Page MenuHomePhabricator
Feed Advanced Search

Wed, Apr 17

tpr committed rG7c55c8d8c347: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)) (authored by tpr).
[AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0))
Wed, Apr 17, 10:26 PM
tpr committed rL358640: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)).
[AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0))
Wed, Apr 17, 10:26 PM
tpr closed D60633: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)).
Wed, Apr 17, 10:26 PM · Restricted Project
tpr committed rG59e8bd3093dc: [AMDGPU] Flag new raw/struct atomic ops as source of divergence (authored by tpr).
[AMDGPU] Flag new raw/struct atomic ops as source of divergence
Wed, Apr 17, 7:03 AM
tpr committed rL358579: [AMDGPU] Flag new raw/struct atomic ops as source of divergence.
[AMDGPU] Flag new raw/struct atomic ops as source of divergence
Wed, Apr 17, 7:03 AM
tpr closed D60731: [AMDGPU] Flag new raw/struct atomic ops as source of divergence.
Wed, Apr 17, 7:03 AM · Restricted Project
tpr added inline comments to D60457: [CodeGen] Fixed de-optimization of legalize subvector extract.
Wed, Apr 17, 3:12 AM · Restricted Project
tpr updated the diff for D60457: [CodeGen] Fixed de-optimization of legalize subvector extract.

V3: Further reduced test case.

Wed, Apr 17, 3:12 AM · Restricted Project
tpr updated the diff for D60457: [CodeGen] Fixed de-optimization of legalize subvector extract.

V2: Addressed review comments.

Wed, Apr 17, 3:05 AM · Restricted Project
tpr updated the diff for D60731: [AMDGPU] Flag new raw/struct atomic ops as source of divergence.

V2: Addressed review comment: moved tests.

Wed, Apr 17, 2:16 AM · Restricted Project

Mon, Apr 15

tpr added reviewers for D60600: [InstCombine] Fix a vector-of-pointers instcombine undef bug.: reames, spatel.
Mon, Apr 15, 2:51 PM · Restricted Project
tpr added a reviewer for D60731: [AMDGPU] Flag new raw/struct atomic ops as source of divergence: nhaehnle.
Mon, Apr 15, 2:30 PM · Restricted Project
tpr created D60731: [AMDGPU] Flag new raw/struct atomic ops as source of divergence.
Mon, Apr 15, 2:30 PM · Restricted Project
tpr added a comment to D60633: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)).

Bug for adding remaining test cases is https://bugs.llvm.org/show_bug.cgi?id=41500

Mon, Apr 15, 7:55 AM · Restricted Project
tpr updated the diff for D60633: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)).

V3: Addressed review comments.

Mon, Apr 15, 7:55 AM · Restricted Project
tpr added a comment to D60682: [AMDGPU] Fixed +DumpCode.

PS I'm only speaking for LLPC. I don't know if Mesa uses it.

Mon, Apr 15, 6:18 AM · Restricted Project
tpr added a comment to D60682: [AMDGPU] Fixed +DumpCode.

I think the only obstacle is getting round to doing the LLPC changes. So this is hopefully a short term fix until we can get around to it.

Mon, Apr 15, 6:18 AM · Restricted Project
tpr committed rG842be38162ec: [AMDGPU] Fixed incorrect test in vcnd/vcmp optimization (authored by tpr).
[AMDGPU] Fixed incorrect test in vcnd/vcmp optimization
Mon, Apr 15, 3:35 AM
tpr committed rL358392: [AMDGPU] Fixed incorrect test in vcnd/vcmp optimization.
[AMDGPU] Fixed incorrect test in vcnd/vcmp optimization
Mon, Apr 15, 3:35 AM
tpr closed D60652: [AMDGPU] Fixed incorrect test in vcnd/vcmp optimization.
Mon, Apr 15, 3:35 AM · Restricted Project
tpr added reviewers for D60682: [AMDGPU] Fixed +DumpCode: scott.linder, kzhuravl.
Mon, Apr 15, 3:18 AM · Restricted Project
tpr created D60682: [AMDGPU] Fixed +DumpCode.
Mon, Apr 15, 3:17 AM · Restricted Project
tpr added a comment to D60633: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)).

I have cut down the test a bit more and put it into fneg-combines.ll. I did not manage to repro any problems with the other cases that I added fixes for.

Mon, Apr 15, 3:06 AM · Restricted Project
tpr updated the diff for D60633: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)).

V2: Cut down test a bit more and put it in fneg-combines.ll.

Mon, Apr 15, 3:03 AM · Restricted Project

Sat, Apr 13

tpr added reviewers for D60652: [AMDGPU] Fixed incorrect test in vcnd/vcmp optimization: vpykhtin, rampitec.

I did not add a test because I believe that we do not currently set the source modifiers in codegen, so this code never gets exercised.

Sat, Apr 13, 9:36 AM · Restricted Project
tpr created D60652: [AMDGPU] Fixed incorrect test in vcnd/vcmp optimization.
Sat, Apr 13, 9:32 AM · Restricted Project

Fri, Apr 12

tpr added reviewers for D60633: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)): arsenm, rampitec.

That's the best bugpoint could do with the test.

Fri, Apr 12, 1:22 PM · Restricted Project
tpr created D60633: [AMDGPU] Avoid DAG combining assert with fneg(fadd(A,0)).
Fri, Apr 12, 1:16 PM · Restricted Project

Tue, Apr 9

tpr added reviewers for D60457: [CodeGen] Fixed de-optimization of legalize subvector extract: hakzsam, arsenm, efriedma.
Tue, Apr 9, 6:10 AM · Restricted Project
tpr added a comment to D58902: [AMDGPU] Support for v3i32/v3f32.

Hopefully fixed by D60457.

Tue, Apr 9, 5:57 AM · Restricted Project
tpr created D60457: [CodeGen] Fixed de-optimization of legalize subvector extract.
Tue, Apr 9, 5:50 AM · Restricted Project

Mon, Apr 1

tpr added a comment to D58902: [AMDGPU] Support for v3i32/v3f32.

Hi Samuel

Mon, Apr 1, 2:11 AM · Restricted Project

Fri, Mar 29

tpr accepted D59295: [AMDGPU] Pre-allocate WWM registers to reduce VGPR pressure..

LGTM modulo the wrong license on the new file.

Fri, Mar 29, 1:53 PM · Restricted Project, Restricted Project

Wed, Mar 27

tpr abandoned D59871: [AMDGPU] Avoid mistaking a ds atomic for a load.

OK thanks Matt.

Wed, Mar 27, 9:46 AM · Restricted Project
tpr added a reviewer for D59871: [AMDGPU] Avoid mistaking a ds atomic for a load: arsenm.
Wed, Mar 27, 2:14 AM · Restricted Project
tpr created D59871: [AMDGPU] Avoid mistaking a ds atomic for a load.
Wed, Mar 27, 2:14 AM · Restricted Project

Fri, Mar 22

tpr committed rG94c163c34e4d: InstCombineSimplifyDemanded: Allow v3 results for AMDGCN buffer and image… (authored by tpr).
InstCombineSimplifyDemanded: Allow v3 results for AMDGCN buffer and image…
Fri, Mar 22, 8:55 AM
tpr committed rL356768: InstCombineSimplifyDemanded: Allow v3 results for AMDGCN buffer and image….
InstCombineSimplifyDemanded: Allow v3 results for AMDGCN buffer and image…
Fri, Mar 22, 8:55 AM
tpr closed D58906: InstCombineSimplifyDemanded: Allow v3 results for amdgcn buffer and image intrinsics.
Fri, Mar 22, 8:54 AM · Restricted Project
tpr committed rG6f0191a55a53: [AMDGPU] Use three- and five-dword result type in image ops (authored by tpr).
[AMDGPU] Use three- and five-dword result type in image ops
Fri, Mar 22, 8:20 AM
tpr committed rL356757: [AMDGPU] Use three- and five-dword result type in image ops.
[AMDGPU] Use three- and five-dword result type in image ops
Fri, Mar 22, 8:20 AM
tpr closed D58905: [AMDGPU] Use three- and five-dword result type in image ops.
Fri, Mar 22, 8:20 AM · Restricted Project
tpr committed rG677387d8dc9e: [AMDGPU] Implemented dwordx3 variants of buffer/tbuffer load/store intrinsics (authored by tpr).
[AMDGPU] Implemented dwordx3 variants of buffer/tbuffer load/store intrinsics
Fri, Mar 22, 7:58 AM
tpr committed rL356755: [AMDGPU] Implemented dwordx3 variants of buffer/tbuffer load/store intrinsics.
[AMDGPU] Implemented dwordx3 variants of buffer/tbuffer load/store intrinsics
Fri, Mar 22, 7:57 AM
tpr closed D58904: [AMDGPU] Implemented dwordx3 variants of buffer/tbuffer load/store intrinsics.
Fri, Mar 22, 7:57 AM · Restricted Project
tpr committed rG033f99a2e567: [AMDGPU] Added v5i32 and v5f32 register classes (authored by tpr).
[AMDGPU] Added v5i32 and v5f32 register classes
Fri, Mar 22, 3:11 AM
tpr committed rL356735: [AMDGPU] Added v5i32 and v5f32 register classes.
[AMDGPU] Added v5i32 and v5f32 register classes
Fri, Mar 22, 3:10 AM
tpr closed D58903: [AMDGPU] Added v5i32 and v5f32 register classes.
Fri, Mar 22, 3:10 AM · Restricted Project

Mar 21 2019

tpr committed rG361b5b219342: [AMDGPU] Support for v3i32/v3f32 (authored by tpr).
[AMDGPU] Support for v3i32/v3f32
Mar 21 2019, 5:01 AM
tpr committed rL356659: [AMDGPU] Support for v3i32/v3f32.
[AMDGPU] Support for v3i32/v3f32
Mar 21 2019, 5:00 AM
tpr closed D58902: [AMDGPU] Support for v3i32/v3f32.
Mar 21 2019, 5:00 AM · Restricted Project

Mar 20 2019

tpr committed rG2327c231d662: [AMDGPU] Do not generate spurious PAL metadata (authored by tpr).
[AMDGPU] Do not generate spurious PAL metadata
Mar 20 2019, 3:02 PM
tpr committed rL356621: [AMDGPU] Do not generate spurious PAL metadata.
[AMDGPU] Do not generate spurious PAL metadata
Mar 20 2019, 3:01 PM
tpr closed D59613: [AMDGPU] Do not generate spurious PAL metadata.
Mar 20 2019, 3:00 PM · Restricted Project
tpr added reviewers for D59613: [AMDGPU] Do not generate spurious PAL metadata: scott.linder, kzhuravl, hliao.
Mar 20 2019, 2:13 PM · Restricted Project
tpr created D59613: [AMDGPU] Do not generate spurious PAL metadata.
Mar 20 2019, 2:10 PM · Restricted Project
tpr abandoned D59556: [AMDGPU] Fixed i64 add/sub used in lowering of i64 srem.

Thanks for the better test Matt. But I'll abandon this one in favor of Michael's improved fix D59608.

Mar 20 2019, 1:15 PM · Restricted Project
tpr accepted D59608: [AMDGPU] Fix clamp bit DAG operand.

Thanks Michael; that is much better than my fix in D59556. Let's go with this one and I'll abandon my one.

Mar 20 2019, 1:14 PM · Restricted Project
tpr committed rGe7bd52f86e04: [AMDGPU] Added MsgPack format PAL metadata (authored by tpr).
[AMDGPU] Added MsgPack format PAL metadata
Mar 20 2019, 11:48 AM
tpr committed rL356591: [AMDGPU] Added MsgPack format PAL metadata.
[AMDGPU] Added MsgPack format PAL metadata
Mar 20 2019, 11:48 AM
tpr closed D57028: [AMDGPU] Added MsgPack format PAL metadata.
Mar 20 2019, 11:48 AM · Restricted Project
tpr committed rGd737b551e97e: [AMDGPU] Factored PAL metadata handling out into its own class (authored by tpr).
[AMDGPU] Factored PAL metadata handling out into its own class
Mar 20 2019, 10:42 AM
tpr committed rL356582: [AMDGPU] Factored PAL metadata handling out into its own class.
[AMDGPU] Factored PAL metadata handling out into its own class
Mar 20 2019, 10:42 AM
tpr closed D57027: [AMDGPU] Factored PAL metadata handling out into its own class.
Mar 20 2019, 10:42 AM · Restricted Project
tpr added a comment to D59556: [AMDGPU] Fixed i64 add/sub used in lowering of i64 srem.

The test is already reduced as much as I can. Removing anything in there makes the problem disappear. Constructing a new test case using llvm.uadd.with.overflow does not show the problem. Can we go with this test case?

Mar 20 2019, 7:03 AM · Restricted Project
tpr updated the diff for D57028: [AMDGPU] Added MsgPack format PAL metadata.

V4: Generate vendor=AMDGPU, type=32 for the metadata note record, as now expected by PAL.

Mar 20 2019, 5:19 AM · Restricted Project

Mar 19 2019

tpr added a reviewer for D59556: [AMDGPU] Fixed i64 add/sub used in lowering of i64 srem: michel.daenzer.
Mar 19 2019, 1:10 PM · Restricted Project
tpr created D59556: [AMDGPU] Fixed i64 add/sub used in lowering of i64 srem.
Mar 19 2019, 1:07 PM · Restricted Project
tpr updated the diff for D57028: [AMDGPU] Added MsgPack format PAL metadata.

V3: Rebased, and addressed "single char in single quotes" review comment from the other change.

Mar 19 2019, 12:57 AM · Restricted Project
tpr added inline comments to D57027: [AMDGPU] Factored PAL metadata handling out into its own class.
Mar 19 2019, 12:47 AM · Restricted Project
tpr updated the diff for D57027: [AMDGPU] Factored PAL metadata handling out into its own class.

V3: Fixed review comments.

Mar 19 2019, 12:45 AM · Restricted Project

Mar 18 2019

tpr committed rGcfdfba996b08: [AMDGPU] Asm/disasm clamp modifier on vop3 int arithmetic (authored by tpr).
[AMDGPU] Asm/disasm clamp modifier on vop3 int arithmetic
Mar 18 2019, 12:36 PM
tpr committed rL356399: [AMDGPU] Asm/disasm clamp modifier on vop3 int arithmetic.
[AMDGPU] Asm/disasm clamp modifier on vop3 int arithmetic
Mar 18 2019, 12:36 PM
tpr closed D59267: [AMDGPU] Asm/disasm clamp modifier on vop3 int arithmetic.
Mar 18 2019, 12:36 PM · Restricted Project
tpr committed rG2e94f6e584d9: [AMDGPU] Asm/disasm v_cndmask_b32_e64 with abs/neg source modifiers (authored by tpr).
[AMDGPU] Asm/disasm v_cndmask_b32_e64 with abs/neg source modifiers
Mar 18 2019, 12:25 PM
tpr committed rL356398: [AMDGPU] Asm/disasm v_cndmask_b32_e64 with abs/neg source modifiers.
[AMDGPU] Asm/disasm v_cndmask_b32_e64 with abs/neg source modifiers
Mar 18 2019, 12:25 PM
tpr closed D59191: [AMDGPU] Asm/disasm v_cndmask_b32_e64 with abs/neg source modifiers.
Mar 18 2019, 12:24 PM · Restricted Project
tpr committed rG8723a5655172: [MsgPack][AMDGPU] Fix unflushed raw_string_ostream bugs on windows expensive… (authored by tpr).
[MsgPack][AMDGPU] Fix unflushed raw_string_ostream bugs on windows expensive…
Mar 18 2019, 12:01 PM
tpr committed rL356394: [MsgPack][AMDGPU] Fix unflushed raw_string_ostream bugs on windows expensive….
[MsgPack][AMDGPU] Fix unflushed raw_string_ostream bugs on windows expensive…
Mar 18 2019, 12:00 PM
tpr closed D59396: [MsgPack] Attempt to fix failure on windows expensive checks bot.
Mar 18 2019, 12:00 PM · Restricted Project
tpr added inline comments to D57027: [AMDGPU] Factored PAL metadata handling out into its own class.
Mar 18 2019, 8:56 AM · Restricted Project
tpr updated the diff for D57028: [AMDGPU] Added MsgPack format PAL metadata.

V2: Like the previous commit, no longer supports a method for LLPC to

call to write the PAL metadata into IR metadata. The plan now is
that LLPC will use MsgPackDocument and put the msgpack binary blob
into IR metadata itself.
Mar 18 2019, 8:55 AM · Restricted Project
tpr updated the diff for D57027: [AMDGPU] Factored PAL metadata handling out into its own class.

V2: Removed AMDGPUPALMetadata API used directly by LLPC. The plan now is

that LLPC will continue to put the PAL metadata binary blob
into IR metadata itself.
Mar 18 2019, 8:52 AM · Restricted Project

Mar 17 2019

tpr committed rGc4e128e2219b: [CodeGen] Defined MVTs v3i32, v3f32, v5i32, v5f32 (authored by tpr).
[CodeGen] Defined MVTs v3i32, v3f32, v5i32, v5f32
Mar 17 2019, 3:57 PM
tpr committed rL356351: [CodeGen] Defined MVTs v3i32, v3f32, v5i32, v5f32.
[CodeGen] Defined MVTs v3i32, v3f32, v5i32, v5f32
Mar 17 2019, 3:57 PM
tpr closed D58901: [CodeGen] Defined MVTs v3i32, v3f32, v5i32, v5f32.
Mar 17 2019, 3:56 PM · Restricted Project
tpr committed rGc302b9b5fe0e: [CodeGen] Prepare for introduction of v3 and v5 MVTs (authored by tpr).
[CodeGen] Prepare for introduction of v3 and v5 MVTs
Mar 17 2019, 2:42 PM
tpr committed rL356350: [CodeGen] Prepare for introduction of v3 and v5 MVTs.
[CodeGen] Prepare for introduction of v3 and v5 MVTs
Mar 17 2019, 2:42 PM
tpr closed D58899: [CodeGen] Prepare for introduction of v3 and v5 MVTs.
Mar 17 2019, 2:42 PM · Restricted Project
tpr committed rGe30aa6a13623: [AMDGPU] Prepare for introduction of v3 and v5 MVTs (authored by tpr).
[AMDGPU] Prepare for introduction of v3 and v5 MVTs
Mar 17 2019, 2:03 PM
tpr committed rL356342: [AMDGPU] Prepare for introduction of v3 and v5 MVTs.
[AMDGPU] Prepare for introduction of v3 and v5 MVTs
Mar 17 2019, 2:03 PM
tpr closed D58928: [AMDGPU] Prepare for introduction of v3 and v5 MVTs.
Mar 17 2019, 2:03 PM · Restricted Project
tpr committed rGd1477e989cef: [ARM] Fixed an assumption of power-of-2 vector MVT (authored by tpr).
[ARM] Fixed an assumption of power-of-2 vector MVT
Mar 17 2019, 1:48 PM
tpr committed rL356341: [ARM] Fixed an assumption of power-of-2 vector MVT.
[ARM] Fixed an assumption of power-of-2 vector MVT
Mar 17 2019, 1:47 PM
tpr closed D58927: [ARM] Fixed an assumption of power-of-2 vector MVT.
Mar 17 2019, 1:47 PM · Restricted Project

Mar 16 2019

tpr updated the diff for D58902: [AMDGPU] Support for v3i32/v3f32.

V5: Added vec3 to AMDGPUCallingConvention.td. Fixed call-return-types test for dwordx3 buffer/flat.
(I will add vec5 to AMDGPUCallingConvention.td in the vec5 change.)

Mar 16 2019, 11:55 AM · Restricted Project
tpr updated the diff for D58928: [AMDGPU] Prepare for introduction of v3 and v5 MVTs.

V5: Added vec3/vec5 non-shader arg and ret type test cases, including passing by stack.

Mar 16 2019, 11:49 AM · Restricted Project
tpr added inline comments to D58927: [ARM] Fixed an assumption of power-of-2 vector MVT.
Mar 16 2019, 6:09 AM · Restricted Project
tpr updated the diff for D58927: [ARM] Fixed an assumption of power-of-2 vector MVT.

V4: Properly addressed review comment.

Mar 16 2019, 6:09 AM · Restricted Project
tpr updated the diff for D59396: [MsgPack] Attempt to fix failure on windows expensive checks bot.

V3: Adjusted the MIR roundtrip fix to use scoping instead of explicit flush.

Mar 16 2019, 5:26 AM · Restricted Project