Page MenuHomePhabricator
Feed Advanced Search

Today

critson added a comment to D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.

Sorry, I just realized you had not accepted this when I landed it (after addressing the nit). Please let me know if you have any concerns.

Mon, Aug 2, 5:11 AM · Restricted Project
critson committed rG675c9423736d: [AMDGPU] Disable NSA for BVH instructions when appropriate (authored by critson).
[AMDGPU] Disable NSA for BVH instructions when appropriate
Mon, Aug 2, 4:11 AM
critson closed D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
Mon, Aug 2, 4:11 AM · Restricted Project

Yesterday

critson committed rGa441de6d94dc: [AMDGPU][GlobalISel] Add missing default mapping for BVH intrinsics (authored by critson).
[AMDGPU][GlobalISel] Add missing default mapping for BVH intrinsics
Sun, Aug 1, 8:44 PM
critson closed D107211: [AMDGPU][GlobalISel] Add missing default mapping for BVH intrinsics.
Sun, Aug 1, 8:44 PM · Restricted Project

Fri, Jul 30

critson updated the diff for D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
  • Rebase
Fri, Jul 30, 9:23 PM · Restricted Project
critson requested review of D107211: [AMDGPU][GlobalISel] Add missing default mapping for BVH intrinsics.
Fri, Jul 30, 9:22 PM · Restricted Project
critson updated the diff for D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
  • Address reviewer comments
Fri, Jul 30, 12:59 AM · Restricted Project
critson added inline comments to D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
Fri, Jul 30, 12:59 AM · Restricted Project

Tue, Jul 27

critson added inline comments to D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
Tue, Jul 27, 1:29 AM · Restricted Project
critson updated the diff for D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
  • Remove unnecessary padding.
  • Properly set bank size
Tue, Jul 27, 1:25 AM · Restricted Project

Mon, Jul 26

critson committed rGfbaa35e16947: [AMDGPU] Add SelectionDAG support for insert_subvector on v4f64 (authored by critson).
[AMDGPU] Add SelectionDAG support for insert_subvector on v4f64
Mon, Jul 26, 6:12 PM
critson closed D105385: [AMDGPU] Add SelectionDAG support for insert_subvector on v4f64.
Mon, Jul 26, 6:11 PM · Restricted Project

Sun, Jul 25

critson updated the diff for D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.

Ping.
It would be good to get this merged now the other NSA limiting features are in.

Sun, Jul 25, 11:59 PM · Restricted Project

Fri, Jul 23

critson committed rG7d4baf25aae5: [AMDGPU] Add maximum NSA size limit ISA feature (authored by critson).
[AMDGPU] Add maximum NSA size limit ISA feature
Fri, Jul 23, 12:17 AM
critson closed D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
Fri, Jul 23, 12:17 AM · Restricted Project

Thu, Jul 22

critson added a comment to D105385: [AMDGPU] Add SelectionDAG support for insert_subvector on v4f64.

Should I find more reviewers?
Not sure who is versed in Selection DAG?

Thu, Jul 22, 1:37 AM · Restricted Project

Wed, Jul 21

critson added inline comments to D106471: [DAGCombiner][RISCV][AMDGPU] Call SimplifyDemandedBits at the end of visitMULHU to enable known bits contant folding..
Wed, Jul 21, 8:03 PM · Restricted Project
critson added inline comments to D106471: [DAGCombiner][RISCV][AMDGPU] Call SimplifyDemandedBits at the end of visitMULHU to enable known bits contant folding..
Wed, Jul 21, 7:36 PM · Restricted Project
critson committed rG6efb3220b472: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions (authored by critson).
[AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions
Wed, Jul 21, 6:43 PM
critson closed D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.
Wed, Jul 21, 6:42 PM · Restricted Project
critson committed rG9dcd75f86f24: [AMDGPU] Allow frontends to disable null export for pixel shaders (authored by critson).
[AMDGPU] Allow frontends to disable null export for pixel shaders
Wed, Jul 21, 6:21 PM
critson closed D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
Wed, Jul 21, 6:21 PM · Restricted Project

Tue, Jul 20

critson added a comment to D106176: [Scheduler] Treat weak edges uniformly at entry.

Is it possible to add a test case?

Do you know a way to create a synthetic schedule DAG for testing? Without that, I would say no, since I don't think we want to create a new schedule mutator purely for testing purposes?

Tue, Jul 20, 7:21 PM · Restricted Project
critson accepted D106176: [Scheduler] Treat weak edges uniformly at entry.

I agree manipulation of the available queue here needs to be made consistent with the scheduling loop below it.

Tue, Jul 20, 1:10 AM · Restricted Project

Mon, Jul 19

critson updated the diff for D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
  • Fix typo in comment
Mon, Jul 19, 1:09 AM · Restricted Project

Fri, Jul 16

critson added a comment to D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.

@arsenm are you happy for this to proceed now?

Fri, Jul 16, 7:39 PM · Restricted Project
critson updated the diff for D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
  • Add comments
Fri, Jul 16, 7:37 PM · Restricted Project
critson committed rGc7f2f81f5e2a: [AMDGPU] Tidy SReg/SGPR definitions using template class (authored by critson).
[AMDGPU] Tidy SReg/SGPR definitions using template class
Fri, Jul 16, 7:27 PM
critson closed D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
Fri, Jul 16, 7:27 PM · Restricted Project
critson updated the diff for D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
  • Address nits
  • Fix allocation priority of SReg_224
Fri, Jul 16, 7:12 PM · Restricted Project
critson added inline comments to D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
Fri, Jul 16, 7:12 PM · Restricted Project
critson added a comment to D106079: [AMDGPU] Divergence-driven compare operations instruction selection.

Can you explain the motivation for this change further?
Particularly as it increases generated code size.

Fri, Jul 16, 1:01 AM · Restricted Project

Thu, Jul 15

critson updated the summary of D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
Thu, Jul 15, 9:45 PM · Restricted Project
critson updated the diff for D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
  • Rework based on TableGen changes in D105967
  • Define all applicable TTMP register classes
Thu, Jul 15, 9:45 PM · Restricted Project
critson committed rGf5917e0312ed: [TableGen] Allow isAllocatable inheritence from any superclass (authored by critson).
[TableGen] Allow isAllocatable inheritence from any superclass
Thu, Jul 15, 9:03 PM
critson closed D105967: [TableGen] Allow isAllocatable inheritence from any superclass.
Thu, Jul 15, 9:03 PM · Restricted Project

Wed, Jul 14

critson added a comment to D105967: [TableGen] Allow isAllocatable inheritence from any superclass.

I think it's ok, but it makes me wonder about the other properties. If this one can vary depending on the topological ordering, can VTs or CopyCost (for example) change too? Logically, VTs should be the intersection of VTs over all super classes, other properties may need different treatment.

Wed, Jul 14, 6:09 PM · Restricted Project
critson updated the diff for D105967: [TableGen] Allow isAllocatable inheritence from any superclass.
  • Rework to use any_of
Wed, Jul 14, 6:04 PM · Restricted Project
critson added a comment to D101633: [AMDGPU] Set number vgprs used in PS shaders based on input registers actually used.

To confirm, this reduces the minimum number of VGPRs used by PS?
I am not sure the language in the description makes this clear.

Wed, Jul 14, 1:30 AM · Restricted Project
critson added a comment to D105385: [AMDGPU] Add SelectionDAG support for insert_subvector on v4f64.

Ping

Wed, Jul 14, 12:58 AM · Restricted Project
critson requested review of D105967: [TableGen] Allow isAllocatable inheritence from any superclass.
Wed, Jul 14, 12:58 AM · Restricted Project

Tue, Jul 13

critson updated the summary of D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
Tue, Jul 13, 8:30 PM · Restricted Project
critson updated the diff for D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
  • Rename function attributes to amdgpu-color-export and amdgpu-depth-export
Tue, Jul 13, 8:28 PM · Restricted Project
critson added a comment to D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.

Should we be defining TTMP classes for all sizes instead of working around it like this?

Tue, Jul 13, 7:55 PM · Restricted Project
critson added inline comments to D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
Tue, Jul 13, 2:25 AM · Restricted Project
critson updated the diff for D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
  • Address reviewer comments
Tue, Jul 13, 2:23 AM · Restricted Project

Mon, Jul 12

critson updated the summary of D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
Mon, Jul 12, 8:42 PM · Restricted Project
critson updated the diff for D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
  • Rename attributes to HasColorExport and HasDepthExport
Mon, Jul 12, 8:42 PM · Restricted Project
critson accepted D105610: [AMDGPU] Don't handle export done when unify exit nodes.

Please confirm it passes the usual graphics tests.

Mon, Jul 12, 7:45 PM · Restricted Project
critson updated the diff for D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
  • Only allow no export PS on GFX10
Mon, Jul 12, 2:17 AM · Restricted Project
critson requested review of D105800: [AMDGPU] Tidy SReg/SGPR definitions using template class.
Mon, Jul 12, 1:42 AM · Restricted Project

Sun, Jul 11

critson added inline comments to D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
Sun, Jul 11, 6:28 PM · Restricted Project

Fri, Jul 9

critson added inline comments to D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
Fri, Jul 9, 3:32 AM · Restricted Project
critson abandoned D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.
Fri, Jul 9, 1:58 AM · Restricted Project
critson requested review of D105683: [AMDGPU] Allow frontends to disable null export for pixel shaders.
Fri, Jul 9, 1:56 AM · Restricted Project

Thu, Jul 8

critson added a comment to D105610: [AMDGPU] Don't handle export done when unify exit nodes.

From my own recent work, I think this should be fine.
The only cases where a null export are "required" occur in control flow following an infinite loop, which obviously is never executed.

Thu, Jul 8, 1:39 AM · Restricted Project
critson added a comment to D105610: [AMDGPU] Don't handle export done when unify exit nodes.

This is an alternative to D102830 ?

Thu, Jul 8, 12:13 AM · Restricted Project

Tue, Jul 6

critson updated the diff for D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.
  • Remove redundant comments
Tue, Jul 6, 8:02 PM · Restricted Project
critson updated the diff for D105385: [AMDGPU] Add SelectionDAG support for insert_subvector on v4f64.
  • llvm-reduce tests
Tue, Jul 6, 8:01 PM · Restricted Project
critson updated the summary of D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.
Tue, Jul 6, 4:15 AM · Restricted Project
critson updated the diff for D102830: [AMDGPU] Avoid null export insertion when unifying exit blocks.

Rework based on the assumption there is only one "done" export.
This is true for existing front-ends.

Tue, Jul 6, 4:11 AM · Restricted Project

Jul 2 2021

critson requested review of D105385: [AMDGPU] Add SelectionDAG support for insert_subvector on v4f64.
Jul 2 2021, 6:47 PM · Restricted Project

Jun 24 2021

critson added inline comments to D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.
Jun 24 2021, 5:09 AM · Restricted Project
critson updated the diff for D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.
  • Refactor tablegen code in MIMG
Jun 24 2021, 5:08 AM · Restricted Project

Jun 23 2021

critson added a comment to D104622: [AMDGPU] Add 224-bit vector types and link 192-bit types to MVTs.

I have committed this but I am happy to add the TTMP classes as a follow up.
I was unsure whether it was appropriate to add more classes that probably are not even used?

Jun 23 2021, 8:43 PM · Restricted Project
critson committed rG98f48723f2ca: [AMDGPU] Add 224-bit vector types and link 192-bit types to MVTs (authored by critson).
[AMDGPU] Add 224-bit vector types and link 192-bit types to MVTs
Jun 23 2021, 8:42 PM
critson committed rG6b0f98d442e0: [ValueTypes] Define MVTs for v3i64/v3f64 to complement v6i32/v6f32 (authored by critson).
[ValueTypes] Define MVTs for v3i64/v3f64 to complement v6i32/v6f32
Jun 23 2021, 8:42 PM
critson closed D104622: [AMDGPU] Add 224-bit vector types and link 192-bit types to MVTs.
Jun 23 2021, 8:42 PM · Restricted Project
critson closed D104621: [ValueTypes] Define MVTs for v3i64/v3f64 to complement v6i32/v6f32.
Jun 23 2021, 8:41 PM · Restricted Project
critson committed rGae266e743c91: [LVI] Remove recursion from getValueForCondition (NFCI) (authored by critson).
[LVI] Remove recursion from getValueForCondition (NFCI)
Jun 23 2021, 5:59 PM
critson closed D104191: [LVI] Remove recursion from getValueForCondition (NFC).
Jun 23 2021, 5:58 PM · Restricted Project

Jun 22 2021

critson added a comment to D104622: [AMDGPU] Add 224-bit vector types and link 192-bit types to MVTs.

LGTM, although v3 change can have unexpected impact. Could you please run PSDB on this?

Jun 22 2021, 6:01 PM · Restricted Project
critson added inline comments to D104621: [ValueTypes] Define MVTs for v3i64/v3f64 to complement v6i32/v6f32.
Jun 22 2021, 5:59 PM · Restricted Project
critson updated the diff for D104621: [ValueTypes] Define MVTs for v3i64/v3f64 to complement v6i32/v6f32.
  • Address reviewer comments.
Jun 22 2021, 5:59 PM · Restricted Project

Jun 21 2021

critson updated the diff for D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.
  • Address reviewer comments
Jun 21 2021, 7:38 PM · Restricted Project
critson added inline comments to D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.
Jun 21 2021, 7:37 PM · Restricted Project
critson updated the diff for D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.
  • Incorporate reviewer feedback
Jun 21 2021, 7:33 PM · Restricted Project
critson added inline comments to D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
Jun 21 2021, 3:59 AM · Restricted Project
critson updated the diff for D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
  • Remove casts
Jun 21 2021, 3:59 AM · Restricted Project
critson updated the diff for D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
  • Use array for opcode look up
Jun 21 2021, 3:52 AM · Restricted Project
critson added a comment to D104191: [LVI] Remove recursion from getValueForCondition (NFC).

Ping.
As this is NFC, I will submit in a few days if there are no comments.

Jun 21 2021, 1:23 AM · Restricted Project
critson retitled D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate from [AMDGPU] Add options to disable NSA for BVH instructions to [AMDGPU] Disable NSA for BVH instructions when appropriate.
Jun 21 2021, 1:23 AM · Restricted Project
critson updated the diff for D103230: [AMDGPU] Disable NSA for BVH instructions when appropriate.
  • Rework this based on NSA limit functionality
Jun 21 2021, 1:22 AM · Restricted Project
critson updated the diff for D103348: [AMDGPU] Add maximum NSA size limit ISA feature.
  • Simplify based on MIMG support for v5/v6/v7
  • Update with gfx1013 target
Jun 21 2021, 1:21 AM · Restricted Project
critson retitled D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions from [AMDGPU] Add VReg_192 support for MIMG instructions to [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.
Jun 21 2021, 1:19 AM · Restricted Project
critson updated the diff for D103800: [AMDGPU] Add VReg_192/VReg_224 support for MIMG instructions.

Update based on new MVT type support.

Jun 21 2021, 1:19 AM · Restricted Project
critson requested review of D104622: [AMDGPU] Add 224-bit vector types and link 192-bit types to MVTs.
Jun 21 2021, 1:16 AM · Restricted Project
critson requested review of D104621: [ValueTypes] Define MVTs for v3i64/v3f64 to complement v6i32/v6f32.
Jun 21 2021, 1:13 AM · Restricted Project

Jun 17 2021

critson committed rGa10aeb3b32b5: [AMDGPU] Remove duplicate setOperationAction for v4i16/v4f16 (NFC) (authored by critson).
[AMDGPU] Remove duplicate setOperationAction for v4i16/v4f16 (NFC)
Jun 17 2021, 8:39 PM

Jun 15 2021

critson added a comment to D103881: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32.

Since there is a bug and a repro case, I'd like to ask if this change can be rolled back until a proper fix is in order.

It looks like @craig.topper fixed the bug with rG4017d0335a35334835bfae6fc3e258adcd9ed2dc so probably no need for a revert

Jun 15 2021, 1:36 AM · Restricted Project

Jun 12 2021

critson requested review of D104191: [LVI] Remove recursion from getValueForCondition (NFC).
Jun 12 2021, 11:37 PM · Restricted Project

Jun 10 2021

critson committed rG2c2d2922a24b: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32 (authored by critson).
[ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32
Jun 10 2021, 4:59 PM
critson committed rGcfbb92441f17: [SDAG] Fix pow2 assumption when splitting vectors (authored by critson).
[SDAG] Fix pow2 assumption when splitting vectors
Jun 10 2021, 4:59 PM
critson closed D103881: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32.
Jun 10 2021, 4:59 PM · Restricted Project
critson closed D103880: [SDAG] Fix pow2 assumption when splitting vectors.
Jun 10 2021, 4:59 PM · Restricted Project
critson updated the diff for D103880: [SDAG] Fix pow2 assumption when splitting vectors.
  • Replace getVectorMinNumElements with getVectorNumElements
Jun 10 2021, 3:54 AM · Restricted Project

Jun 9 2021

critson added a comment to D103880: [SDAG] Fix pow2 assumption when splitting vectors.

test case?

Jun 9 2021, 8:03 PM · Restricted Project
critson accepted D102212: [AMDGPU] Add Optimize VGPR LiveRange Pass..

LGTM, with one nit

Jun 9 2021, 6:59 PM · Restricted Project
critson updated the diff for D103881: [ValueTypes] Define MVTs for v6i32, v6f32, v7i32, v7f32.
  • Rebase
Jun 9 2021, 4:47 AM · Restricted Project