SecBSD/ports
2
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions
ports/multimedia/dav1d/patches/patch-src_x86_ipred_avx512_asm

375 lines
11 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/ipred_avx512.asm
--- src/x86/ipred_avx512.asm.orig
+++ src/x86/ipred_avx512.asm
@@ -168,18 +168,23 @@ cglobal ipred_dc_left_8bpc, 3, 7, 5, dst, stride, tl,
add wq, r5
jmp r6
.h64:
+ _CET_ENDBR
movu ym1, [tlq+32] ; unaligned when jumping here from dc_top
vpdpbusd ym0, ym1, ym2
.h32:
+ _CET_ENDBR
vextracti32x4 xm1, ym0, 1
paddd xm0, xm1
.h16:
+ _CET_ENDBR
punpckhqdq xm1, xm0, xm0
paddd xm0, xm1
.h8:
+ _CET_ENDBR
psrlq xm1, xm0, 32
paddd xm0, xm1
.h4:
+ _CET_ENDBR
vpsrlvd xm0, xmm3
lea stride3q, [strideq*3]
vpbroadcastb m0, xm0
@@ -204,10 +209,12 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
lea stride3q, [strideq*3]
jmp r6
.h4:
+ _CET_ENDBR
movd xmm1, [tlq-4]
vpdpbusd xm0, xmm1, xm3
jmp wq
.w4:
+ _CET_ENDBR
movd xmm1, [tlq+1]
vpdpbusd xm0, xmm1, xm3
cmp hd, 4
@@ -228,6 +235,7 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
.w4_end:
vpbroadcastb xm0, xmm0
.s4:
+ _CET_ENDBR
movd [dstq+strideq*0], xm0
movd [dstq+strideq*1], xm0
movd [dstq+strideq*2], xm0
@@ -237,10 +245,12 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
jg .s4
RET
.h8:
+ _CET_ENDBR
movq xmm1, [tlq-8]
vpdpbusd xm0, xmm1, xm3
jmp wq
.w8:
+ _CET_ENDBR
movq xmm1, [tlq+1]
vextracti32x4 xm2, ym0, 1
vpdpbusd xm0, xmm1, xm3
@@ -261,6 +271,7 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
.w8_end:
vpbroadcastb xm0, xmm0
.s8:
+ _CET_ENDBR
movq [dstq+strideq*0], xm0
movq [dstq+strideq*1], xm0
movq [dstq+strideq*2], xm0
@@ -270,10 +281,12 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
jg .s8
RET
.h16:
+ _CET_ENDBR
mova xmm1, [tlq-16]
vpdpbusd xm0, xmm1, xm3
jmp wq
.w16:
+ _CET_ENDBR
movu xmm1, [tlq+1]
vextracti32x4 xm2, ym0, 1
vpdpbusd xm0, xmm1, xm3
@@ -294,6 +307,7 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
.w16_end:
vpbroadcastb xm0, xmm0
.s16:
+ _CET_ENDBR
mova [dstq+strideq*0], xm0
mova [dstq+strideq*1], xm0
mova [dstq+strideq*2], xm0
@@ -303,10 +317,12 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
jg .s16
RET
.h32:
+ _CET_ENDBR
mova ym1, [tlq-32]
vpdpbusd ym0, ym1, ym3
jmp wq
.w32:
+ _CET_ENDBR
movu ym1, [tlq+1]
vpdpbusd ym0, ym1, ym3
vextracti32x4 xm1, ym0, 1
@@ -326,6 +342,7 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
.w32_end:
vpbroadcastb ym0, xmm0
.s32:
+ _CET_ENDBR
mova [dstq+strideq*0], ym0
mova [dstq+strideq*1], ym0
mova [dstq+strideq*2], ym0
@@ -335,12 +352,14 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
jg .s32
RET
.h64:
+ _CET_ENDBR
mova ym1, [tlq-64]
mova ym2, [tlq-32]
vpdpbusd ym0, ym1, ym3
vpdpbusd ym0, ym2, ym3
jmp wq
.w64:
+ _CET_ENDBR
movu ym1, [tlq+ 1]
movu ym2, [tlq+33]
vpdpbusd ym0, ym1, ym3
@@ -361,6 +380,7 @@ cglobal ipred_dc_8bpc, 3, 7, 5, dst, stride, tl, w, h,
.w64_end:
vpbroadcastb m0, xmm0
.s64:
+ _CET_ENDBR
mova [dstq+strideq*0], m0
mova [dstq+strideq*1], m0
mova [dstq+strideq*2], m0
@@ -401,6 +421,7 @@ cglobal ipred_h_8bpc, 3, 7, 8, dst, stride, tl, w, h,
add wq, r6
jmp wq
.w4:
+ _CET_ENDBR
mova xmm1, [base+ipred_h_shuf+16]
.w4_loop:
movd xmm0, [tlq+hq-4]
@@ -414,6 +435,7 @@ cglobal ipred_h_8bpc, 3, 7, 8, dst, stride, tl, w, h,
jg .w4_loop
RET
.w8:
+ _CET_ENDBR
movsldup xmm2, [base+ipred_h_shuf+16]
movshdup xmm3, [base+ipred_h_shuf+16]
.w8_loop:
@@ -429,6 +451,7 @@ cglobal ipred_h_8bpc, 3, 7, 8, dst, stride, tl, w, h,
jg .w8_loop
RET
.w16:
+ _CET_ENDBR
movsldup m1, [base+smooth_shuf]
.w16_loop:
vpbroadcastd m0, [tlq+hq-4]
@@ -442,6 +465,7 @@ cglobal ipred_h_8bpc, 3, 7, 8, dst, stride, tl, w, h,
jg .w16
RET
.w32:
+ _CET_ENDBR
vpbroadcastd ym3, [base+pb_1]
vpord m2, m3, [base+pb_2] {1to16}
.w32_loop:
@@ -457,6 +481,7 @@ cglobal ipred_h_8bpc, 3, 7, 8, dst, stride, tl, w, h,
jg .w32_loop
RET
.w64:
+ _CET_ENDBR
vpbroadcastd m4, [base+pb_3]
vpbroadcastd m5, [base+pb_2]
vpbroadcastd m6, [base+pb_1]
@@ -509,6 +534,7 @@ cglobal ipred_paeth_8bpc, 3, 7, 10, dst, stride, tl, w
jmp wq
INIT_YMM avx512icl
.w4:
+ _CET_ENDBR
vpbroadcastd m6, [topq]
mova m9, [ipred_h_shuf]
psubusb m7, m5, m6
@@ -536,6 +562,7 @@ INIT_YMM avx512icl
RET
INIT_ZMM avx512icl
.w8:
+ _CET_ENDBR
vpbroadcastq m6, [topq]
movsldup m9, [smooth_shuf]
psubusb m7, m5, m6
@@ -564,6 +591,7 @@ INIT_ZMM avx512icl
.w8_ret:
RET
.w16:
+ _CET_ENDBR
vbroadcasti32x4 m6, [topq]
movsldup m9, [smooth_shuf]
psubusb m7, m5, m6
@@ -582,6 +610,7 @@ INIT_ZMM avx512icl
jg .w16_loop
RET
.w32:
+ _CET_ENDBR
vbroadcasti32x8 m6, [topq]
mova ym9, ym8
psubusb m7, m5, m6
@@ -598,6 +627,7 @@ INIT_ZMM avx512icl
jg .w32_loop
RET
.w64:
+ _CET_ENDBR
movu m6, [topq]
psubusb m7, m5, m6
psubusb m0, m6, m5
@@ -626,6 +656,7 @@ cglobal ipred_smooth_v_8bpc, 3, 7, 7, dst, stride, tl,
lea stride3q, [strideq*3]
jmp wq
.w4:
+ _CET_ENDBR
vpbroadcastd m2, [tlq+1]
movshdup m5, [smooth_shuf]
mova ym6, [smooth_endA]
@@ -656,6 +687,7 @@ cglobal ipred_smooth_v_8bpc, 3, 7, 7, dst, stride, tl,
.ret:
RET
.w8:
+ _CET_ENDBR
vpbroadcastq m2, [tlq+1]
movshdup m5, [smooth_shuf]
mova ym6, [smooth_endA]
@@ -679,6 +711,7 @@ cglobal ipred_smooth_v_8bpc, 3, 7, 7, dst, stride, tl,
jl .w8_loop
RET
.w16:
+ _CET_ENDBR
vbroadcasti32x4 m3, [tlq+1]
movshdup m6, [smooth_shuf]
mova m7, [smooth_endB]
@@ -707,6 +740,7 @@ cglobal ipred_smooth_v_8bpc, 3, 7, 7, dst, stride, tl,
jl .w16_loop
RET
.w32:
+ _CET_ENDBR
vbroadcasti32x8 m3, [tlq+1]
movshdup m6, [smooth_shuf]
mova m7, [smooth_endB]
@@ -733,6 +767,7 @@ cglobal ipred_smooth_v_8bpc, 3, 7, 7, dst, stride, tl,
jl .w32_loop
RET
.w64:
+ _CET_ENDBR
movu m3, [tlq+1]
mova m6, [smooth_endB]
punpcklbw m2, m3, m4
@@ -772,6 +807,7 @@ cglobal ipred_smooth_h_8bpc, 4, 7, 11, dst, stride, tl
lea stride3q, [strideq*3]
jmp wq
.w4:
+ _CET_ENDBR
movsldup m3, [smooth_shuf]
vpbroadcastq m7, [smooth_weights+4*2]
mova ym8, [smooth_endA]
@@ -802,6 +838,7 @@ cglobal ipred_smooth_h_8bpc, 4, 7, 11, dst, stride, tl
.ret:
RET
.w8:
+ _CET_ENDBR
movsldup m3, [smooth_shuf]
vbroadcasti32x4 m7, [smooth_weights+8*2]
mova ym8, [smooth_endA]
@@ -825,6 +862,7 @@ cglobal ipred_smooth_h_8bpc, 4, 7, 11, dst, stride, tl
jg .w8_loop
RET
.w16:
+ _CET_ENDBR
movsldup m7, [smooth_shuf]
vbroadcasti32x4 m8, [smooth_weights+16*2]
vbroadcasti32x4 m9, [smooth_weights+16*3]
@@ -850,6 +888,7 @@ cglobal ipred_smooth_h_8bpc, 4, 7, 11, dst, stride, tl
jg .w16_loop
RET
.w32:
+ _CET_ENDBR
mova m10, [smooth_endA]
vpbroadcastd ym7, [pb_1]
vbroadcasti32x8 m8, [smooth_weights+32*2]
@@ -874,6 +913,7 @@ cglobal ipred_smooth_h_8bpc, 4, 7, 11, dst, stride, tl
jg .w32_loop
RET
.w64:
+ _CET_ENDBR
mova m7, [smooth_weights+64*2]
mova m8, [smooth_weights+64*3]
mova m9, [smooth_endA]
@@ -912,6 +952,7 @@ cglobal ipred_smooth_8bpc, 4, 7, 16, dst, stride, tl,
lea stride3q, [strideq*3]
jmp wq
.w4:
+ _CET_ENDBR
vpbroadcastd m8, [tlq+hq+1]
movsldup m4, [smooth_shuf]
movshdup m5, [smooth_shuf]
@@ -954,6 +995,7 @@ cglobal ipred_smooth_8bpc, 4, 7, 16, dst, stride, tl,
.ret:
RET
.w8:
+ _CET_ENDBR
vpbroadcastq m8, [tlq+hq+1]
movsldup m4, [smooth_shuf]
movshdup m5, [smooth_shuf]
@@ -988,6 +1030,7 @@ cglobal ipred_smooth_8bpc, 4, 7, 16, dst, stride, tl,
jg .w8_loop
RET
.w16:
+ _CET_ENDBR
vbroadcasti32x4 m9, [tlq+hq+1]
movsldup m5, [smooth_shuf]
movshdup m10, [smooth_shuf]
@@ -1031,6 +1074,7 @@ cglobal ipred_smooth_8bpc, 4, 7, 16, dst, stride, tl,
jg .w16_loop
RET
.w32:
+ _CET_ENDBR
vbroadcasti32x8 m9, [tlq+hq+1]
movshdup m10, [smooth_shuf]
mova m12, [smooth_weights+32*2]
@@ -1073,6 +1117,7 @@ cglobal ipred_smooth_8bpc, 4, 7, 16, dst, stride, tl,
jg .w32_loop
RET
.w64:
+ _CET_ENDBR
movu m9, [tlq+hq+1]
mova m11, [smooth_weights+64*2]
mova m2, [smooth_weights+64*3]
@@ -1122,6 +1167,7 @@ cglobal pal_pred_8bpc, 4, 7, 5, dst, stride, pal, idx,
lea stride3q, [strideq*3]
jmp wq
.w4:
+ _CET_ENDBR
pshufb xmm0, xm4, [idxq]
add idxq, 16
movd [dstq+strideq*0], xmm0
@@ -1133,6 +1179,7 @@ cglobal pal_pred_8bpc, 4, 7, 5, dst, stride, pal, idx,
jg .w4
RET
.w8:
+ _CET_ENDBR
pshufb xmm0, xm4, [idxq+16*0]
pshufb xmm1, xm4, [idxq+16*1]
add idxq, 16*2
@@ -1145,6 +1192,7 @@ cglobal pal_pred_8bpc, 4, 7, 5, dst, stride, pal, idx,
jg .w8
RET
.w16:
+ _CET_ENDBR
pshufb m0, m4, [idxq]
add idxq, 64
mova [dstq+strideq*0], xm0
@@ -1156,6 +1204,7 @@ cglobal pal_pred_8bpc, 4, 7, 5, dst, stride, pal, idx,
jg .w16
RET
.w32:
+ _CET_ENDBR
pshufb m0, m4, [idxq+64*0]
pshufb m1, m4, [idxq+64*1]
add idxq, 64*2
@@ -1168,6 +1217,7 @@ cglobal pal_pred_8bpc, 4, 7, 5, dst, stride, pal, idx,
jg .w32
RET
.w64:
+ _CET_ENDBR
pshufb m0, m4, [idxq+64*0]
pshufb m1, m4, [idxq+64*1]
pshufb m2, m4, [idxq+64*2]
Reference in a new issue Copy permalink