From 8a823faeba2da8a77740b0cd9eaf1acf473067b1 Mon Sep 17 00:00:00 2001 From: Mike Klein Date: Wed, 5 Apr 2017 17:29:26 -0400 Subject: jumper, kill off F4 Its alignment (sometimes 4, sometimes 16) has proven to be error-prone. This also means we don't really need LazyCtx::load(). I think I only had it there to make sure we were doing unaligned loads of F4; the better way is to just never declare the data as aligned... The generated code isn't quite as good, but I can live with it. Change-Id: I5d57a580ca12c94ca84a5e8b72a66cf8d0c829eb Reviewed-on: https://skia-review.googlesource.com/11406 Reviewed-by: Herb Derby Commit-Queue: Mike Klein --- src/jumper/SkJumper_generated_win.S | 124 ++++++++++++++++++------------------ 1 file changed, 62 insertions(+), 62 deletions(-) (limited to 'src/jumper/SkJumper_generated_win.S') diff --git a/src/jumper/SkJumper_generated_win.S b/src/jumper/SkJumper_generated_win.S index c4872ae7e6..4ec22db65b 100644 --- a/src/jumper/SkJumper_generated_win.S +++ b/src/jumper/SkJumper_generated_win.S @@ -5147,19 +5147,19 @@ _sk_linear_gradient_2stops_avx LABEL PROC DB 196,226,125,24,72,16 ; vbroadcastss 0x10(%rax),%ymm1 DB 196,226,125,24,16 ; vbroadcastss (%rax),%ymm2 DB 197,244,89,200 ; vmulps %ymm0,%ymm1,%ymm1 - DB 197,108,88,193 ; vaddps %ymm1,%ymm2,%ymm8 + DB 197,116,88,194 ; vaddps %ymm2,%ymm1,%ymm8 DB 196,226,125,24,72,20 ; vbroadcastss 0x14(%rax),%ymm1 DB 196,226,125,24,80,4 ; vbroadcastss 0x4(%rax),%ymm2 DB 197,244,89,200 ; vmulps %ymm0,%ymm1,%ymm1 - DB 197,236,88,201 ; vaddps %ymm1,%ymm2,%ymm1 + DB 197,244,88,202 ; vaddps %ymm2,%ymm1,%ymm1 DB 196,226,125,24,80,24 ; vbroadcastss 0x18(%rax),%ymm2 DB 196,226,125,24,88,8 ; vbroadcastss 0x8(%rax),%ymm3 DB 197,236,89,208 ; vmulps %ymm0,%ymm2,%ymm2 - DB 197,228,88,210 ; vaddps %ymm2,%ymm3,%ymm2 + DB 197,236,88,211 ; vaddps %ymm3,%ymm2,%ymm2 DB 196,226,125,24,88,28 ; vbroadcastss 0x1c(%rax),%ymm3 DB 196,98,125,24,72,12 ; vbroadcastss 0xc(%rax),%ymm9 DB 197,228,89,192 ; vmulps %ymm0,%ymm3,%ymm0 - DB 197,180,88,216 ; vaddps %ymm0,%ymm9,%ymm3 + DB 196,193,124,88,217 ; vaddps %ymm9,%ymm0,%ymm3 DB 72,173 ; lods %ds:(%rsi),%rax DB 197,124,41,192 ; vmovaps %ymm8,%ymm0 DB 255,224 ; jmpq *%rax @@ -5267,14 +5267,14 @@ _sk_seed_shader_sse41 LABEL PROC PUBLIC _sk_constant_color_sse41 _sk_constant_color_sse41 LABEL PROC DB 72,173 ; lods %ds:(%rsi),%rax - DB 15,16,24 ; movups (%rax),%xmm3 - DB 15,40,195 ; movaps %xmm3,%xmm0 + DB 243,15,16,0 ; movss (%rax),%xmm0 + DB 243,15,16,72,4 ; movss 0x4(%rax),%xmm1 DB 15,198,192,0 ; shufps $0x0,%xmm0,%xmm0 - DB 15,40,203 ; movaps %xmm3,%xmm1 - DB 15,198,201,85 ; shufps $0x55,%xmm1,%xmm1 - DB 15,40,211 ; movaps %xmm3,%xmm2 - DB 15,198,210,170 ; shufps $0xaa,%xmm2,%xmm2 - DB 15,198,219,255 ; shufps $0xff,%xmm3,%xmm3 + DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1 + DB 243,15,16,80,8 ; movss 0x8(%rax),%xmm2 + DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2 + DB 243,15,16,88,12 ; movss 0xc(%rax),%xmm3 + DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3 DB 72,173 ; lods %ds:(%rsi),%rax DB 255,224 ; jmpq *%rax @@ -7449,33 +7449,33 @@ _sk_matrix_perspective_sse41 LABEL PROC PUBLIC _sk_linear_gradient_2stops_sse41 _sk_linear_gradient_2stops_sse41 LABEL PROC + DB 68,15,40,192 ; movaps %xmm0,%xmm8 DB 72,173 ; lods %ds:(%rsi),%rax - DB 68,15,16,8 ; movups (%rax),%xmm9 - DB 15,16,88,16 ; movups 0x10(%rax),%xmm3 - DB 68,15,40,195 ; movaps %xmm3,%xmm8 - DB 69,15,198,192,0 ; shufps $0x0,%xmm8,%xmm8 - DB 65,15,40,201 ; movaps %xmm9,%xmm1 + DB 243,15,16,64,16 ; movss 0x10(%rax),%xmm0 + DB 15,198,192,0 ; shufps $0x0,%xmm0,%xmm0 + DB 243,15,16,8 ; movss (%rax),%xmm1 + DB 243,15,16,80,4 ; movss 0x4(%rax),%xmm2 DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1 - DB 68,15,89,192 ; mulps %xmm0,%xmm8 - DB 68,15,88,193 ; addps %xmm1,%xmm8 - DB 15,40,203 ; movaps %xmm3,%xmm1 - DB 15,198,201,85 ; shufps $0x55,%xmm1,%xmm1 - DB 65,15,40,209 ; movaps %xmm9,%xmm2 - DB 15,198,210,85 ; shufps $0x55,%xmm2,%xmm2 - DB 15,89,200 ; mulps %xmm0,%xmm1 + DB 65,15,89,192 ; mulps %xmm8,%xmm0 + DB 15,88,193 ; addps %xmm1,%xmm0 + DB 243,15,16,72,20 ; movss 0x14(%rax),%xmm1 + DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1 + DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2 + DB 65,15,89,200 ; mulps %xmm8,%xmm1 DB 15,88,202 ; addps %xmm2,%xmm1 - DB 15,40,211 ; movaps %xmm3,%xmm2 - DB 15,198,210,170 ; shufps $0xaa,%xmm2,%xmm2 - DB 69,15,40,209 ; movaps %xmm9,%xmm10 - DB 69,15,198,210,170 ; shufps $0xaa,%xmm10,%xmm10 - DB 15,89,208 ; mulps %xmm0,%xmm2 - DB 65,15,88,210 ; addps %xmm10,%xmm2 - DB 15,198,219,255 ; shufps $0xff,%xmm3,%xmm3 - DB 69,15,198,201,255 ; shufps $0xff,%xmm9,%xmm9 - DB 15,89,216 ; mulps %xmm0,%xmm3 + DB 243,15,16,80,24 ; movss 0x18(%rax),%xmm2 + DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2 + DB 243,15,16,88,8 ; movss 0x8(%rax),%xmm3 + DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3 + DB 65,15,89,208 ; mulps %xmm8,%xmm2 + DB 15,88,211 ; addps %xmm3,%xmm2 + DB 243,15,16,88,28 ; movss 0x1c(%rax),%xmm3 + DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3 + DB 243,68,15,16,72,12 ; movss 0xc(%rax),%xmm9 + DB 69,15,198,201,0 ; shufps $0x0,%xmm9,%xmm9 + DB 65,15,89,216 ; mulps %xmm8,%xmm3 DB 65,15,88,217 ; addps %xmm9,%xmm3 DB 72,173 ; lods %ds:(%rsi),%rax - DB 65,15,40,192 ; movaps %xmm8,%xmm0 DB 255,224 ; jmpq *%rax PUBLIC _sk_start_pipeline_sse2 @@ -7581,14 +7581,14 @@ _sk_seed_shader_sse2 LABEL PROC PUBLIC _sk_constant_color_sse2 _sk_constant_color_sse2 LABEL PROC DB 72,173 ; lods %ds:(%rsi),%rax - DB 15,16,24 ; movups (%rax),%xmm3 - DB 15,40,195 ; movaps %xmm3,%xmm0 + DB 243,15,16,0 ; movss (%rax),%xmm0 + DB 243,15,16,72,4 ; movss 0x4(%rax),%xmm1 DB 15,198,192,0 ; shufps $0x0,%xmm0,%xmm0 - DB 15,40,203 ; movaps %xmm3,%xmm1 - DB 15,198,201,85 ; shufps $0x55,%xmm1,%xmm1 - DB 15,40,211 ; movaps %xmm3,%xmm2 - DB 15,198,210,170 ; shufps $0xaa,%xmm2,%xmm2 - DB 15,198,219,255 ; shufps $0xff,%xmm3,%xmm3 + DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1 + DB 243,15,16,80,8 ; movss 0x8(%rax),%xmm2 + DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2 + DB 243,15,16,88,12 ; movss 0xc(%rax),%xmm3 + DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3 DB 72,173 ; lods %ds:(%rsi),%rax DB 255,224 ; jmpq *%rax @@ -9866,33 +9866,33 @@ _sk_matrix_perspective_sse2 LABEL PROC PUBLIC _sk_linear_gradient_2stops_sse2 _sk_linear_gradient_2stops_sse2 LABEL PROC + DB 68,15,40,192 ; movaps %xmm0,%xmm8 DB 72,173 ; lods %ds:(%rsi),%rax - DB 68,15,16,8 ; movups (%rax),%xmm9 - DB 15,16,88,16 ; movups 0x10(%rax),%xmm3 - DB 68,15,40,195 ; movaps %xmm3,%xmm8 - DB 69,15,198,192,0 ; shufps $0x0,%xmm8,%xmm8 - DB 65,15,40,201 ; movaps %xmm9,%xmm1 + DB 243,15,16,64,16 ; movss 0x10(%rax),%xmm0 + DB 15,198,192,0 ; shufps $0x0,%xmm0,%xmm0 + DB 243,15,16,8 ; movss (%rax),%xmm1 + DB 243,15,16,80,4 ; movss 0x4(%rax),%xmm2 DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1 - DB 68,15,89,192 ; mulps %xmm0,%xmm8 - DB 68,15,88,193 ; addps %xmm1,%xmm8 - DB 15,40,203 ; movaps %xmm3,%xmm1 - DB 15,198,201,85 ; shufps $0x55,%xmm1,%xmm1 - DB 65,15,40,209 ; movaps %xmm9,%xmm2 - DB 15,198,210,85 ; shufps $0x55,%xmm2,%xmm2 - DB 15,89,200 ; mulps %xmm0,%xmm1 + DB 65,15,89,192 ; mulps %xmm8,%xmm0 + DB 15,88,193 ; addps %xmm1,%xmm0 + DB 243,15,16,72,20 ; movss 0x14(%rax),%xmm1 + DB 15,198,201,0 ; shufps $0x0,%xmm1,%xmm1 + DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2 + DB 65,15,89,200 ; mulps %xmm8,%xmm1 DB 15,88,202 ; addps %xmm2,%xmm1 - DB 15,40,211 ; movaps %xmm3,%xmm2 - DB 15,198,210,170 ; shufps $0xaa,%xmm2,%xmm2 - DB 69,15,40,209 ; movaps %xmm9,%xmm10 - DB 69,15,198,210,170 ; shufps $0xaa,%xmm10,%xmm10 - DB 15,89,208 ; mulps %xmm0,%xmm2 - DB 65,15,88,210 ; addps %xmm10,%xmm2 - DB 15,198,219,255 ; shufps $0xff,%xmm3,%xmm3 - DB 69,15,198,201,255 ; shufps $0xff,%xmm9,%xmm9 - DB 15,89,216 ; mulps %xmm0,%xmm3 + DB 243,15,16,80,24 ; movss 0x18(%rax),%xmm2 + DB 15,198,210,0 ; shufps $0x0,%xmm2,%xmm2 + DB 243,15,16,88,8 ; movss 0x8(%rax),%xmm3 + DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3 + DB 65,15,89,208 ; mulps %xmm8,%xmm2 + DB 15,88,211 ; addps %xmm3,%xmm2 + DB 243,15,16,88,28 ; movss 0x1c(%rax),%xmm3 + DB 15,198,219,0 ; shufps $0x0,%xmm3,%xmm3 + DB 243,68,15,16,72,12 ; movss 0xc(%rax),%xmm9 + DB 69,15,198,201,0 ; shufps $0x0,%xmm9,%xmm9 + DB 65,15,89,216 ; mulps %xmm8,%xmm3 DB 65,15,88,217 ; addps %xmm9,%xmm3 DB 72,173 ; lods %ds:(%rsi),%rax - DB 65,15,40,192 ; movaps %xmm8,%xmm0 DB 255,224 ; jmpq *%rax ENDIF END -- cgit v1.2.3