 = uint8_t __attribute__((ext_vector_type(8)));
static F mad(F f, F m, F a) { return _mm256_fmadd_ps(f,m,a);}
@ -98,6 +118,12 @@ using K = const SkJumper_constants;
static F rsqrt(F v) { return _mm256_rsqrt_ps(v); }
static U32 round(F v, F scale) { return _mm256_cvtps_epi32(v*scale); }
static U16 pack(U32 v) {
__m128i lo = _mm256_extractf128_si256(v, 0),
hi = _mm256_extractf128_si256(v, 1);
return _mm_packus_epi32(lo, hi);
}
static F if_then_else(I32 c, F t, F e) { return _mm256_blendv_ps(e,t,c); }
static F gather(const float* p, U32 ix) { return _mm256_i32gather_ps(p, ix, 4); }
@ -110,6 +136,7 @@ using K = const SkJumper_constants;
using F = float __attribute__((ext_vector_type(8)));
using I32 = int32_t __attribute__((ext_vector_type(8)));
using U32 = uint32_t __attribute__((ext_vector_type(8)));
using U16 = uint16_t __attribute__((ext_vector_type(8)));
using U8 = uint8_t __attribute__((ext_vector_type(8)));
static F mad(F f, F m, F a) { return f*m+a; }
@ -119,6 +146,12 @@ using K = const SkJumper_constants;
static F rsqrt(F v) { return _mm256_rsqrt_ps(v); }
static U32 round(F v, F scale) { return _mm256_cvtps_epi32(v*scale); }
static U16 pack(U32 v) {
__m128i lo = _mm256_extractf128_si256(v, 0),
hi = _mm256_extractf128_si256(v, 1);
return _mm_packus_epi32(lo, hi);
}
static F if_then_else(I32 c, F t, F e) { return _mm256_blendv_ps(e,t,c); }
static F gather(const float* p, U32 ix) {
@ -134,6 +167,7 @@ using K = const SkJumper_constants;
using F = float __attribute__((ext_vector_type(4)));
using I32 = int32_t __attribute__((ext_vector_type(4)));
using U32 = uint32_t __attribute__((ext_vector_type(4)));
using U16 = uint16_t __attribute__((ext_vector_type(4)));
using U8 = uint8_t __attribute__((ext_vector_type(4)));
static F mad(F f, F m, F a) { return f*m+a; }
@ -143,6 +177,17 @@ using K = const SkJumper_constants;
static F rsqrt(F v) { return _mm_rsqrt_ps(v); }
static U32 round(F v, F scale) { return _mm_cvtps_epi32(v*scale); }
static U16 pack(U32 v) {
#if defined(__SSE4_1__)
auto p = _mm_packus_epi32(v,v);
#else
// Sign extend so that _mm_packs_epi32() does the pack we want.
auto p = _mm_srai_epi32(_mm_slli_epi32(v, 16), 16);
p = _mm_packs_epi32(p,p);
#endif
return unaligned_load<U16>(&p); // We have two copies. Return (the lower) one.
}
static F if_then_else(I32 c, F t, F e) {
#if defined(__SSE4_1__)
return _mm_blendv_ps(e,t,c);
@ -160,32 +205,29 @@ using K = const SkJumper_constants;
#endif
#endif
static F lerp(F from, F to, F t) {
return mad(to-from, t, from);
}
// We need to be a careful with casts.
// (F)x means cast x to float in the portable path, but bit_cast x to float in the others.
// These named casts and bit_cast() are always what they seem to be.
#if defined(JUMPER)
static F cast (U32 v) { return __builtin_convertvector((I32)v, F); }
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 U32 expand(U16 v) { return (U32)v; }
static U32 expand(U8 v) { return (U32)v; }
#endif
template <typename T, typename P>
static T unaligned_load(const P* p) {
T v;
memcpy(&v, p, sizeof(v));
return v;
static F lerp(F from, F to, F t) {
return mad(to-from, t, from);
}
template <typename Dst, typename Src>
static Dst bit_cast(const Src& src) {
static_assert(sizeof(Dst) == sizeof(Src), "");
return unaligned_load<Dst>(&src);
static void from_565(U16 _565, F* r, F* g, F* b, K* k) {
U32 wide = expand(_565);
*r = cast(wide & k->r_565_mask) * k->r_565_scale;
*g = cast(wide & k->g_565_mask) * k->g_565_scale;
*b = cast(wide & k->b_565_mask) * k->b_565_scale;
}
// Sometimes we want to work with 4 floats directly, regardless of the depth of the F vector.
@ -455,6 +497,22 @@ STAGE(load_tables) {
a = cast( (px >> 24)) * k->_1_255;
}
STAGE(load_565) {
auto ptr = *(const uint16_t**)ctx + x;
auto px = unaligned_load<U16>(ptr);
from_565(px, &r,&g,&b, k);
a = k->_1;
}
STAGE(store_565) {
auto ptr = *(uint16_t**)ctx + x;
U16 px = pack( round(r, k->_31) << 11
| round(g, k->_63) << 5
| round(b, k->_31) );
memcpy(ptr, &px, sizeof(px));
}
STAGE(load_8888) {
auto ptr = *(const uint32_t**)ctx + x;