SecBSD/ports
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
ports/multimedia/dav1d/patches/patch-src_x86_ipred16_avx512_asm

204 lines
6.4 KiB
Text
Raw Normal View History

As we discussed in the last meeting, we reset the ports tree and began from scratch, even though this change involves porting all the packages. Starting small and growing gradually, this approach will reduce build times and consequently lower energy consumption in a world affected by climate change. We will add new ports as users needs arise; ok h3artbl33d@
2024-05-26 03:08:12 +00:00
Index: src/x86/ipred16_avx512.asm
--- src/x86/ipred16_avx512.asm.orig
+++ src/x86/ipred16_avx512.asm
@@ -104,6 +104,7 @@ cglobal ipred_paeth_16bpc, 3, 7, 10, dst, stride, tl,
add wq, r6
jmp wq
.w4:
+ _CET_ENDBR
vpbroadcastq m4, [tlq+2] ; top
movsldup m7, [base+ipred_shuf]
lea r6, [strideq*3]
@@ -133,6 +134,7 @@ cglobal ipred_paeth_16bpc, 3, 7, 10, dst, stride, tl,
.w4_end:
RET
.w8:
+ _CET_ENDBR
vbroadcasti32x4 m4, [tlq+2]
movsldup m7, [base+ipred_shuf]
lea r6, [strideq*3]
@@ -152,6 +154,7 @@ cglobal ipred_paeth_16bpc, 3, 7, 10, dst, stride, tl,
jg .w8_loop
RET
.w16:
+ _CET_ENDBR
vbroadcasti32x8 m4, [tlq+2]
movsldup m7, [base+ipred_shuf]
psubw m5, m4, m3
@@ -168,6 +171,7 @@ cglobal ipred_paeth_16bpc, 3, 7, 10, dst, stride, tl,
jg .w16_loop
RET
.w32:
+ _CET_ENDBR
movu m4, [tlq+2]
psubw m5, m4, m3
pabsw m6, m5
@@ -181,6 +185,7 @@ cglobal ipred_paeth_16bpc, 3, 7, 10, dst, stride, tl,
jg .w32_loop
RET
.w64:
+ _CET_ENDBR
movu m4, [tlq+ 2]
movu m7, [tlq+66]
psubw m5, m4, m3
@@ -212,6 +217,7 @@ cglobal ipred_smooth_v_16bpc, 3, 7, 7, dst, stride, tl
lea stride3q, [strideq*3]
jmp wq
.w4:
+ _CET_ENDBR
vpbroadcastq m5, [tlq+2] ; top
movsldup m4, [ipred_shuf]
psubw m5, m6 ; top - bottom
@@ -239,6 +245,7 @@ cglobal ipred_smooth_v_16bpc, 3, 7, 7, dst, stride, tl
.end:
RET
.w8:
+ _CET_ENDBR
vbroadcasti32x4 m5, [tlq+2] ; top
movsldup m4, [ipred_shuf]
psubw m5, m6 ; top - bottom
@@ -256,6 +263,7 @@ cglobal ipred_smooth_v_16bpc, 3, 7, 7, dst, stride, tl
jl .w8_loop
RET
.w16:
+ _CET_ENDBR
vbroadcasti32x8 m5, [tlq+2] ; top
movsldup m4, [ipred_shuf]
psubw m5, m6 ; top - bottom
@@ -277,6 +285,7 @@ cglobal ipred_smooth_v_16bpc, 3, 7, 7, dst, stride, tl
jl .w16_loop
RET
.w32:
+ _CET_ENDBR
movu m5, [tlq+2]
psubw m5, m6
.w32_loop:
@@ -295,6 +304,7 @@ cglobal ipred_smooth_v_16bpc, 3, 7, 7, dst, stride, tl
jl .w32_loop
RET
.w64:
+ _CET_ENDBR
movu m4, [tlq+ 2]
movu m5, [tlq+66]
psubw m4, m6
@@ -329,6 +339,7 @@ cglobal ipred_smooth_h_16bpc, 3, 7, 7, dst, stride, tl
lea wq, [base+ipred_smooth_h_16bpc_avx512icl_table+wq]
jmp wq
.w4:
+ _CET_ENDBR
movsldup m4, [base+ipred_shuf]
vpbroadcastq m5, [base+smooth_weights_1d_16bpc+4*2]
.w4_loop:
@@ -356,6 +367,7 @@ cglobal ipred_smooth_h_16bpc, 3, 7, 7, dst, stride, tl
.end:
RET
.w8:
+ _CET_ENDBR
movsldup m4, [base+ipred_shuf]
vbroadcasti32x4 m5, [base+smooth_weights_1d_16bpc+8*2]
.w8_loop:
@@ -373,6 +385,7 @@ cglobal ipred_smooth_h_16bpc, 3, 7, 7, dst, stride, tl
jg .w8_loop
RET
.w16:
+ _CET_ENDBR
movsldup m4, [base+ipred_shuf]
vbroadcasti32x8 m5, [base+smooth_weights_1d_16bpc+16*2]
.w16_loop:
@@ -395,6 +408,7 @@ cglobal ipred_smooth_h_16bpc, 3, 7, 7, dst, stride, tl
jg .w16_loop
RET
.w32:
+ _CET_ENDBR
movu m5, [base+smooth_weights_1d_16bpc+32*2]
.w32_loop:
vpbroadcastq m3, [tlq+hq-8]
@@ -415,6 +429,7 @@ cglobal ipred_smooth_h_16bpc, 3, 7, 7, dst, stride, tl
jg .w32_loop
RET
.w64:
+ _CET_ENDBR
movu m4, [base+smooth_weights_1d_16bpc+64*2]
movu m5, [base+smooth_weights_1d_16bpc+64*3]
.w64_loop:
@@ -456,6 +471,7 @@ cglobal ipred_smooth_16bpc, 3, 7, 16, dst, stride, tl,
lea v_weightsq, [base+smooth_weights_2d_16bpc+hq*2]
jmp wq
.w4:
+ _CET_ENDBR
vpbroadcastq m5, [tlq+hq+2]
movshdup m3, [base+ipred_shuf]
movsldup m4, [base+ipred_shuf]
@@ -483,6 +499,7 @@ cglobal ipred_smooth_16bpc, 3, 7, 16, dst, stride, tl,
jg .w4_loop
RET
.w8:
+ _CET_ENDBR
vbroadcasti32x4 ym5, [tlq+hq+2]
movshdup m6, [base+ipred_shuf]
movsldup m7, [base+ipred_shuf]
@@ -517,6 +534,7 @@ cglobal ipred_smooth_16bpc, 3, 7, 16, dst, stride, tl,
jg .w8_loop
RET
.w16:
+ _CET_ENDBR
pmovzxwd m5, [tlq+hq+2]
mova m6, [base+smooth_weights_2d_16bpc+16*4]
vpblendmw m5{k1}, m0, m5 ; top, bottom
@@ -541,6 +559,7 @@ cglobal ipred_smooth_16bpc, 3, 7, 16, dst, stride, tl,
jg .w16_loop
RET
.w32:
+ _CET_ENDBR
pmovzxwd m5, [tlq+hq+ 2]
pmovzxwd m6, [tlq+hq+34]
mova m7, [base+smooth_weights_2d_16bpc+32*4]
@@ -574,6 +593,7 @@ cglobal ipred_smooth_16bpc, 3, 7, 16, dst, stride, tl,
jg .w32_loop
RET
.w64:
+ _CET_ENDBR
pmovzxwd m5, [tlq+hq+ 2]
pmovzxwd m6, [tlq+hq+34]
pmovzxwd m7, [tlq+hq+66]
@@ -621,6 +641,7 @@ cglobal pal_pred_16bpc, 4, 7, 4, dst, stride, pal, idx
lea stride3q, [strideq*3]
jmp wq
.w4:
+ _CET_ENDBR
pmovzxbw ym0, [idxq]
add idxq, 16
vpermw ym0, ym0, ym3
@@ -634,6 +655,7 @@ cglobal pal_pred_16bpc, 4, 7, 4, dst, stride, pal, idx
jg .w4
RET
.w8:
+ _CET_ENDBR
pmovzxbw m0, [idxq]
add idxq, 32
vpermw m0, m0, m3
@@ -646,6 +668,7 @@ cglobal pal_pred_16bpc, 4, 7, 4, dst, stride, pal, idx
jg .w8
RET
.w16:
+ _CET_ENDBR
vpermb m1, m2, [idxq]
add idxq, 64
vpermw m0, m1, m3
@@ -660,6 +683,7 @@ cglobal pal_pred_16bpc, 4, 7, 4, dst, stride, pal, idx
jg .w16
RET
.w32:
+ _CET_ENDBR
vpermb m1, m2, [idxq]
add idxq, 64
vpermw m0, m1, m3
@@ -672,6 +696,7 @@ cglobal pal_pred_16bpc, 4, 7, 4, dst, stride, pal, idx
jg .w32
RET
.w64:
+ _CET_ENDBR
vpermb m1, m2, [idxq]
add idxq, 64
vpermw m0, m1, m3
Reference in a new issue Copy permalink