From 14987ebb97f491f1b5bc776252b5ddbf65b8fca0 Mon Sep 17 00:00:00 2001 From: Mike Klein Date: Thu, 6 Apr 2017 10:22:26 -0400 Subject: jumper, add load_f32() Change-Id: I71d85ffe29bc11678ff1e696fa4a2c93d0b4fcbe Reviewed-on: https://skia-review.googlesource.com/11446 Reviewed-by: Herb Derby Commit-Queue: Mike Klein --- src/jumper/SkJumper.cpp | 1 + src/jumper/SkJumper_generated.S | 206 +++++++++++++++++++++++++++++++++--- src/jumper/SkJumper_generated_win.S | 183 +++++++++++++++++++++++++++++--- src/jumper/SkJumper_stages.cpp | 4 + src/jumper/SkJumper_vectors.h | 60 +++++++++++ 5 files changed, 420 insertions(+), 34 deletions(-) (limited to 'src') diff --git a/src/jumper/SkJumper.cpp b/src/jumper/SkJumper.cpp index 4ad2afd896..e4836aac11 100644 --- a/src/jumper/SkJumper.cpp +++ b/src/jumper/SkJumper.cpp @@ -99,6 +99,7 @@ static K kConstants = { M(store_f16) \ M(load_u16_be) \ M(store_u16_be) \ + M(load_f32) \ M(store_f32) \ M(luminance_to_alpha) \ M(matrix_2x3) \ diff --git a/src/jumper/SkJumper_generated.S b/src/jumper/SkJumper_generated.S index 75e1ad2c45..cff090471e 100644 --- a/src/jumper/SkJumper_generated.S +++ b/src/jumper/SkJumper_generated.S @@ -1678,6 +1678,15 @@ _sk_store_u16_be_aarch64: .long 0x91004021 // add x1, x1, #0x10 .long 0xd61f0060 // br x3 +HIDDEN _sk_load_f32_aarch64 +.globl _sk_load_f32_aarch64 +_sk_load_f32_aarch64: + .long 0xa8c10c28 // ldp x8, x3, [x1], #16 + .long 0xf9400108 // ldr x8, [x8] + .long 0x8b001108 // add x8, x8, x0, lsl #4 + .long 0x4c400900 // ld4 {v0.4s-v3.4s}, [x8] + .long 0xd61f0060 // br x3 + HIDDEN _sk_store_f32_aarch64 .globl _sk_store_f32_aarch64 _sk_store_f32_aarch64: @@ -3760,6 +3769,16 @@ _sk_store_u16_be_vfp4: .long 0x477fff00 // .word 0x477fff00 .long 0x477fff00 // .word 0x477fff00 +HIDDEN _sk_load_f32_vfp4 +.globl _sk_load_f32_vfp4 +_sk_load_f32_vfp4: + .long 0xe8911008 // ldm r1, {r3, ip} + .long 0xe2811008 // add r1, r1, #8 + .long 0xe5933000 // ldr r3, [r3] + .long 0xe0833200 // add r3, r3, r0, lsl #4 + .long 0xf423008f // vld4.32 {d0-d3}, [r3] + .long 0xe12fff1c // bx ip + HIDDEN _sk_store_f32_vfp4 .globl _sk_store_f32_vfp4 _sk_store_f32_vfp4: @@ -5768,7 +5787,7 @@ _sk_load_4444_hsw: .byte 255 // (bad) .byte 255 // (bad) .byte 255 // (bad) - .byte 233,255,255,255,225 // jmpq ffffffffe2001650 <_sk_linear_gradient_2stops_hsw+0xffffffffe1fff563> + .byte 233,255,255,255,225 // jmpq ffffffffe2001650 <_sk_linear_gradient_2stops_hsw+0xffffffffe1fff4a4> .byte 255 // (bad) .byte 255 // (bad) .byte 255 // (bad) @@ -6190,6 +6209,56 @@ _sk_store_u16_be_hsw: .byte 196,65,121,214,68,248,48 // vmovq %xmm8,0x30(%r8,%rdi,8) .byte 235,174 // jmp 1c57 <_sk_store_u16_be_hsw+0xef> +HIDDEN _sk_load_f32_hsw +.globl _sk_load_f32_hsw +_sk_load_f32_hsw: + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 72,131,249,7 // cmp $0x7,%rcx + .byte 119,110 // ja 1d1f <_sk_load_f32_hsw+0x76> + .byte 76,139,0 // mov (%rax),%r8 + .byte 76,141,12,189,0,0,0,0 // lea 0x0(,%rdi,4),%r9 + .byte 76,141,21,133,0,0,0 // lea 0x85(%rip),%r10 # 1d48 <_sk_load_f32_hsw+0x9f> + .byte 73,99,4,138 // movslq (%r10,%rcx,4),%rax + .byte 76,1,208 // add %r10,%rax + .byte 255,224 // jmpq *%rax + .byte 196,3,125,24,68,136,112,1 // vinsertf128 $0x1,0x70(%r8,%r9,4),%ymm0,%ymm8 + .byte 196,131,125,24,92,136,96,1 // vinsertf128 $0x1,0x60(%r8,%r9,4),%ymm0,%ymm3 + .byte 196,131,125,24,76,136,80,1 // vinsertf128 $0x1,0x50(%r8,%r9,4),%ymm0,%ymm1 + .byte 196,131,125,24,84,136,64,1 // vinsertf128 $0x1,0x40(%r8,%r9,4),%ymm0,%ymm2 + .byte 196,129,121,16,68,136,48 // vmovupd 0x30(%r8,%r9,4),%xmm0 + .byte 196,195,125,13,192,12 // vblendpd $0xc,%ymm8,%ymm0,%ymm0 + .byte 196,1,121,16,68,136,32 // vmovupd 0x20(%r8,%r9,4),%xmm8 + .byte 196,99,61,13,203,12 // vblendpd $0xc,%ymm3,%ymm8,%ymm9 + .byte 196,129,121,16,92,136,16 // vmovupd 0x10(%r8,%r9,4),%xmm3 + .byte 196,99,101,13,209,12 // vblendpd $0xc,%ymm1,%ymm3,%ymm10 + .byte 196,129,121,16,12,136 // vmovupd (%r8,%r9,4),%xmm1 + .byte 196,227,117,13,202,12 // vblendpd $0xc,%ymm2,%ymm1,%ymm1 + .byte 196,193,116,20,210 // vunpcklps %ymm10,%ymm1,%ymm2 + .byte 196,193,116,21,218 // vunpckhps %ymm10,%ymm1,%ymm3 + .byte 197,180,20,200 // vunpcklps %ymm0,%ymm9,%ymm1 + .byte 197,52,21,192 // vunpckhps %ymm0,%ymm9,%ymm8 + .byte 197,237,20,193 // vunpcklpd %ymm1,%ymm2,%ymm0 + .byte 197,237,21,201 // vunpckhpd %ymm1,%ymm2,%ymm1 + .byte 196,193,101,20,208 // vunpcklpd %ymm8,%ymm3,%ymm2 + .byte 196,193,101,21,216 // vunpckhpd %ymm8,%ymm3,%ymm3 + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 255,224 // jmpq *%rax + .byte 144 // nop + .byte 132,255 // test %bh,%bh + .byte 255 // (bad) + .byte 255,203 // dec %ebx + .byte 255 // (bad) + .byte 255 // (bad) + .byte 255 // (bad) + .byte 190,255,255,255,177 // mov $0xb1ffffff,%esi + .byte 255 // (bad) + .byte 255 // (bad) + .byte 255,164,255,255,255,156,255 // jmpq *-0x630001(%rdi,%rdi,8) + .byte 255 // (bad) + .byte 255,148,255,255,255,140,255 // callq *-0x730001(%rdi,%rdi,8) + .byte 255 // (bad) + .byte 255 // .byte 0xff + HIDDEN _sk_store_f32_hsw .globl _sk_store_f32_hsw _sk_store_f32_hsw: @@ -6205,7 +6274,7 @@ _sk_store_f32_hsw: .byte 196,65,37,20,196 // vunpcklpd %ymm12,%ymm11,%ymm8 .byte 196,65,37,21,220 // vunpckhpd %ymm12,%ymm11,%ymm11 .byte 72,133,201 // test %rcx,%rcx - .byte 117,55 // jne 1d16 <_sk_store_f32_hsw+0x6d> + .byte 117,55 // jne 1dd5 <_sk_store_f32_hsw+0x6d> .byte 196,67,45,24,225,1 // vinsertf128 $0x1,%xmm9,%ymm10,%ymm12 .byte 196,67,61,24,235,1 // vinsertf128 $0x1,%xmm11,%ymm8,%ymm13 .byte 196,67,45,6,201,49 // vperm2f128 $0x31,%ymm9,%ymm10,%ymm9 @@ -6218,22 +6287,22 @@ _sk_store_f32_hsw: .byte 255,224 // jmpq *%rax .byte 196,65,121,17,20,128 // vmovupd %xmm10,(%r8,%rax,4) .byte 72,131,249,1 // cmp $0x1,%rcx - .byte 116,240 // je 1d12 <_sk_store_f32_hsw+0x69> + .byte 116,240 // je 1dd1 <_sk_store_f32_hsw+0x69> .byte 196,65,121,17,76,128,16 // vmovupd %xmm9,0x10(%r8,%rax,4) .byte 72,131,249,3 // cmp $0x3,%rcx - .byte 114,227 // jb 1d12 <_sk_store_f32_hsw+0x69> + .byte 114,227 // jb 1dd1 <_sk_store_f32_hsw+0x69> .byte 196,65,121,17,68,128,32 // vmovupd %xmm8,0x20(%r8,%rax,4) - .byte 116,218 // je 1d12 <_sk_store_f32_hsw+0x69> + .byte 116,218 // je 1dd1 <_sk_store_f32_hsw+0x69> .byte 196,65,121,17,92,128,48 // vmovupd %xmm11,0x30(%r8,%rax,4) .byte 72,131,249,5 // cmp $0x5,%rcx - .byte 114,205 // jb 1d12 <_sk_store_f32_hsw+0x69> + .byte 114,205 // jb 1dd1 <_sk_store_f32_hsw+0x69> .byte 196,67,125,25,84,128,64,1 // vextractf128 $0x1,%ymm10,0x40(%r8,%rax,4) - .byte 116,195 // je 1d12 <_sk_store_f32_hsw+0x69> + .byte 116,195 // je 1dd1 <_sk_store_f32_hsw+0x69> .byte 196,67,125,25,76,128,80,1 // vextractf128 $0x1,%ymm9,0x50(%r8,%rax,4) .byte 72,131,249,7 // cmp $0x7,%rcx - .byte 114,181 // jb 1d12 <_sk_store_f32_hsw+0x69> + .byte 114,181 // jb 1dd1 <_sk_store_f32_hsw+0x69> .byte 196,67,125,25,68,128,96,1 // vextractf128 $0x1,%ymm8,0x60(%r8,%rax,4) - .byte 235,171 // jmp 1d12 <_sk_store_f32_hsw+0x69> + .byte 235,171 // jmp 1dd1 <_sk_store_f32_hsw+0x69> HIDDEN _sk_clamp_x_hsw .globl _sk_clamp_x_hsw @@ -9016,6 +9085,57 @@ _sk_store_u16_be_avx: .byte 196,65,121,214,68,248,48 // vmovq %xmm8,0x30(%r8,%rdi,8) .byte 235,174 // jmp 255a <_sk_store_u16_be_avx+0xf6> +HIDDEN _sk_load_f32_avx +.globl _sk_load_f32_avx +_sk_load_f32_avx: + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 72,131,249,7 // cmp $0x7,%rcx + .byte 119,110 // ja 2622 <_sk_load_f32_avx+0x76> + .byte 76,139,0 // mov (%rax),%r8 + .byte 76,141,12,189,0,0,0,0 // lea 0x0(,%rdi,4),%r9 + .byte 76,141,21,134,0,0,0 // lea 0x86(%rip),%r10 # 264c <_sk_load_f32_avx+0xa0> + .byte 73,99,4,138 // movslq (%r10,%rcx,4),%rax + .byte 76,1,208 // add %r10,%rax + .byte 255,224 // jmpq *%rax + .byte 196,3,125,24,68,136,112,1 // vinsertf128 $0x1,0x70(%r8,%r9,4),%ymm0,%ymm8 + .byte 196,131,125,24,92,136,96,1 // vinsertf128 $0x1,0x60(%r8,%r9,4),%ymm0,%ymm3 + .byte 196,131,125,24,76,136,80,1 // vinsertf128 $0x1,0x50(%r8,%r9,4),%ymm0,%ymm1 + .byte 196,131,125,24,84,136,64,1 // vinsertf128 $0x1,0x40(%r8,%r9,4),%ymm0,%ymm2 + .byte 196,129,121,16,68,136,48 // vmovupd 0x30(%r8,%r9,4),%xmm0 + .byte 196,195,125,13,192,12 // vblendpd $0xc,%ymm8,%ymm0,%ymm0 + .byte 196,1,121,16,68,136,32 // vmovupd 0x20(%r8,%r9,4),%xmm8 + .byte 196,99,61,13,203,12 // vblendpd $0xc,%ymm3,%ymm8,%ymm9 + .byte 196,129,121,16,92,136,16 // vmovupd 0x10(%r8,%r9,4),%xmm3 + .byte 196,99,101,13,209,12 // vblendpd $0xc,%ymm1,%ymm3,%ymm10 + .byte 196,129,121,16,12,136 // vmovupd (%r8,%r9,4),%xmm1 + .byte 196,227,117,13,202,12 // vblendpd $0xc,%ymm2,%ymm1,%ymm1 + .byte 196,193,116,20,210 // vunpcklps %ymm10,%ymm1,%ymm2 + .byte 196,193,116,21,218 // vunpckhps %ymm10,%ymm1,%ymm3 + .byte 197,180,20,200 // vunpcklps %ymm0,%ymm9,%ymm1 + .byte 197,52,21,192 // vunpckhps %ymm0,%ymm9,%ymm8 + .byte 197,237,20,193 // vunpcklpd %ymm1,%ymm2,%ymm0 + .byte 197,237,21,201 // vunpckhpd %ymm1,%ymm2,%ymm1 + .byte 196,193,101,20,208 // vunpcklpd %ymm8,%ymm3,%ymm2 + .byte 196,193,101,21,216 // vunpckhpd %ymm8,%ymm3,%ymm3 + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 255,224 // jmpq *%rax + .byte 102,144 // xchg %ax,%ax + .byte 131,255,255 // cmp $0xffffffff,%edi + .byte 255,202 // dec %edx + .byte 255 // (bad) + .byte 255 // (bad) + .byte 255 // (bad) + .byte 189,255,255,255,176 // mov $0xb0ffffff,%ebp + .byte 255 // (bad) + .byte 255 // (bad) + .byte 255,163,255,255,255,155 // jmpq *-0x64000001(%rbx) + .byte 255 // (bad) + .byte 255 // (bad) + .byte 255,147,255,255,255,139 // callq *-0x74000001(%rbx) + .byte 255 // (bad) + .byte 255 // (bad) + .byte 255 // .byte 0xff + HIDDEN _sk_store_f32_avx .globl _sk_store_f32_avx _sk_store_f32_avx: @@ -9031,7 +9151,7 @@ _sk_store_f32_avx: .byte 196,65,37,20,196 // vunpcklpd %ymm12,%ymm11,%ymm8 .byte 196,65,37,21,220 // vunpckhpd %ymm12,%ymm11,%ymm11 .byte 72,133,201 // test %rcx,%rcx - .byte 117,55 // jne 2619 <_sk_store_f32_avx+0x6d> + .byte 117,55 // jne 26d9 <_sk_store_f32_avx+0x6d> .byte 196,67,45,24,225,1 // vinsertf128 $0x1,%xmm9,%ymm10,%ymm12 .byte 196,67,61,24,235,1 // vinsertf128 $0x1,%xmm11,%ymm8,%ymm13 .byte 196,67,45,6,201,49 // vperm2f128 $0x31,%ymm9,%ymm10,%ymm9 @@ -9044,22 +9164,22 @@ _sk_store_f32_avx: .byte 255,224 // jmpq *%rax .byte 196,65,121,17,20,128 // vmovupd %xmm10,(%r8,%rax,4) .byte 72,131,249,1 // cmp $0x1,%rcx - .byte 116,240 // je 2615 <_sk_store_f32_avx+0x69> + .byte 116,240 // je 26d5 <_sk_store_f32_avx+0x69> .byte 196,65,121,17,76,128,16 // vmovupd %xmm9,0x10(%r8,%rax,4) .byte 72,131,249,3 // cmp $0x3,%rcx - .byte 114,227 // jb 2615 <_sk_store_f32_avx+0x69> + .byte 114,227 // jb 26d5 <_sk_store_f32_avx+0x69> .byte 196,65,121,17,68,128,32 // vmovupd %xmm8,0x20(%r8,%rax,4) - .byte 116,218 // je 2615 <_sk_store_f32_avx+0x69> + .byte 116,218 // je 26d5 <_sk_store_f32_avx+0x69> .byte 196,65,121,17,92,128,48 // vmovupd %xmm11,0x30(%r8,%rax,4) .byte 72,131,249,5 // cmp $0x5,%rcx - .byte 114,205 // jb 2615 <_sk_store_f32_avx+0x69> + .byte 114,205 // jb 26d5 <_sk_store_f32_avx+0x69> .byte 196,67,125,25,84,128,64,1 // vextractf128 $0x1,%ymm10,0x40(%r8,%rax,4) - .byte 116,195 // je 2615 <_sk_store_f32_avx+0x69> + .byte 116,195 // je 26d5 <_sk_store_f32_avx+0x69> .byte 196,67,125,25,76,128,80,1 // vextractf128 $0x1,%ymm9,0x50(%r8,%rax,4) .byte 72,131,249,7 // cmp $0x7,%rcx - .byte 114,181 // jb 2615 <_sk_store_f32_avx+0x69> + .byte 114,181 // jb 26d5 <_sk_store_f32_avx+0x69> .byte 196,67,125,25,68,128,96,1 // vextractf128 $0x1,%ymm8,0x60(%r8,%rax,4) - .byte 235,171 // jmp 2615 <_sk_store_f32_avx+0x69> + .byte 235,171 // jmp 26d5 <_sk_store_f32_avx+0x69> HIDDEN _sk_clamp_x_avx .globl _sk_clamp_x_avx @@ -11340,6 +11460,32 @@ _sk_store_u16_be_sse41: .byte 72,173 // lods %ds:(%rsi),%rax .byte 255,224 // jmpq *%rax +HIDDEN _sk_load_f32_sse41 +.globl _sk_load_f32_sse41 +_sk_load_f32_sse41: + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 72,139,0 // mov (%rax),%rax + .byte 72,137,249 // mov %rdi,%rcx + .byte 72,193,225,4 // shl $0x4,%rcx + .byte 68,15,16,4,8 // movups (%rax,%rcx,1),%xmm8 + .byte 15,16,68,8,16 // movups 0x10(%rax,%rcx,1),%xmm0 + .byte 15,16,92,8,32 // movups 0x20(%rax,%rcx,1),%xmm3 + .byte 68,15,16,76,8,48 // movups 0x30(%rax,%rcx,1),%xmm9 + .byte 65,15,40,208 // movaps %xmm8,%xmm2 + .byte 15,20,208 // unpcklps %xmm0,%xmm2 + .byte 15,40,203 // movaps %xmm3,%xmm1 + .byte 65,15,20,201 // unpcklps %xmm9,%xmm1 + .byte 68,15,21,192 // unpckhps %xmm0,%xmm8 + .byte 65,15,21,217 // unpckhps %xmm9,%xmm3 + .byte 15,40,194 // movaps %xmm2,%xmm0 + .byte 102,15,20,193 // unpcklpd %xmm1,%xmm0 + .byte 15,18,202 // movhlps %xmm2,%xmm1 + .byte 65,15,40,208 // movaps %xmm8,%xmm2 + .byte 102,15,20,211 // unpcklpd %xmm3,%xmm2 + .byte 65,15,18,216 // movhlps %xmm8,%xmm3 + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 255,224 // jmpq *%rax + HIDDEN _sk_store_f32_sse41 .globl _sk_store_f32_sse41 _sk_store_f32_sse41: @@ -13771,6 +13917,32 @@ _sk_store_u16_be_sse2: .byte 72,173 // lods %ds:(%rsi),%rax .byte 255,224 // jmpq *%rax +HIDDEN _sk_load_f32_sse2 +.globl _sk_load_f32_sse2 +_sk_load_f32_sse2: + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 72,139,0 // mov (%rax),%rax + .byte 72,137,249 // mov %rdi,%rcx + .byte 72,193,225,4 // shl $0x4,%rcx + .byte 68,15,16,4,8 // movups (%rax,%rcx,1),%xmm8 + .byte 15,16,68,8,16 // movups 0x10(%rax,%rcx,1),%xmm0 + .byte 15,16,92,8,32 // movups 0x20(%rax,%rcx,1),%xmm3 + .byte 68,15,16,76,8,48 // movups 0x30(%rax,%rcx,1),%xmm9 + .byte 65,15,40,208 // movaps %xmm8,%xmm2 + .byte 15,20,208 // unpcklps %xmm0,%xmm2 + .byte 15,40,203 // movaps %xmm3,%xmm1 + .byte 65,15,20,201 // unpcklps %xmm9,%xmm1 + .byte 68,15,21,192 // unpckhps %xmm0,%xmm8 + .byte 65,15,21,217 // unpckhps %xmm9,%xmm3 + .byte 15,40,194 // movaps %xmm2,%xmm0 + .byte 102,15,20,193 // unpcklpd %xmm1,%xmm0 + .byte 15,18,202 // movhlps %xmm2,%xmm1 + .byte 65,15,40,208 // movaps %xmm8,%xmm2 + .byte 102,15,20,211 // unpcklpd %xmm3,%xmm2 + .byte 65,15,18,216 // movhlps %xmm8,%xmm3 + .byte 72,173 // lods %ds:(%rsi),%rax + .byte 255,224 // jmpq *%rax + HIDDEN _sk_store_f32_sse2 .globl _sk_store_f32_sse2 _sk_store_f32_sse2: diff --git a/src/jumper/SkJumper_generated_win.S b/src/jumper/SkJumper_generated_win.S index 4ec22db65b..2fa69f5c8b 100644 --- a/src/jumper/SkJumper_generated_win.S +++ b/src/jumper/SkJumper_generated_win.S @@ -1607,7 +1607,7 @@ _sk_load_4444_hsw LABEL PROC DB 255 ; (bad) DB 255 ; (bad) DB 255 ; (bad) - DB 233,255,255,255,225 ; jmpq ffffffffe20016ec <_sk_linear_gradient_2stops_hsw+0xffffffffe1fff563> + DB 233,255,255,255,225 ; jmpq ffffffffe20016ec <_sk_linear_gradient_2stops_hsw+0xffffffffe1fff4a4> DB 255 ; (bad) DB 255 ; (bad) DB 255 ; (bad) @@ -2022,6 +2022,55 @@ _sk_store_u16_be_hsw LABEL PROC DB 196,65,121,214,68,248,48 ; vmovq %xmm8,0x30(%r8,%rdi,8) DB 235,174 ; jmp 1cf3 <_sk_store_u16_be_hsw+0xef> +PUBLIC _sk_load_f32_hsw +_sk_load_f32_hsw LABEL PROC + DB 72,173 ; lods %ds:(%rsi),%rax + DB 72,131,249,7 ; cmp $0x7,%rcx + DB 119,110 ; ja 1dbb <_sk_load_f32_hsw+0x76> + DB 76,139,0 ; mov (%rax),%r8 + DB 76,141,12,189,0,0,0,0 ; lea 0x0(,%rdi,4),%r9 + DB 76,141,21,133,0,0,0 ; lea 0x85(%rip),%r10 # 1de4 <_sk_load_f32_hsw+0x9f> + DB 73,99,4,138 ; movslq (%r10,%rcx,4),%rax + DB 76,1,208 ; add %r10,%rax + DB 255,224 ; jmpq *%rax + DB 196,3,125,24,68,136,112,1 ; vinsertf128 $0x1,0x70(%r8,%r9,4),%ymm0,%ymm8 + DB 196,131,125,24,92,136,96,1 ; vinsertf128 $0x1,0x60(%r8,%r9,4),%ymm0,%ymm3 + DB 196,131,125,24,76,136,80,1 ; vinsertf128 $0x1,0x50(%r8,%r9,4),%ymm0,%ymm1 + DB 196,131,125,24,84,136,64,1 ; vinsertf128 $0x1,0x40(%r8,%r9,4),%ymm0,%ymm2 + DB 196,129,121,16,68,136,48 ; vmovupd 0x30(%r8,%r9,4),%xmm0 + DB 196,195,125,13,192,12 ; vblendpd $0xc,%ymm8,%ymm0,%ymm0 + DB 196,1,121,16,68,136,32 ; vmovupd 0x20(%r8,%r9,4),%xmm8 + DB 196,99,61,13,203,12 ; vblendpd $0xc,%ymm3,%ymm8,%ymm9 + DB 196,129,121,16,92,136,16 ; vmovupd 0x10(%r8,%r9,4),%xmm3 + DB 196,99,101,13,209,12 ; vblendpd $0xc,%ymm1,%ymm3,%ymm10 + DB 196,129,121,16,12,136 ; vmovupd (%r8,%r9,4),%xmm1 + DB 196,227,117,13,202,12 ; vblendpd $0xc,%ymm2,%ymm1,%ymm1 + DB 196,193,116,20,210 ; vunpcklps %ymm10,%ymm1,%ymm2 + DB 196,193,116,21,218 ; vunpckhps %ymm10,%ymm1,%ymm3 + DB 197,180,20,200 ; vunpcklps %ymm0,%ymm9,%ymm1 + DB 197,52,21,192 ; vunpckhps %ymm0,%ymm9,%ymm8 + DB 197,237,20,193 ; vunpcklpd %ymm1,%ymm2,%ymm0 + DB 197,237,21,201 ; vunpckhpd %ymm1,%ymm2,%ymm1 + DB 196,193,101,20,208 ; vunpcklpd %ymm8,%ymm3,%ymm2 + DB 196,193,101,21,216 ; vunpckhpd %ymm8,%ymm3,%ymm3 + DB 72,173 ; lods %ds:(%rsi),%rax + DB 255,224 ; jmpq *%rax + DB 144 ; nop + DB 132,255 ; test %bh,%bh + DB 255 ; (bad) + DB 255,203 ; dec %ebx + DB 255 ; (bad) + DB 255 ; (bad) + DB 255 ; (bad) + DB 190,255,255,255,177 ; mov $0xb1ffffff,%esi + DB 255 ; (bad) + DB 255 ; (bad) + DB 255,164,255,255,255,156,255 ; jmpq *-0x630001(%rdi,%rdi,8) + DB 255 ; (bad) + DB 255,148,255,255,255,140,255 ; callq *-0x730001(%rdi,%rdi,8) + DB 255 ; (bad) + DB 255 ; .byte 0xff + PUBLIC _sk_store_f32_hsw _sk_store_f32_hsw LABEL PROC DB 72,173 ; lods %ds:(%rsi),%rax @@ -2036,7 +2085,7 @@ _sk_store_f32_hsw LABEL PROC DB 196,65,37,20,196 ; vunpcklpd %ymm12,%ymm11,%ymm8 DB 196,65,37,21,220 ; vunpckhpd %ymm12,%ymm11,%ymm11 DB 72,133,201 ; test %rcx,%rcx - DB 117,55 ; jne 1db2 <_sk_store_f32_hsw+0x6d> + DB 117,55 ; jne 1e71 <_sk_store_f32_hsw+0x6d> DB 196,67,45,24,225,1 ; vinsertf128 $0x1,%xmm9,%ymm10,%ymm12 DB 196,67,61,24,235,1 ; vinsertf128 $0x1,%xmm11,%ymm8,%ymm13 DB 196,67,45,6,201,49 ; vperm2f128 $0x31,%ymm9,%ymm10,%ymm9 @@ -2049,22 +2098,22 @@ _sk_store_f32_hsw LABEL PROC DB 255,224 ; jmpq *%rax DB 196,65,121,17,20,128 ; vmovupd %xmm10,(%r8,%rax,4) DB 72,131,249,1 ; cmp $0x1,%rcx - DB 116,240 ; je 1dae <_sk_store_f32_hsw+0x69> + DB 116,240 ; je 1e6d <_sk_store_f32_hsw+0x69> DB 196,65,121,17,76,128,16 ; vmovupd %xmm9,0x10(%r8,%rax,4) DB 72,131,249,3 ; cmp $0x3,%rcx - DB 114,227 ; jb 1dae <_sk_store_f32_hsw+0x69> + DB 114,227 ; jb 1e6d <_sk_store_f32_hsw+0x69> DB 196,65,121,17,68,128,32 ; vmovupd %xmm8,0x20(%r8,%rax,4) - DB 116,218 ; je 1dae <_sk_store_f32_hsw+0x69> + DB 116,218 ; je 1e6d <_sk_store_f32_hsw+0x69> DB 196,65,121,17,92,128,48 ; vmovupd %xmm11,0x30(%r8,%rax,4) DB 72,131,249,5 ; cmp $0x5,%rcx - DB 114,205 ; jb 1dae <_sk_store_f32_hsw+0x69> + DB 114,205 ; jb 1e6d <_sk_store_f32_hsw+0x69> DB 196,67,125,25,84,128,64,1 ; vextractf128 $0x1,%ymm10,0x40(%r8,%rax,4) - DB 116,195 ; je 1dae <_sk_store_f32_hsw+0x69> + DB 116,195 ; je 1e6d <_sk_store_f32_hsw+0x69> DB 196,67,125,25,76,128,80,1 ; vextractf128 $0x1,%ymm9,0x50(%r8,%rax,4) DB 72,131,249,7 ; cmp $0x7,%rcx - DB 114,181 ; jb 1dae <_sk_store_f32_hsw+0x69> + DB 114,181 ; jb 1e6d <_sk_store_f32_hsw+0x69> DB 196,67,125,25,68,128,96,1 ; vextractf128 $0x1,%ymm8,0x60(%r8,%rax,4) - DB 235,171 ; jmp 1dae <_sk_store_f32_hsw+0x69> + DB 235,171 ; jmp 1e6d <_sk_store_f32_hsw+0x69> PUBLIC _sk_clamp_x_hsw _sk_clamp_x_hsw LABEL PROC @@ -4804,6 +4853,56 @@ _sk_store_u16_be_avx LABEL PROC DB 196,65,121,214,68,248,48 ; vmovq %xmm8,0x30(%r8,%rdi,8) DB 235,174 ; jmp 25f6 <_sk_store_u16_be_avx+0xf6> +PUBLIC _sk_load_f32_avx +_sk_load_f32_avx LABEL PROC + DB 72,173 ; lods %ds:(%rsi),%rax + DB 72,131,249,7 ; cmp $0x7,%rcx + DB 119,110 ; ja 26be <_sk_load_f32_avx+0x76> + DB 76,139,0 ; mov (%rax),%r8 + DB 76,141,12,189,0,0,0,0 ; lea 0x0(,%rdi,4),%r9 + DB 76,141,21,134,0,0,0 ; lea 0x86(%rip),%r10 # 26e8 <_sk_load_f32_avx+0xa0> + DB 73,99,4,138 ; movslq (%r10,%rcx,4),%rax + DB 76,1,208 ; add %r10,%rax + DB 255,224 ; jmpq *%rax + DB 196,3,125,24,68,136,112,1 ; vinsertf128 $0x1,0x70(%r8,%r9,4),%ymm0,%ymm8 + DB 196,131,125,24,92,136,96,1 ; vinsertf128 $0x1,0x60(%r8,%r9,4),%ymm0,%ymm3 + DB 196,131,125,24,76,136,80,1 ; vinsertf128 $0x1,0x50(%r8,%r9,4),%ymm0,%ymm1 + DB 196,131,125,24,84,136,64,1 ; vinsertf128 $0x1,0x40(%r8,%r9,4),%ymm0,%ymm2 + DB 196,129,121,16,68,136,48 ; vmovupd 0x30(%r8,%r9,4),%xmm0 + DB 196,195,125,13,192,12 ; vblendpd $0xc,%ymm8,%ymm0,%ymm0 + DB 196,1,121,16,68,136,32 ; vmovupd 0x20(%r8,%r9,4),%xmm8 + DB 196,99,61,13,203,12 ; vblendpd $0xc,%ymm3,%ymm8,%ymm9 + DB 196,129,121,16,92,136,16 ; vmovupd 0x10(%r8,%r9,4),%xmm3 + DB 196,99,101,13,209,12 ; vblendpd $0xc,%ymm1,%ymm3,%ymm10 + DB 196,129,121,16,12,136 ; vmovupd (%r8,%r9,4),%xmm1 + DB 196,227,117,13,202,12 ; vblendpd $0xc,%ymm2,%ymm1,%ymm1 + DB 196,193,116,20,210 ; vunpcklps %ymm10,%ymm1,%ymm2 + DB 196,193,116,21,218 ; vunpckhps %ymm10,%ymm1,%ymm3 + DB 197,180,20,200 ; vunpcklps %ymm0,%ymm9,%ymm1 + DB 197,52,21,192 ; vunpckhps %ymm0,%ymm9,%ymm8 + DB 197,237,20,193 ; vunpcklpd %ymm1,%ymm2,%ymm0 + DB 197,237,21,201 ; vunpckhpd %ymm1,%ymm2,%ymm1 + DB 196,193,101,20,208 ; vunpcklpd %ymm8,%ymm3,%ymm2 + DB 196,193,101,21,216 ; vunpckhpd %ymm8,%ymm3,%ymm3 + DB 72,173 ; lods %ds:(%rsi),%rax + DB 255,224 ; jmpq *%rax + DB 102,144 ; xchg %ax,%ax + DB 131,255,255 ; cmp $0xffffffff,%edi + DB 255,202 ; dec %edx + DB 255 ; (bad) + DB 255 ; (bad) + DB 255 ; (bad) + DB 189,255,255,255,176 ; mov $0xb0ffffff,%ebp + DB 255 ; (bad) + DB 255 ; (bad) + DB 255,163,255,255,255,155 ; jmpq *-0x64000001(%rbx) + DB 255 ; (bad) + DB 255 ; (bad) + DB 255,147,255,255,255,139 ; callq *-0x74000001(%rbx) + DB 255 ; (bad) + DB 255 ; (bad) + DB 255 ; .byte 0xff + PUBLIC _sk_store_f32_avx _sk_store_f32_avx LABEL PROC DB 72,173 ; lods %ds:(%rsi),%rax @@ -4818,7 +4917,7 @@ _sk_store_f32_avx LABEL PROC DB 196,65,37,20,196 ; vunpcklpd %ymm12,%ymm11,%ymm8 DB 196,65,37,21,220 ; vunpckhpd %ymm12,%ymm11,%ymm11 DB 72,133,201 ; test %rcx,%rcx - DB 117,55 ; jne 26b5 <_sk_store_f32_avx+0x6d> + DB 117,55 ; jne 2775 <_sk_store_f32_avx+0x6d> DB 196,67,45,24,225,1 ; vinsertf128 $0x1,%xmm9,%ymm10,%ymm12 DB 196,67,61,24,235,1 ; vinsertf128 $0x1,%xmm11,%ymm8,%ymm13 DB 196,67,45,6,201,49 ; vperm2f128 $0x31,%ymm9,%ymm10,%ymm9 @@ -4831,22 +4930,22 @@ _sk_store_f32_avx LABEL PROC DB 255,224 ; jmpq *%rax DB 196,65,121,17,20,128 ; vmovupd %xmm10,(%r8,%rax,4) DB 72,131,249,1 ; cmp $0x1,%rcx - DB 116,240 ; je 26b1 <_sk_store_f32_avx+0x69> + DB 116,240 ; je 2771 <_sk_store_f32_avx+0x69> DB 196,65,121,17,76,128,16 ; vmovupd %xmm9,0x10(%r8,%rax,4) DB 72,131,249,3 ; cmp $0x3,%rcx - DB 114,227 ; jb 26b1 <_sk_store_f32_avx+0x69> + DB 114,227 ; jb 2771 <_sk_store_f32_avx+0x69> DB 196,65,121,17,68,128,32 ; vmovupd %xmm8,0x20(%r8,%rax,4) - DB 116,218 ; je 26b1 <_sk_store_f32_avx+0x69> + DB 116,218 ; je 2771 <_sk_store_f32_avx+0x69> DB 196,65,121,17,92,128,48 ; vmovupd %xmm11,0x30(%r8,%rax,4) DB 72,131,249,5 ; cmp $0x5,%rcx - DB 114,205 ; jb 26b1 <_sk_store_f32_avx+0x69> + DB 114,205 ; jb 2771 <_sk_store_f32_avx+0x69> DB 196,67,125,25,84,128,64,1 ; vextractf128 $0x1,%ymm10,0x40(%r8,%rax,4) - DB 116,195 ; je 26b1 <_sk_store_f32_avx+0x69> + DB 116,195 ; je 2771 <_sk_store_f32_avx+0x69> DB 196,67,125,25,76,128,80,1 ; vextractf128 $0x1,%ymm9,0x50(%r8,%rax,4) DB 72,131,249,7 ; cmp $0x7,%rcx - DB 114,181 ; jb 26b1 <_sk_store_f32_avx+0x69> + DB 114,181 ; jb 2771 <_sk_store_f32_avx+0x69> DB 196,67,125,25,68,128,96,1 ; vextractf128 $0x1,%ymm8,0x60(%r8,%rax,4) - DB 235,171 ; jmp 26b1 <_sk_store_f32_avx+0x69> + DB 235,171 ; jmp 2771 <_sk_store_f32_avx+0x69> PUBLIC _sk_clamp_x_avx _sk_clamp_x_avx LABEL PROC @@ -7088,6 +7187,31 @@ _sk_store_u16_be_sse41 LABEL PROC DB 72,173 ; lods %ds:(%rsi),%rax DB 255,224 ; jmpq *%rax +PUBLIC _sk_load_f32_sse41 +_sk_load_f32_sse41 LABEL PROC + DB 72,173 ; lods %ds:(%rsi),%rax + DB 72,139,0 ; mov (%rax),%rax + DB 72,137,249 ; mov %rdi,%rcx + DB 72,193,225,4 ; shl $0x4,%rcx + DB 68,15,16,4,8 ; movups (%rax,%rcx,1),%xmm8 + DB 15,16,68,8,16 ; movups 0x10(%rax,%rcx,1),%xmm0 + DB 15,16,92,8,32 ; movups 0x20(%rax,%rcx,1),%xmm3 + DB 68,15,16,76,8,48 ; movups 0x30(%rax,%rcx,1),%xmm9 + DB 65,15,40,208 ; movaps %xmm8,%xmm2 + DB 15,20,208 ; unpcklps %xmm0,%xmm2 + DB 15,40,203 ; movaps %xmm3,%xmm1 + DB 65,15,20,201 ; unpcklps %xmm9,%xmm1 + DB 68,15,21,192 ; unpckhps %xmm0,%xmm8 + DB 65,15,21,217 ; unpckhps %xmm9,%xmm3 + DB 15,40,194 ; movaps %xmm2,%xmm0 + DB 102,15,20,193 ; unpcklpd %xmm1,%xmm0 + DB 15,18,202 ; movhlps %xmm2,%xmm1 + DB 65,15,40,208 ; movaps %xmm8,%xmm2 + DB 102,15,20,211 ; unpcklpd %xmm3,%xmm2 + DB 65,15,18,216 ; movhlps %xmm8,%xmm3 + DB 72,173 ; lods %ds:(%rsi),%rax + DB 255,224 ; jmpq *%rax + PUBLIC _sk_store_f32_sse41 _sk_store_f32_sse41 LABEL PROC DB 72,173 ; lods %ds:(%rsi),%rax @@ -9477,6 +9601,31 @@ _sk_store_u16_be_sse2 LABEL PROC DB 72,173 ; lods %ds:(%rsi),%rax DB 255,224 ; jmpq *%rax +PUBLIC _sk_load_f32_sse2 +_sk_load_f32_sse2 LABEL PROC + DB 72,173 ; lods %ds:(%rsi),%rax + DB 72,139,0 ; mov (%rax),%rax + DB 72,137,249 ; mov %rdi,%rcx + DB 72,193,225,4 ; shl $0x4,%rcx + DB 68,15,16,4,8 ; movups (%rax,%rcx,1),%xmm8 + DB 15,16,68,8,16 ; movups 0x10(%rax,%rcx,1),%xmm0 + DB 15,16,92,8,32 ; movups 0x20(%rax,%rcx,1),%xmm3 + DB 68,15,16,76,8,48 ; movups 0x30(%rax,%rcx,1),%xmm9 + DB 65,15,40,208 ; movaps %xmm8,%xmm2 + DB 15,20,208 ; unpcklps %xmm0,%xmm2 + DB 15,40,203 ; movaps %xmm3,%xmm1 + DB 65,15,20,201 ; unpcklps %xmm9,%xmm1 + DB 68,15,21,192 ; unpckhps %xmm0,%xmm8 + DB 65,15,21,217 ; unpckhps %xmm9,%xmm3 + DB 15,40,194 ; movaps %xmm2,%xmm0 + DB 102,15,20,193 ; unpcklpd %xmm1,%xmm0 + DB 15,18,202 ; movhlps %xmm2,%xmm1 + DB 65,15,40,208 ; movaps %xmm8,%xmm2 + DB 102,15,20,211 ; unpcklpd %xmm3,%xmm2 + DB 65,15,18,216 ; movhlps %xmm8,%xmm3 + DB 72,173 ; lods %ds:(%rsi),%rax + DB 255,224 ; jmpq *%rax + PUBLIC _sk_store_f32_sse2 _sk_store_f32_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 a8d5a29bb9..6e0c908593 100644 --- a/src/jumper/SkJumper_stages.cpp +++ b/src/jumper/SkJumper_stages.cpp @@ -673,6 +673,10 @@ STAGE(store_u16_be) { store4((uint16_t*)ptr,tail, R,G,B,A); } +STAGE(load_f32) { + auto ptr = *(const float**)ctx + 4*x; + load4(ptr,tail, &r,&g,&b,&a); +} STAGE(store_f32) { auto ptr = *(float**)ctx + 4*x; store4(ptr,tail, r,g,b,a); diff --git a/src/jumper/SkJumper_vectors.h b/src/jumper/SkJumper_vectors.h index a829e7241c..01c0ccf1fd 100644 --- a/src/jumper/SkJumper_vectors.h +++ b/src/jumper/SkJumper_vectors.h @@ -53,6 +53,13 @@ ptr[2] = b; ptr[3] = a; } + + SI void load4(const float* ptr, size_t tail, F* r, F* g, F* b, F* a) { + *r = ptr[0]; + *g = ptr[1]; + *b = ptr[2]; + *a = ptr[3]; + } SI void store4(float* ptr, size_t tail, F r, F g, F b, F a) { ptr[0] = r; ptr[1] = g; @@ -106,6 +113,14 @@ SI void store4(uint16_t* ptr, size_t tail, U16 r, U16 g, U16 b, U16 a) { vst4_u16(ptr, (uint16x4x4_t{{r,g,b,a}})); } + + SI void load4(const float* ptr, size_t tail, F* r, F* g, F* b, F* a) { + float32x4x4_t rgba = vld4q_f32(ptr); + *r = rgba.val[0]; + *g = rgba.val[1]; + *b = rgba.val[2]; + *a = rgba.val[3]; + } SI void store4(float* ptr, size_t tail, F r, F g, F b, F a) { vst4q_f32(ptr, (float32x4x4_t{{r,g,b,a}})); } @@ -164,6 +179,14 @@ vst4_lane_u16(ptr + 0, rgba, 0); vst4_lane_u16(ptr + 4, rgba, 1); } + + SI void load4(const float* ptr, size_t tail, F* r, F* g, F* b, F* a) { + float32x2x4_t rgba = vld4_f32(ptr); + *r = rgba.val[0]; + *g = rgba.val[1]; + *b = rgba.val[2]; + *a = rgba.val[3]; + } SI void store4(float* ptr, size_t tail, F r, F g, F b, F a) { vst4_f32(ptr, (float32x2x4_t{{r,g,b,a}})); } @@ -285,6 +308,31 @@ _mm_storeu_si128((__m128i*)ptr + 3, _67); } } + + SI void load4(const float* ptr, size_t tail, F* r, F* g, F* b, F* a) { + F _04, _15, _26, _37; + + switch (tail) { + case 0: _37 = _mm256_insertf128_ps(_37, _mm_loadu_ps(ptr+28), 1); + case 7: _26 = _mm256_insertf128_ps(_26, _mm_loadu_ps(ptr+24), 1); + case 6: _15 = _mm256_insertf128_ps(_15, _mm_loadu_ps(ptr+20), 1); + case 5: _04 = _mm256_insertf128_ps(_04, _mm_loadu_ps(ptr+16), 1); + case 4: _37 = _mm256_insertf128_ps(_37, _mm_loadu_ps(ptr+12), 0); + case 3: _26 = _mm256_insertf128_ps(_26, _mm_loadu_ps(ptr+ 8), 0); + case 2: _15 = _mm256_insertf128_ps(_15, _mm_loadu_ps(ptr+ 4), 0); + case 1: _04 = _mm256_insertf128_ps(_04, _mm_loadu_ps(ptr+ 0), 0); + } + + F rg0145 = _mm256_unpacklo_ps(_04,_15), // r0 r1 g0 g1 | r4 r5 g4 g5 + ba0145 = _mm256_unpackhi_ps(_04,_15), + rg2367 = _mm256_unpacklo_ps(_26,_37), + ba2367 = _mm256_unpackhi_ps(_26,_37); + + *r = _mm256_unpacklo_pd(rg0145, rg2367); + *g = _mm256_unpackhi_pd(rg0145, rg2367); + *b = _mm256_unpacklo_pd(ba0145, ba2367); + *a = _mm256_unpackhi_pd(ba0145, ba2367); + } SI void store4(float* ptr, size_t tail, F r, F g, F b, F a) { F rg0145 = _mm256_unpacklo_ps(r, g), // r0 g0 r1 g1 | r4 g4 r5 g5 rg2367 = _mm256_unpackhi_ps(r, g), // r2 ... | r6 ... @@ -408,6 +456,18 @@ _mm_storeu_si128((__m128i*)ptr + 0, _mm_unpacklo_epi32(rg, ba)); _mm_storeu_si128((__m128i*)ptr + 1, _mm_unpackhi_epi32(rg, ba)); } + + SI void load4(const float* ptr, size_t tail, F* r, F* g, F* b, F* a) { + auto _0 = _mm_loadu_ps(ptr+ 0), + _1 = _mm_loadu_ps(ptr+ 4), + _2 = _mm_loadu_ps(ptr+ 8), + _3 = _mm_loadu_ps(ptr+12); + _MM_TRANSPOSE4_PS(_0,_1,_2,_3); + *r = _0; + *g = _1; + *b = _2; + *a = _3; + } SI void store4(float* ptr, size_t tail, F r, F g, F b, F a) { _MM_TRANSPOSE4_PS(r,g,b,a); _mm_storeu_ps(ptr+ 0, r); -- cgit v1.2.3