Page MenuHomePhabricator

matejam (Mateja Marjanovic)
User

Projects

User does not belong to any projects.

User Details

User Since
Nov 4 2019, 3:49 AM (138 w, 4 d)

Recent Activity

Feb 25 2022

matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Remove a lambda that is used only once and a part of the code that covers only a specially constructed test.

Feb 25 2022, 9:57 AM · Restricted Project
matejam added inline comments to D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.
Feb 25 2022, 9:33 AM · Restricted Project

Feb 24 2022

matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Typos in tests and a few minor changes.

Feb 24 2022, 8:31 AM · Restricted Project
matejam added inline comments to D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.
Feb 24 2022, 8:25 AM · Restricted Project
matejam added inline comments to D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.
Feb 24 2022, 8:21 AM · Restricted Project
matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Minor changes.

Feb 24 2022, 5:06 AM · Restricted Project
matejam added inline comments to D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.
Feb 24 2022, 3:12 AM · Restricted Project

Feb 22 2022

matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Code formatting.

Feb 22 2022, 9:00 AM · Restricted Project
matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Instead of a combiner in regbankselect pass, add a combiner in amdgpu-postlegalizer-combiner pass.

Feb 22 2022, 8:57 AM · Restricted Project

Feb 4 2022

matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Narrow capture lists in lambdas.

Feb 4 2022, 6:43 AM · Restricted Project
matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Instead of using a combiner in the legalizer, change the implementation of apply mapping for G_BUILD_VECTOR_TRUNC in regbankselect.

Feb 4 2022, 4:38 AM · Restricted Project

Jan 27 2022

matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Added .mir test.

Jan 27 2022, 5:17 AM · Restricted Project
matejam updated the diff for D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.

Minor change in the patch format.

Jan 27 2022, 4:59 AM · Restricted Project
matejam requested review of D118351: [AMDGPU][GlobalISel] Code quality: don't expand G_BUILD_VECTOR_TRUNC if not neccessary.
Jan 27 2022, 4:55 AM · Restricted Project

Nov 30 2021

matejam closed D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

commit ca57b80cd6767b97477fd157831a2b099b5f8f75

Nov 30 2021, 8:29 AM · Restricted Project
matejam committed rGca57b80cd676: Code quality: Combine V_RSQ (authored by matejam).
Code quality: Combine V_RSQ
Nov 30 2021, 8:18 AM
matejam committed rG0c1b18f0f3e5: Test commit (authored by matejam).
Test commit
Nov 30 2021, 6:06 AM

Nov 29 2021

matejam updated the diff for D98050: [AMDGPU][GlobalISel] Transform (fsub (fpext (fneg (fmul x, y))), z) -> (fneg (fma (fpext x), (fpext y), z)).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:22 AM · Restricted Project
matejam updated the diff for D98049: [AMDGPU][GlobalISel] Transform (fsub (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), (fneg z)).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:21 AM · Restricted Project
matejam updated the diff for D98048: [AMDGPU][GlobalISel] Transform (fsub (fneg (fmul, x, y)), z) -> (fma (fneg x), y, (fneg z)).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:21 AM · Restricted Project
matejam updated the diff for D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:20 AM · Restricted Project
matejam updated the diff for D98047: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fpext (fmul u, v))), z) -> (fma x, y, (fma (fpext u), (fpext v), z)).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:10 AM · Restricted Project
matejam updated the diff for D97938: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y, (fma u, v, z)).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:09 AM · Restricted Project
matejam updated the diff for D97937: [AMDGPU][GlobalISel] Transform (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:09 AM · Restricted Project
matejam updated the diff for D93305: [AMDGPU][GlobalISel] Transform (fadd (fmul x, y), z) -> (fma x, y, z).

Added missing comments and an FMA combiner group in table gen.

Nov 29 2021, 6:09 AM · Restricted Project

Nov 26 2021

matejam updated the diff for D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

Added .ll test. Don't cover the G_FDIV + G_FSQRT case, only with rcp intrinsic (by the time it gets to the postlegalizer it will be transformed to that).

Nov 26 2021, 7:52 AM · Restricted Project

Nov 4 2021

matejam updated the diff for D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Minor changes in tests.

Nov 4 2021, 8:28 AM · Restricted Project
matejam added a comment to D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

Added implementation for all possible cases which should be combined into rsq (rcp(sqrt(x)), sqrt(rcp(x)), 1/sqrt(x), sqrt(1/x)).

I thought this would be two separate combines:

  1. (1.0 / x) -> (rcp x)
  2. (sqrt (rcp x)) or (rcp (sqrt x)) -> (rsq x)

Is there some reason we don't implement the first combine, e.g. because of the precision of the rcp instruction is not good enough? What does SelectionDAG do?

Nov 4 2021, 5:29 AM · Restricted Project

Nov 3 2021

matejam updated the diff for D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

Formatting.

Nov 3 2021, 10:27 AM · Restricted Project
matejam updated the diff for D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

Formatting.

Nov 3 2021, 9:03 AM · Restricted Project
matejam updated the diff for D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

Added implementation for all possible cases which should be combined into rsq (rcp(sqrt(x)), sqrt(rcp(x)), 1/sqrt(x), sqrt(1/x)).

Nov 3 2021, 6:31 AM · Restricted Project

Oct 28 2021

matejam updated the diff for D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

Delete RsqPat pattern definition and uses and copy the flags from the original instruction to the newly built instruction (fast math flags...).

Oct 28 2021, 6:12 AM · Restricted Project

Oct 7 2021

matejam updated the summary of D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.
Oct 7 2021, 5:43 AM · Restricted Project
matejam updated the diff for D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.

Instead of a pattern, use a combiner on AMDGPU for GlobalISel.

Oct 7 2021, 5:42 AM · Restricted Project

Oct 4 2021

matejam added inline comments to D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.
Oct 4 2021, 6:20 AM · Restricted Project

Oct 1 2021

matejam updated the diff for D98048: [AMDGPU][GlobalISel] Transform (fsub (fneg (fmul, x, y)), z) -> (fma (fneg x), y, (fneg z)).

Changes in tests.

Oct 1 2021, 7:27 AM · Restricted Project
matejam updated the diff for D98050: [AMDGPU][GlobalISel] Transform (fsub (fpext (fneg (fmul x, y))), z) -> (fneg (fma (fpext x), (fpext y), z)).

Changes in tests.

Oct 1 2021, 7:24 AM · Restricted Project
matejam updated the diff for D98049: [AMDGPU][GlobalISel] Transform (fsub (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), (fneg z)).

Changes in tests.

Oct 1 2021, 7:24 AM · Restricted Project
matejam updated the diff for D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Changes in tests.

Oct 1 2021, 7:23 AM · Restricted Project
matejam updated the diff for D98047: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fpext (fmul u, v))), z) -> (fma x, y, (fma (fpext u), (fpext v), z)).

Changes in tests.

Oct 1 2021, 7:23 AM · Restricted Project
matejam updated the diff for D97938: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y, (fma u, v, z)).

Changes in tests.

Oct 1 2021, 7:23 AM · Restricted Project
matejam updated the diff for D97937: [AMDGPU][GlobalISel] Transform (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z).

Changes in tests.

Oct 1 2021, 7:23 AM · Restricted Project
matejam updated the diff for D93305: [AMDGPU][GlobalISel] Transform (fadd (fmul x, y), z) -> (fma x, y, z).

Changes in tests.

Oct 1 2021, 7:22 AM · Restricted Project

Sep 30 2021

matejam updated the diff for D98050: [AMDGPU][GlobalISel] Transform (fsub (fpext (fneg (fmul x, y))), z) -> (fneg (fma (fpext x), (fpext y), z)).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:10 AM · Restricted Project
matejam updated the diff for D98048: [AMDGPU][GlobalISel] Transform (fsub (fneg (fmul, x, y)), z) -> (fma (fneg x), y, (fneg z)).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:10 AM · Restricted Project
matejam updated the diff for D98049: [AMDGPU][GlobalISel] Transform (fsub (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), (fneg z)).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:10 AM · Restricted Project
matejam updated the diff for D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:09 AM · Restricted Project
matejam updated the diff for D98047: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fpext (fmul u, v))), z) -> (fma x, y, (fma (fpext u), (fpext v), z)).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:09 AM · Restricted Project
matejam updated the diff for D97938: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y, (fma u, v, z)).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:09 AM · Restricted Project
matejam updated the diff for D97937: [AMDGPU][GlobalISel] Transform (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:08 AM · Restricted Project
matejam updated the diff for D93305: [AMDGPU][GlobalISel] Transform (fadd (fmul x, y), z) -> (fma x, y, z).

Use applyBuildFn instead of writing my own apply.

Sep 30 2021, 7:07 AM · Restricted Project

Sep 29 2021

matejam updated the diff for D98050: [AMDGPU][GlobalISel] Transform (fsub (fpext (fneg (fmul x, y))), z) -> (fneg (fma (fpext x), (fpext y), z)).

Use m_MInstr instead of m_Reg in matching patterns (mi_match).
Formatting and refactoring.

Sep 29 2021, 7:10 AM · Restricted Project
matejam updated the diff for D98049: [AMDGPU][GlobalISel] Transform (fsub (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), (fneg z)).

Use m_MInstr instead of m_Reg in matching patterns (mi_match).
Instead of getting MachineInstr's, get only the Register values (no need for MachineInstr in this case).
A few minor bug fixes.
Formatting and refactoring.

Sep 29 2021, 7:09 AM · Restricted Project
matejam updated the diff for D98048: [AMDGPU][GlobalISel] Transform (fsub (fneg (fmul, x, y)), z) -> (fma (fneg x), y, (fneg z)).

Use m_MInstr instead of m_Reg in matching patterns (mi_match).
Instead of getting MachineInstr's, get only the Register values (no need for MachineInstr in this case).
Formatting and refactoring.

Sep 29 2021, 7:08 AM · Restricted Project
matejam updated the diff for D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Use m_MInstr instead of m_Reg in matching patterns (mi_match).
A few minor bug fixes.
Formatting and refactoring.

Sep 29 2021, 7:05 AM · Restricted Project
matejam updated the diff for D98047: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fpext (fmul u, v))), z) -> (fma x, y, (fma (fpext u), (fpext v), z)).

Use m_MInstr instead of m_Reg in matching patterns (mi_match).
A few minor bug fixes.
Formatting and refactoring.

Sep 29 2021, 7:04 AM · Restricted Project
matejam updated the diff for D97938: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y, (fma u, v, z)).

Use m_MInstr instead of m_Reg in matching patterns (mi_match).
Use CanReassociate as a parameter in canCombineFMadOrFMA with a default value.
Formatting and refactoring.

Sep 29 2021, 7:03 AM · Restricted Project
matejam updated the diff for D97937: [AMDGPU][GlobalISel] Transform (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z).

Use m_MInstr instead of m_Reg in matching patterns (mi_match).
Formatting and refactoring.

Sep 29 2021, 6:59 AM · Restricted Project
matejam updated the diff for D93305: [AMDGPU][GlobalISel] Transform (fadd (fmul x, y), z) -> (fma x, y, z).

Move isFPExtFoldable() to the patch where it is used for the first time.
Updated to reflect changes in D104247.
Update tests.

Sep 29 2021, 6:56 AM · Restricted Project

Sep 21 2021

matejam added inline comments to D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.
Sep 21 2021, 5:54 AM · Restricted Project

Sep 20 2021

matejam requested review of D110076: [AMDGPU][GlobalISel] Code quality: Combine V_RSQ.
Sep 20 2021, 8:16 AM · Restricted Project

Sep 7 2021

matejam abandoned D98784: [AMDGPU][GlobalISel] Fix unable to legalize G_FSHL error.
Sep 7 2021, 7:41 AM · Restricted Project
matejam updated the summary of D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.
Sep 7 2021, 7:23 AM · Restricted Project
matejam updated the diff for D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.

Formatting.

Sep 7 2021, 6:08 AM · Restricted Project
matejam updated the diff for D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.

Instead of clampScalar, use minScalar and maxScalar before and after widening to the next multiple of 32.

Sep 7 2021, 5:54 AM · Restricted Project

Sep 6 2021

matejam updated the diff for D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.

Changes in comments, refactoring and formatting.

Sep 6 2021, 8:29 AM · Restricted Project
matejam updated the diff for D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.

Refactoring and formatting. Using parameters instead of hardcoding.

Sep 6 2021, 7:37 AM · Restricted Project
matejam updated the diff for D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.

Refactoring and formatting.

Sep 6 2021, 6:18 AM · Restricted Project

Sep 3 2021

matejam updated the diff for D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.

Include G_ADD and G_SUB along with G_MUL. Instead of widening the scalar to the next power of 2, widen it to the next
multiple of 32 (if the type is i65, widen it to i96, instead of i128).
Refactoring and formatting.

Sep 3 2021, 10:04 AM · Restricted Project
matejam requested review of D109228: [AMDGPU][GlobalISel] Legalize G_MUL for non-standard types.
Sep 3 2021, 5:26 AM · Restricted Project

Aug 27 2021

matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Refactoring.

Aug 27 2021, 7:36 AM · Restricted Project
matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

In case the funnel shift instruction is not legal and the type is not a power of 2, use the old way of lowering (G_ROTL/G_ROTR -> G_SHL/G_LSHR).

Aug 27 2021, 6:10 AM · Restricted Project

Aug 24 2021

matejam added inline comments to D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.
Aug 24 2021, 6:50 AM · Restricted Project
matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Minor bug fix.

Aug 24 2021, 6:27 AM · Restricted Project
matejam added inline comments to D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.
Aug 24 2021, 5:01 AM · Restricted Project

Aug 12 2021

matejam updated the diff for D98050: [AMDGPU][GlobalISel] Transform (fsub (fpext (fneg (fmul x, y))), z) -> (fneg (fma (fpext x), (fpext y), z)).

Rebased.

Aug 12 2021, 5:22 AM · Restricted Project
matejam updated the diff for D98049: [AMDGPU][GlobalISel] Transform (fsub (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), (fneg z)).

Rebased.

Aug 12 2021, 5:21 AM · Restricted Project
matejam updated the diff for D98048: [AMDGPU][GlobalISel] Transform (fsub (fneg (fmul, x, y)), z) -> (fma (fneg x), y, (fneg z)).

Rebased.

Aug 12 2021, 5:20 AM · Restricted Project
matejam updated the diff for D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Rebased.

Aug 12 2021, 5:19 AM · Restricted Project
matejam updated the diff for D97937: [AMDGPU][GlobalISel] Transform (fadd (fpext (fmul x, y)), z) -> (fma (fpext x), (fpext y), z).

Rebased.

Aug 12 2021, 5:19 AM · Restricted Project
matejam updated the diff for D98047: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fpext (fmul u, v))), z) -> (fma x, y, (fma (fpext u), (fpext v), z)).

Rebased.

Aug 12 2021, 5:18 AM · Restricted Project
matejam updated the diff for D97938: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y, (fma u, v, z)).

Rebased.

Aug 12 2021, 5:18 AM · Restricted Project
matejam updated the diff for D93305: [AMDGPU][GlobalISel] Transform (fadd (fmul x, y), z) -> (fma x, y, z).

Rebased.

Aug 12 2021, 5:17 AM · Restricted Project

Aug 11 2021

matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Formatting.

Aug 11 2021, 6:08 AM · Restricted Project
matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Added a few comments and few minor updates.

Aug 11 2021, 4:52 AM · Restricted Project
matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Move the legalization of G_ROTL and G_ROTR from AMDGPULegalizerInfo to LegalizerHelper.

Aug 11 2021, 4:38 AM · Restricted Project

Jul 30 2021

matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Formatting and refactoring.

Jul 30 2021, 7:02 AM · Restricted Project
matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Formatting and refactoring.

Jul 30 2021, 4:00 AM · Restricted Project
matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Implement integer negation buildNeg() and use it instead of buildFNeg().

Jul 30 2021, 2:58 AM · Restricted Project

Jul 28 2021

matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Add tests for non power of 2 types (i15, i31...).

Jul 28 2021, 4:34 AM · Restricted Project
matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

When rotating left, use the right negated rotation (negate the number of shifts) only in cases when the destination type size is a power of 2.

Jul 28 2021, 3:44 AM · Restricted Project

Jul 27 2021

matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.
Jul 27 2021, 9:10 AM · Restricted Project

Jul 23 2021

matejam requested review of D106672: [AMDGPU][GlobalIsel] Fix unable to legalize G_CONCAT_VECTORS when loading a vector..
Jul 23 2021, 8:24 AM · Restricted Project

Jul 7 2021

matejam added inline comments to D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.
Jul 7 2021, 6:37 AM · Restricted Project

Jul 2 2021

matejam updated the diff for D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.

Changes in tests.

Jul 2 2021, 7:16 AM · Restricted Project
matejam requested review of D105347: [AMDGPU][GlobalISel] Legalization of G_ROTL and G_ROTR.
Jul 2 2021, 5:36 AM · Restricted Project

Jun 4 2021

matejam updated the diff for D97938: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fmul u, v)), z) -> (fma x, y, (fma u, v, z)).

Typos and refactoring.

Jun 4 2021, 7:44 AM · Restricted Project
matejam updated the diff for D98047: [AMDGPU][GlobalISel] Transform (fadd (fma x, y, (fpext (fmul u, v))), z) -> (fma x, y, (fma (fpext u), (fpext v), z)).

Typos and refactoring.

Jun 4 2021, 7:43 AM · Restricted Project
matejam updated the diff for D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Typos and refactoring.

Jun 4 2021, 7:10 AM · Restricted Project
matejam added a comment to D96614: [AMDGPU][GlobalISel] Transform (fsub (fmul x, y), z) -> (fma x, y, -z).

Does this assume that all targets can do the fneg for free? Or can a target choose to fold only (fadd (fmul x, y), z), not (fsub (fmul x, y), z) ?

Jun 4 2021, 7:09 AM · Restricted Project