-
Notifications
You must be signed in to change notification settings - Fork 30
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
- Loading branch information
1 parent
7044410
commit 7ab9457
Showing
14 changed files
with
265 additions
and
119 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file was deleted.
Oops, something went wrong.
This file was deleted.
Oops, something went wrong.
This file was deleted.
Oops, something went wrong.
This file was deleted.
Oops, something went wrong.
This file was deleted.
Oops, something went wrong.
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,32 @@ | ||
Vulkan xf-barrier-cbar-rlx-1 | ||
"Adapted from Figure 2 in | ||
Portable Inter-workgroup Barrier Synchronisation for GPUs | ||
https://dl.acm.org/doi/pdf/10.1145/2983990.2984032" | ||
{ | ||
x0=0; x1=0; x2=0; x3=0; x4=0; x5=0; | ||
f1=0; f2=0; | ||
} | ||
P0@sg 0, wg 0, qf 0 | P1@sg 0, wg 0, qf 0 | P2@sg 0, wg 1, qf 0 | P3@sg 0, wg 1, qf 0 | P4@sg 0, wg 2, qf 0 | P5@sg 0, wg 2, qf 0 ; | ||
st.av.dv.sc0 x0, 1 | st.av.dv.sc0 x1, 1 | st.av.dv.sc0 x2, 1 | st.av.dv.sc0 x3, 1 | st.av.dv.sc0 x4, 1 | st.av.dv.sc0 x5, 1 ; | ||
LC00: | LC10: | cbar.acq_rel.wg.semsc0 10 | cbar.acq_rel.wg.semsc0 10 | cbar.acq_rel.wg.semsc0 20 | cbar.acq_rel.wg.semsc0 20 ; | ||
ld.atom.acq.dv.sc0.semsc0 r9, f1 | ld.atom.acq.dv.sc0.semsc0 r9, f2 | st.atom.rel.dv.sc0.semsc0 f1, 1 | cbar.acq_rel.wg.semsc0 11 | st.atom.rel.dv.sc0.semsc0 f2, 1 | cbar.acq_rel.wg.semsc0 21 ; | ||
bne r9, 0, LC01 | bne r9, 0, LC11 | LC20: | ld.vis.dv.sc0 r0, x0 | LC40: | ld.vis.dv.sc0 r0, x0 ; | ||
goto LC00 | goto LC10 | ld.atom.acq.dv.sc0.semsc0 r9, f1 | ld.vis.dv.sc0 r1, x1 | ld.atom.acq.dv.sc0.semsc0 r9, f2 | ld.vis.dv.sc0 r1, x1 ; | ||
LC01: | LC11: | bne r9, 1, LC21 | ld.vis.dv.sc0 r2, x2 | bne r9, 1, LC41 | ld.vis.dv.sc0 r2, x2 ; | ||
cbar.wg.semsc0 00 | cbar.wg.semsc0 00 | goto LC20 | ld.vis.dv.sc0 r3, x3 | goto LC40 | ld.vis.dv.sc0 r3, x3 ; | ||
st.atom.rel.dv.sc0.semsc0 f1, 0 | st.atom.rel.dv.sc0.semsc0 f2, 0 | LC21: | ld.vis.dv.sc0 r4, x4 | LC41: | ld.vis.dv.sc0 r4, x4 ; | ||
ld.vis.dv.sc0 r0, x0 | ld.vis.dv.sc0 r0, x0 | cbar.acq_rel.wg.semsc0 11 | ld.vis.dv.sc0 r5, x5 | cbar.acq_rel.wg.semsc0 21 | ld.vis.dv.sc0 r5, x5 ; | ||
ld.vis.dv.sc0 r1, x1 | ld.vis.dv.sc0 r1, x1 | ld.vis.dv.sc0 r0, x0 | | ld.vis.dv.sc0 r0, x0 | ; | ||
ld.vis.dv.sc0 r2, x2 | ld.vis.dv.sc0 r2, x2 | ld.vis.dv.sc0 r1, x1 | | ld.vis.dv.sc0 r1, x1 | ; | ||
ld.vis.dv.sc0 r3, x3 | ld.vis.dv.sc0 r3, x3 | ld.vis.dv.sc0 r2, x2 | | ld.vis.dv.sc0 r2, x2 | ; | ||
ld.vis.dv.sc0 r4, x4 | ld.vis.dv.sc0 r4, x4 | ld.vis.dv.sc0 r3, x3 | | ld.vis.dv.sc0 r3, x3 | ; | ||
ld.vis.dv.sc0 r5, x5 | ld.vis.dv.sc0 r5, x5 | ld.vis.dv.sc0 r4, x4 | | ld.vis.dv.sc0 r4, x4 | ; | ||
| | ld.vis.dv.sc0 r5, x5 | | ld.vis.dv.sc0 r5, x5 | ; | ||
|
||
forall( | ||
P0:r0 == 1 /\ P0:r1 == 1 /\ P0:r2 == 1 /\ P0:r3 == 1 /\ P0:r4 == 1 /\ P0:r5 == 1 /\ | ||
P1:r0 == 1 /\ P1:r1 == 1 /\ P1:r2 == 1 /\ P1:r3 == 1 /\ P1:r4 == 1 /\ P1:r5 == 1 /\ | ||
P2:r0 == 1 /\ P2:r1 == 1 /\ P2:r2 == 1 /\ P2:r3 == 1 /\ P2:r4 == 1 /\ P2:r5 == 1 /\ | ||
P3:r0 == 1 /\ P3:r1 == 1 /\ P3:r2 == 1 /\ P3:r3 == 1 /\ P3:r4 == 1 /\ P3:r5 == 1 /\ | ||
P4:r0 == 1 /\ P4:r1 == 1 /\ P4:r2 == 1 /\ P4:r3 == 1 /\ P4:r4 == 1 /\ P4:r5 == 1 /\ | ||
P5:r0 == 1 /\ P5:r1 == 1 /\ P5:r2 == 1 /\ P5:r3 == 1 /\ P5:r4 == 1 /\ P5:r5 == 1) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,32 @@ | ||
Vulkan xf-barrier-cbar-rlx-2 | ||
"Adapted from Figure 2 in | ||
Portable Inter-workgroup Barrier Synchronisation for GPUs | ||
https://dl.acm.org/doi/pdf/10.1145/2983990.2984032" | ||
{ | ||
x0=0; x1=0; x2=0; x3=0; x4=0; x5=0; | ||
f1=0; f2=0; | ||
} | ||
P0@sg 0, wg 0, qf 0 | P1@sg 0, wg 0, qf 0 | P2@sg 0, wg 1, qf 0 | P3@sg 0, wg 1, qf 0 | P4@sg 0, wg 2, qf 0 | P5@sg 0, wg 2, qf 0 ; | ||
st.av.dv.sc0 x0, 1 | st.av.dv.sc0 x1, 1 | st.av.dv.sc0 x2, 1 | st.av.dv.sc0 x3, 1 | st.av.dv.sc0 x4, 1 | st.av.dv.sc0 x5, 1 ; | ||
LC00: | LC10: | cbar.wg.semsc0 10 | cbar.wg.semsc0 10 | cbar.wg.semsc0 20 | cbar.wg.semsc0 20 ; | ||
ld.atom.acq.dv.sc0.semsc0 r9, f1 | ld.atom.acq.dv.sc0.semsc0 r9, f2 | st.atom.rel.dv.sc0.semsc0 f1, 1 | cbar.acq_rel.wg.semsc0 11 | st.atom.rel.dv.sc0.semsc0 f2, 1 | cbar.acq_rel.wg.semsc0 21 ; | ||
bne r9, 0, LC01 | bne r9, 0, LC11 | LC20: | ld.vis.dv.sc0 r0, x0 | LC40: | ld.vis.dv.sc0 r0, x0 ; | ||
goto LC00 | goto LC10 | ld.atom.acq.dv.sc0.semsc0 r9, f1 | ld.vis.dv.sc0 r1, x1 | ld.atom.acq.dv.sc0.semsc0 r9, f2 | ld.vis.dv.sc0 r1, x1 ; | ||
LC01: | LC11: | bne r9, 1, LC21 | ld.vis.dv.sc0 r2, x2 | bne r9, 1, LC41 | ld.vis.dv.sc0 r2, x2 ; | ||
cbar.acq_rel.wg.semsc0 00 | cbar.acq_rel.wg.semsc0 00 | goto LC20 | ld.vis.dv.sc0 r3, x3 | goto LC40 | ld.vis.dv.sc0 r3, x3 ; | ||
st.atom.rel.dv.sc0.semsc0 f1, 0 | st.atom.rel.dv.sc0.semsc0 f2, 0 | LC21: | ld.vis.dv.sc0 r4, x4 | LC41: | ld.vis.dv.sc0 r4, x4 ; | ||
ld.vis.dv.sc0 r0, x0 | ld.vis.dv.sc0 r0, x0 | cbar.acq_rel.wg.semsc0 11 | ld.vis.dv.sc0 r5, x5 | cbar.acq_rel.wg.semsc0 21 | ld.vis.dv.sc0 r5, x5 ; | ||
ld.vis.dv.sc0 r1, x1 | ld.vis.dv.sc0 r1, x1 | ld.vis.dv.sc0 r0, x0 | | ld.vis.dv.sc0 r0, x0 | ; | ||
ld.vis.dv.sc0 r2, x2 | ld.vis.dv.sc0 r2, x2 | ld.vis.dv.sc0 r1, x1 | | ld.vis.dv.sc0 r1, x1 | ; | ||
ld.vis.dv.sc0 r3, x3 | ld.vis.dv.sc0 r3, x3 | ld.vis.dv.sc0 r2, x2 | | ld.vis.dv.sc0 r2, x2 | ; | ||
ld.vis.dv.sc0 r4, x4 | ld.vis.dv.sc0 r4, x4 | ld.vis.dv.sc0 r3, x3 | | ld.vis.dv.sc0 r3, x3 | ; | ||
ld.vis.dv.sc0 r5, x5 | ld.vis.dv.sc0 r5, x5 | ld.vis.dv.sc0 r4, x4 | | ld.vis.dv.sc0 r4, x4 | ; | ||
| | ld.vis.dv.sc0 r5, x5 | | ld.vis.dv.sc0 r5, x5 | ; | ||
|
||
forall( | ||
P0:r0 == 1 /\ P0:r1 == 1 /\ P0:r2 == 1 /\ P0:r3 == 1 /\ P0:r4 == 1 /\ P0:r5 == 1 /\ | ||
P1:r0 == 1 /\ P1:r1 == 1 /\ P1:r2 == 1 /\ P1:r3 == 1 /\ P1:r4 == 1 /\ P1:r5 == 1 /\ | ||
P2:r0 == 1 /\ P2:r1 == 1 /\ P2:r2 == 1 /\ P2:r3 == 1 /\ P2:r4 == 1 /\ P2:r5 == 1 /\ | ||
P3:r0 == 1 /\ P3:r1 == 1 /\ P3:r2 == 1 /\ P3:r3 == 1 /\ P3:r4 == 1 /\ P3:r5 == 1 /\ | ||
P4:r0 == 1 /\ P4:r1 == 1 /\ P4:r2 == 1 /\ P4:r3 == 1 /\ P4:r4 == 1 /\ P4:r5 == 1 /\ | ||
P5:r0 == 1 /\ P5:r1 == 1 /\ P5:r2 == 1 /\ P5:r3 == 1 /\ P5:r4 == 1 /\ P5:r5 == 1) |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Original file line number | Diff line number | Diff line change |
---|---|---|
@@ -0,0 +1,32 @@ | ||
Vulkan xf-barrier-cbar-rlx-3 | ||
"Adapted from Figure 2 in | ||
Portable Inter-workgroup Barrier Synchronisation for GPUs | ||
https://dl.acm.org/doi/pdf/10.1145/2983990.2984032" | ||
{ | ||
x0=0; x1=0; x2=0; x3=0; x4=0; x5=0; | ||
f1=0; f2=0; | ||
} | ||
P0@sg 0, wg 0, qf 0 | P1@sg 0, wg 0, qf 0 | P2@sg 0, wg 1, qf 0 | P3@sg 0, wg 1, qf 0 | P4@sg 0, wg 2, qf 0 | P5@sg 0, wg 2, qf 0 ; | ||
st.av.dv.sc0 x0, 1 | st.av.dv.sc0 x1, 1 | st.av.dv.sc0 x2, 1 | st.av.dv.sc0 x3, 1 | st.av.dv.sc0 x4, 1 | st.av.dv.sc0 x5, 1 ; | ||
LC00: | LC10: | cbar.acq_rel.wg.semsc0 10 | cbar.acq_rel.wg.semsc0 10 | cbar.acq_rel.wg.semsc0 20 | cbar.acq_rel.wg.semsc0 20 ; | ||
ld.atom.acq.dv.sc0.semsc0 r9, f1 | ld.atom.acq.dv.sc0.semsc0 r9, f2 | st.atom.rel.dv.sc0.semsc0 f1, 1 | cbar.wg.semsc0 11 | st.atom.rel.dv.sc0.semsc0 f2, 1 | cbar.wg.semsc0 21 ; | ||
bne r9, 0, LC01 | bne r9, 0, LC11 | LC20: | ld.vis.dv.sc0 r0, x0 | LC40: | ld.vis.dv.sc0 r0, x0 ; | ||
goto LC00 | goto LC10 | ld.atom.acq.dv.sc0.semsc0 r9, f1 | ld.vis.dv.sc0 r1, x1 | ld.atom.acq.dv.sc0.semsc0 r9, f2 | ld.vis.dv.sc0 r1, x1 ; | ||
LC01: | LC11: | bne r9, 1, LC21 | ld.vis.dv.sc0 r2, x2 | bne r9, 1, LC41 | ld.vis.dv.sc0 r2, x2 ; | ||
cbar.acq_rel.wg.semsc0 00 | cbar.acq_rel.wg.semsc0 00 | goto LC20 | ld.vis.dv.sc0 r3, x3 | goto LC40 | ld.vis.dv.sc0 r3, x3 ; | ||
st.atom.rel.dv.sc0.semsc0 f1, 0 | st.atom.rel.dv.sc0.semsc0 f2, 0 | LC21: | ld.vis.dv.sc0 r4, x4 | LC41: | ld.vis.dv.sc0 r4, x4 ; | ||
ld.vis.dv.sc0 r0, x0 | ld.vis.dv.sc0 r0, x0 | cbar.wg.semsc0 11 | ld.vis.dv.sc0 r5, x5 | cbar.wg.semsc0 21 | ld.vis.dv.sc0 r5, x5 ; | ||
ld.vis.dv.sc0 r1, x1 | ld.vis.dv.sc0 r1, x1 | ld.vis.dv.sc0 r0, x0 | | ld.vis.dv.sc0 r0, x0 | ; | ||
ld.vis.dv.sc0 r2, x2 | ld.vis.dv.sc0 r2, x2 | ld.vis.dv.sc0 r1, x1 | | ld.vis.dv.sc0 r1, x1 | ; | ||
ld.vis.dv.sc0 r3, x3 | ld.vis.dv.sc0 r3, x3 | ld.vis.dv.sc0 r2, x2 | | ld.vis.dv.sc0 r2, x2 | ; | ||
ld.vis.dv.sc0 r4, x4 | ld.vis.dv.sc0 r4, x4 | ld.vis.dv.sc0 r3, x3 | | ld.vis.dv.sc0 r3, x3 | ; | ||
ld.vis.dv.sc0 r5, x5 | ld.vis.dv.sc0 r5, x5 | ld.vis.dv.sc0 r4, x4 | | ld.vis.dv.sc0 r4, x4 | ; | ||
| | ld.vis.dv.sc0 r5, x5 | | ld.vis.dv.sc0 r5, x5 | ; | ||
|
||
forall( | ||
P0:r0 == 1 /\ P0:r1 == 1 /\ P0:r2 == 1 /\ P0:r3 == 1 /\ P0:r4 == 1 /\ P0:r5 == 1 /\ | ||
P1:r0 == 1 /\ P1:r1 == 1 /\ P1:r2 == 1 /\ P1:r3 == 1 /\ P1:r4 == 1 /\ P1:r5 == 1 /\ | ||
P2:r0 == 1 /\ P2:r1 == 1 /\ P2:r2 == 1 /\ P2:r3 == 1 /\ P2:r4 == 1 /\ P2:r5 == 1 /\ | ||
P3:r0 == 1 /\ P3:r1 == 1 /\ P3:r2 == 1 /\ P3:r3 == 1 /\ P3:r4 == 1 /\ P3:r5 == 1 /\ | ||
P4:r0 == 1 /\ P4:r1 == 1 /\ P4:r2 == 1 /\ P4:r3 == 1 /\ P4:r4 == 1 /\ P4:r5 == 1 /\ | ||
P5:r0 == 1 /\ P5:r1 == 1 /\ P5:r2 == 1 /\ P5:r3 == 1 /\ P5:r4 == 1 /\ P5:r5 == 1) |
Oops, something went wrong.