Skip to content

LLVM optimization on PR 162595 [bot] #76

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 1 commit into
base: pr-162595-before-1760438192
Choose a base branch
from
Open

LLVM optimization on PR 162595 [bot] #76

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
29 changes: 12 additions & 17 deletions results/MultiSource/Applications/JM/ldecod/CMakeFiles/ldecod.dir/block.s
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -7635,7 +7635,6 @@ itrans_sp_chroma: # @itrans_sp_chroma
# %bb.8: # %.split301.us
slli.d $a7, $a1, 8
add.d $a7, $a0, $a7
vneg.w $vr1, $vr0
ori $t8, $zero, 2664
ldx.w $t8, $a7, $t8
ori $fp, $zero, 2728
Expand All @@ -7644,15 +7643,14 @@ itrans_sp_chroma: # @itrans_sp_chroma
ldx.w $s0, $a7, $s0
ori $s1, $zero, 3496
ldx.w $a7, $a7, $s1
vldrepl.w $vr2, $t7, 0
vmax.w $vr1, $vr0, $vr1
vldrepl.w $vr1, $t7, 0
vsigncov.w $vr2, $vr0, $vr0
vslti.w $vr0, $vr0, 0
vreplgr2vr.w $vr3, $t6
vmadd.w $vr3, $vr1, $vr2
vmadd.w $vr3, $vr2, $vr1
vreplgr2vr.w $vr1, $t5
vsra.w $vr1, $vr3, $vr1
vneg.w $vr2, $vr1
vmax.w $vr1, $vr1, $vr2
vsigncov.w $vr1, $vr1, $vr1
vneg.w $vr2, $vr1
vbitsel.v $vr0, $vr1, $vr2, $vr0
vinsgr2vr.w $vr1, $a7, 0
Expand All @@ -7661,7 +7659,6 @@ itrans_sp_chroma: # @itrans_sp_chroma
vinsgr2vr.w $vr1, $s0, 2
vinsgr2vr.w $vr1, $t8, 3
vadd.w $vr0, $vr0, $vr1
vmul.w $vr0, $vr0, $vr2
b .LBB9_10
.LBB9_9: # %.split301
ldx.w $a7, $t2, $a6
Expand All @@ -7685,22 +7682,20 @@ itrans_sp_chroma: # @itrans_sp_chroma
vreplgr2vr.w $vr2, $t0
vsll.w $vr1, $vr1, $vr2
vsrai.w $vr1, $vr1, 5
vldrepl.w $vr2, $t7, 0
vadd.w $vr0, $vr1, $vr0
vldrepl.w $vr1, $t7, 0
vneg.w $vr2, $vr0
vmax.w $vr2, $vr0, $vr2
vsigncov.w $vr1, $vr0, $vr0
vreplgr2vr.w $vr3, $t6
vmadd.w $vr3, $vr2, $vr1
vmadd.w $vr3, $vr1, $vr2
vreplgr2vr.w $vr1, $t5
vsra.w $vr1, $vr3, $vr1
vslti.w $vr0, $vr0, 0
vneg.w $vr2, $vr1
vldrepl.w $vr3, $t4, 0
vmax.w $vr1, $vr1, $vr2
vneg.w $vr2, $vr1
vbitsel.v $vr0, $vr1, $vr2, $vr0
vmul.w $vr0, $vr0, $vr3
vldrepl.w $vr2, $t4, 0
vsigncov.w $vr1, $vr1, $vr1
vneg.w $vr3, $vr1
vbitsel.v $vr0, $vr1, $vr3, $vr0
.LBB9_10: # %.preheader278
vmul.w $vr0, $vr0, $vr2
add.d $a6, $t2, $a6
addi.d $t1, $t1, 12
add.d $a7, $t2, $t1
Expand Down
77 changes: 35 additions & 42 deletions results/MultiSource/Applications/JM/lencod/CMakeFiles/lencod.dir/macroblock.s
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -16985,9 +16985,9 @@ find_sad_16x16: # @find_sad_16x16
vadd.w $vr15, $vr14, $vr13
vsub.w $vr14, $vr13, $vr14
vsub.w $vr8, $vr8, $vr10
vadd.w $vr16, $vr15, $vr11
vadd.w $vr17, $vr15, $vr11
vsub.w $vr13, $vr11, $vr15
vadd.w $vr18, $vr14, $vr8
vadd.w $vr16, $vr14, $vr8
vsub.w $vr8, $vr8, $vr14
ld.w $a0, $a2, -960
ld.w $a4, $a2, -896
Expand Down Expand Up @@ -17018,21 +17018,21 @@ find_sad_16x16: # @find_sad_16x16
ld.w $a4, $a2, -384
ld.w $a5, $a2, -320
ld.w $t0, $a2, -256
vinsgr2vr.w $vr17, $a0, 0
vinsgr2vr.w $vr17, $a4, 1
vinsgr2vr.w $vr17, $a5, 2
vinsgr2vr.w $vr17, $t0, 3
vadd.w $vr19, $vr17, $vr15
vsub.w $vr15, $vr15, $vr17
vinsgr2vr.w $vr18, $a0, 0
vinsgr2vr.w $vr18, $a4, 1
vinsgr2vr.w $vr18, $a5, 2
vinsgr2vr.w $vr18, $t0, 3
vadd.w $vr19, $vr18, $vr15
vsub.w $vr15, $vr15, $vr18
vsub.w $vr10, $vr10, $vr11
vadd.w $vr17, $vr19, $vr14
vadd.w $vr18, $vr19, $vr14
vsub.w $vr19, $vr14, $vr19
vadd.w $vr20, $vr15, $vr10
vsub.w $vr15, $vr10, $vr15
vadd.w $vr10, $vr17, $vr7
vadd.w $vr11, $vr16, $vr12
vsub.w $vr12, $vr12, $vr16
vsub.w $vr7, $vr7, $vr17
vadd.w $vr10, $vr18, $vr7
vadd.w $vr11, $vr17, $vr12
vsub.w $vr12, $vr12, $vr17
vsub.w $vr7, $vr7, $vr18
vadd.w $vr14, $vr11, $vr10
vpickve2gr.w $a0, $vr14, 0
st.w $a0, $a2, -1008
Expand Down Expand Up @@ -17060,20 +17060,20 @@ find_sad_16x16: # @find_sad_16x16
st.w $a0, $a2, -864
vpickve2gr.w $a0, $vr14, 3
st.w $a0, $a2, -800
vsub.w $vr16, $vr7, $vr12
vpickve2gr.w $a0, $vr16, 0
vsub.w $vr17, $vr7, $vr12
vpickve2gr.w $a0, $vr17, 0
st.w $a0, $a2, -960
vpickve2gr.w $a0, $vr16, 1
vpickve2gr.w $a0, $vr17, 1
st.w $a0, $a2, -896
vpickve2gr.w $a0, $vr16, 2
vpickve2gr.w $a0, $vr17, 2
st.w $a0, $a2, -832
vpickve2gr.w $a0, $vr16, 3
vpickve2gr.w $a0, $vr17, 3
st.w $a0, $a2, -768
vadd.w $vr16, $vr20, $vr5
vadd.w $vr17, $vr18, $vr9
vsub.w $vr9, $vr9, $vr18
vsub.w $vr18, $vr5, $vr20
vadd.w $vr5, $vr17, $vr16
vadd.w $vr17, $vr20, $vr5
vadd.w $vr18, $vr16, $vr9
vsub.w $vr9, $vr9, $vr16
vsub.w $vr16, $vr5, $vr20
vadd.w $vr5, $vr18, $vr17
vpickve2gr.w $a0, $vr5, 0
st.w $a0, $a2, -752
vpickve2gr.w $a0, $vr5, 1
Expand All @@ -17082,7 +17082,7 @@ find_sad_16x16: # @find_sad_16x16
st.w $a0, $a2, -624
vpickve2gr.w $a0, $vr5, 3
st.w $a0, $a2, -560
vsub.w $vr20, $vr16, $vr17
vsub.w $vr20, $vr17, $vr18
vpickve2gr.w $a0, $vr20, 0
st.w $a0, $a2, -720
vpickve2gr.w $a0, $vr20, 1
Expand All @@ -17091,7 +17091,7 @@ find_sad_16x16: # @find_sad_16x16
st.w $a0, $a2, -592
vpickve2gr.w $a0, $vr20, 3
st.w $a0, $a2, -528
vadd.w $vr20, $vr9, $vr18
vadd.w $vr20, $vr9, $vr16
vpickve2gr.w $a0, $vr20, 0
st.w $a0, $a2, -736
vpickve2gr.w $a0, $vr20, 1
Expand All @@ -17100,7 +17100,7 @@ find_sad_16x16: # @find_sad_16x16
st.w $a0, $a2, -608
vpickve2gr.w $a0, $vr20, 3
st.w $a0, $a2, -544
vsub.w $vr21, $vr18, $vr9
vsub.w $vr21, $vr16, $vr9
vpickve2gr.w $a0, $vr21, 0
st.w $a0, $a2, -704
vpickve2gr.w $a0, $vr21, 1
Expand Down Expand Up @@ -17158,26 +17158,19 @@ find_sad_16x16: # @find_sad_16x16
vstelm.w $vr25, $a2, 0, 3
vabsd.w $vr2, $vr2, $vr4
vabsd.w $vr4, $vr23, $vr24
vneg.w $vr23, $vr15
vmax.w $vr15, $vr15, $vr23
vneg.w $vr23, $vr8
vmax.w $vr8, $vr8, $vr23
vsigncov.w $vr15, $vr15, $vr15
vsigncov.w $vr8, $vr8, $vr8
vabsd.w $vr3, $vr3, $vr6
vabsd.w $vr6, $vr21, $vr22
vneg.w $vr21, $vr19
vmax.w $vr19, $vr19, $vr21
vneg.w $vr21, $vr13
vmax.w $vr13, $vr13, $vr21
vabsd.w $vr9, $vr18, $vr9
vabsd.w $vr16, $vr16, $vr17
vneg.w $vr17, $vr20
vmax.w $vr17, $vr20, $vr17
vneg.w $vr18, $vr5
vmax.w $vr5, $vr5, $vr18
vsigncov.w $vr19, $vr19, $vr19
vsigncov.w $vr13, $vr13, $vr13
vabsd.w $vr9, $vr16, $vr9
vabsd.w $vr16, $vr17, $vr18
vsigncov.w $vr17, $vr20, $vr20
vsigncov.w $vr5, $vr5, $vr5
vabsd.w $vr7, $vr7, $vr12
vabsd.w $vr10, $vr10, $vr11
vneg.w $vr11, $vr14
vmax.w $vr11, $vr14, $vr11
vsigncov.w $vr11, $vr14, $vr14
vadd.w $vr1, $vr11, $vr1
vadd.w $vr1, $vr10, $vr1
vadd.w $vr1, $vr7, $vr1
Expand Down
18 changes: 6 additions & 12 deletions results/MultiSource/Applications/JM/lencod/CMakeFiles/lencod.dir/me_distortion.s
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -753,18 +753,12 @@ HadamardSAD8x8: # @HadamardSAD8x8
xvinsgr2vr.w $xr0, $a3, 5
xvinsgr2vr.w $xr0, $a2, 6
ld.w $a0, $a0, 224
xvneg.w $xr8, $xr1
xvmax.w $xr1, $xr1, $xr8
xvneg.w $xr8, $xr2
xvmax.w $xr2, $xr2, $xr8
xvneg.w $xr8, $xr3
xvmax.w $xr3, $xr3, $xr8
xvneg.w $xr8, $xr4
xvmax.w $xr4, $xr4, $xr8
xvneg.w $xr8, $xr5
xvmax.w $xr5, $xr5, $xr8
xvneg.w $xr8, $xr6
xvmax.w $xr6, $xr6, $xr8
xvsigncov.w $xr1, $xr1, $xr1
xvsigncov.w $xr2, $xr2, $xr2
xvsigncov.w $xr3, $xr3, $xr3
xvsigncov.w $xr4, $xr4, $xr4
xvsigncov.w $xr5, $xr5, $xr5
xvsigncov.w $xr6, $xr6, $xr6
xvinsgr2vr.w $xr0, $a0, 7
xvadda.w $xr0, $xr7, $xr0
xvadd.w $xr0, $xr6, $xr0
Expand Down
3 changes: 1 addition & 2 deletions results/MultiSource/Applications/JM/lencod/CMakeFiles/lencod.dir/me_epzs.s
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1814,8 +1814,7 @@ EPZSSliceInit: # @EPZSSliceInit
vsrli.b $vr12, $vr11, 7
vadd.b $vr11, $vr11, $vr12
vsrai.b $vr11, $vr11, 1
vneg.b $vr12, $vr11
vmax.b $vr11, $vr11, $vr12
vsigncov.b $vr11, $vr11, $vr11
vilvl.b $vr11, $vr5, $vr11
vbitseti.h $vr11, $vr11, 14
vpickev.h $vr9, $vr9, $vr9
Expand Down
Loading