Skip to content
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

Jump to bottom

[AArch64][GlobalISel] Improve codegen for G_VECREDUCE_{SMIN,SMAX,UMIN,UMAX} for odd-sized vectors #81831

Closed
wants to merge 1 commit into from
Closed

[AArch64][GlobalISel] Improve codegen for G_VECREDUCE_{SMIN,SMAX,UMIN,UMAX} for odd-sized vectors #81831

wants to merge 1 commit into from

Conversation

dc03-work
Copy link
Contributor

i8 vectors do not have their sizes changed as I noticed regressions in some
tests when that was done.

@dc03-work
[𝘀𝗽𝗿] initial version
02b11c8 
Created using spr 1.3.5
@github-actions GitHub Actions
Copy link

Thank you for submitting a Pull Request (PR) to the LLVM Project!

This PR will be automatically labeled and the relevant teams will be
notified.

If you wish to, you can add reviewers by using the "Reviewers" section on this page.

If this is not working for you, it is probably because you do not have write
permissions for the repository. In which case you can instead tag reviewers by
name in a comment by using @ followed by their GitHub username.

If you have received no comments on your PR for a week, you can request a review
by "ping"ing the PR by adding a comment “Ping”. The common courtesy "ping" rate
is once a week. Please remember that you are asking for valuable time from other developers.

If you have further questions, they may be answered by the LLVM GitHub User Guide.

You can also ask questions in a comment on this PR, on the LLVM Discord or on the forums.

@dc03-work
Copy link
Contributor Author

This PR is stacked on top of #81830. Sorry for the long branch names on both, I do not know how to change the default branch names with SPR

@llvmbot
Copy link
Collaborator

llvmbot commented Feb 15, 2024

@llvm/pr-subscribers-backend-aarch64

Author: Dhruv Chawla (work) (dc03-work)

Changes

i8 vectors do not have their sizes changed as I noticed regressions in some
tests when that was done.

Full diff: https://github.com/llvm/llvm-project/pull/81831.diff

3 Files Affected:

  • (modified) llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp (+7)
  • (modified) llvm/test/CodeGen/AArch64/aarch64-minmaxv.ll (+47-144)
  • (modified) llvm/test/CodeGen/AArch64/vecreduce-umax-legalization.ll (+6-21)
diff --git a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
index 933f13dd5a19af..622a2b9cceb4bb 100644
--- a/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
+++ b/llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -1070,6 +1070,13 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
                  {s16, v8s16},
                  {s32, v2s32},
                  {s32, v4s32}})
+      .moreElementsIf(
+          [=](const LegalityQuery &Query) {
+            return Query.Types[1].isVector() &&
+                   Query.Types[1].getElementType() != s8 &&
+                   Query.Types[1].getNumElements() & 1;
+          },
+          LegalizeMutations::moreElementsToNextPow2(1))
       .clampMaxNumElements(1, s64, 2)
       .clampMaxNumElements(1, s32, 4)
       .clampMaxNumElements(1, s16, 8)
diff --git a/llvm/test/CodeGen/AArch64/aarch64-minmaxv.ll b/llvm/test/CodeGen/AArch64/aarch64-minmaxv.ll
index 194fe5be40c2bd..76790d128d066c 100644
--- a/llvm/test/CodeGen/AArch64/aarch64-minmaxv.ll
+++ b/llvm/test/CodeGen/AArch64/aarch64-minmaxv.ll
@@ -595,30 +595,14 @@ entry:
 }
 
 define i16 @sminv_v3i16(<3 x i16> %a) {
-; CHECK-SD-LABEL: sminv_v3i16:
-; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-SD-NEXT:    mov w8, #32767 // =0x7fff
-; CHECK-SD-NEXT:    mov v0.h[3], w8
-; CHECK-SD-NEXT:    sminv h0, v0.4h
-; CHECK-SD-NEXT:    fmov w0, s0
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: sminv_v3i16:
-; CHECK-GI:       // %bb.0: // %entry
-; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-GI-NEXT:    mov h1, v0.h[1]
-; CHECK-GI-NEXT:    smov w8, v0.h[0]
-; CHECK-GI-NEXT:    umov w9, v0.h[0]
-; CHECK-GI-NEXT:    umov w10, v0.h[1]
-; CHECK-GI-NEXT:    smov w11, v0.h[2]
-; CHECK-GI-NEXT:    umov w13, v0.h[2]
-; CHECK-GI-NEXT:    fmov w12, s1
-; CHECK-GI-NEXT:    cmp w8, w12, sxth
-; CHECK-GI-NEXT:    csel w8, w9, w10, lt
-; CHECK-GI-NEXT:    cmp w11, w8, sxth
-; CHECK-GI-NEXT:    csel w0, w8, w13, gt
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: sminv_v3i16:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    // kill: def $d0 killed $d0 def $q0
+; CHECK-NEXT:    mov w8, #32767 // =0x7fff
+; CHECK-NEXT:    mov v0.h[3], w8
+; CHECK-NEXT:    sminv h0, v0.4h
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    ret
 entry:
   %arg1 = call i16 @llvm.vector.reduce.smin.v3i16(<3 x i16> %a)
   ret i16 %arg1
@@ -670,28 +654,13 @@ entry:
 }
 
 define i32 @sminv_v3i32(<3 x i32> %a) {
-; CHECK-SD-LABEL: sminv_v3i32:
-; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    mov w8, #2147483647 // =0x7fffffff
-; CHECK-SD-NEXT:    mov v0.s[3], w8
-; CHECK-SD-NEXT:    sminv s0, v0.4s
-; CHECK-SD-NEXT:    fmov w0, s0
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: sminv_v3i32:
-; CHECK-GI:       // %bb.0: // %entry
-; CHECK-GI-NEXT:    mov s1, v0.s[1]
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    mov s2, v0.s[2]
-; CHECK-GI-NEXT:    fmov w9, s1
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fmov w9, s2
-; CHECK-GI-NEXT:    fcsel s0, s0, s1, lt
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fcsel s0, s0, s2, lt
-; CHECK-GI-NEXT:    fmov w0, s0
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: sminv_v3i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    mov w8, #2147483647 // =0x7fffffff
+; CHECK-NEXT:    mov v0.s[3], w8
+; CHECK-NEXT:    sminv s0, v0.4s
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    ret
 entry:
   %arg1 = call i32 @llvm.vector.reduce.smin.v3i32(<3 x i32> %a)
   ret i32 %arg1
@@ -972,17 +941,10 @@ define i16 @smaxv_v3i16(<3 x i16> %a) {
 ; CHECK-GI-LABEL: smaxv_v3i16:
 ; CHECK-GI:       // %bb.0: // %entry
 ; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-GI-NEXT:    mov h1, v0.h[1]
-; CHECK-GI-NEXT:    smov w8, v0.h[0]
-; CHECK-GI-NEXT:    umov w9, v0.h[0]
-; CHECK-GI-NEXT:    umov w10, v0.h[1]
-; CHECK-GI-NEXT:    smov w11, v0.h[2]
-; CHECK-GI-NEXT:    umov w13, v0.h[2]
-; CHECK-GI-NEXT:    fmov w12, s1
-; CHECK-GI-NEXT:    cmp w8, w12, sxth
-; CHECK-GI-NEXT:    csel w8, w9, w10, gt
-; CHECK-GI-NEXT:    cmp w11, w8, sxth
-; CHECK-GI-NEXT:    csel w0, w8, w13, lt
+; CHECK-GI-NEXT:    mov w8, #32768 // =0x8000
+; CHECK-GI-NEXT:    mov v0.h[3], w8
+; CHECK-GI-NEXT:    smaxv h0, v0.4h
+; CHECK-GI-NEXT:    fmov w0, s0
 ; CHECK-GI-NEXT:    ret
 entry:
   %arg1 = call i16 @llvm.vector.reduce.smax.v3i16(<3 x i16> %a)
@@ -1035,28 +997,13 @@ entry:
 }
 
 define i32 @smaxv_v3i32(<3 x i32> %a) {
-; CHECK-SD-LABEL: smaxv_v3i32:
-; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    mov w8, #-2147483648 // =0x80000000
-; CHECK-SD-NEXT:    mov v0.s[3], w8
-; CHECK-SD-NEXT:    smaxv s0, v0.4s
-; CHECK-SD-NEXT:    fmov w0, s0
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: smaxv_v3i32:
-; CHECK-GI:       // %bb.0: // %entry
-; CHECK-GI-NEXT:    mov s1, v0.s[1]
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    mov s2, v0.s[2]
-; CHECK-GI-NEXT:    fmov w9, s1
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fmov w9, s2
-; CHECK-GI-NEXT:    fcsel s0, s0, s1, gt
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fcsel s0, s0, s2, gt
-; CHECK-GI-NEXT:    fmov w0, s0
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: smaxv_v3i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    mov w8, #-2147483648 // =0x80000000
+; CHECK-NEXT:    mov v0.s[3], w8
+; CHECK-NEXT:    smaxv s0, v0.4s
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    ret
 entry:
   %arg1 = call i32 @llvm.vector.reduce.smax.v3i32(<3 x i32> %a)
   ret i32 %arg1
@@ -1335,17 +1282,10 @@ define i16 @uminv_v3i16(<3 x i16> %a) {
 ; CHECK-GI-LABEL: uminv_v3i16:
 ; CHECK-GI:       // %bb.0: // %entry
 ; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-GI-NEXT:    mov h1, v0.h[1]
-; CHECK-GI-NEXT:    umov w8, v0.h[0]
-; CHECK-GI-NEXT:    umov w9, v0.h[0]
-; CHECK-GI-NEXT:    umov w10, v0.h[1]
-; CHECK-GI-NEXT:    umov w11, v0.h[2]
-; CHECK-GI-NEXT:    umov w13, v0.h[2]
-; CHECK-GI-NEXT:    fmov w12, s1
-; CHECK-GI-NEXT:    cmp w8, w12, uxth
-; CHECK-GI-NEXT:    csel w8, w9, w10, lo
-; CHECK-GI-NEXT:    cmp w11, w8, uxth
-; CHECK-GI-NEXT:    csel w0, w8, w13, hi
+; CHECK-GI-NEXT:    mov w8, #65535 // =0xffff
+; CHECK-GI-NEXT:    mov v0.h[3], w8
+; CHECK-GI-NEXT:    uminv h0, v0.4h
+; CHECK-GI-NEXT:    fmov w0, s0
 ; CHECK-GI-NEXT:    ret
 entry:
   %arg1 = call i16 @llvm.vector.reduce.umin.v3i16(<3 x i16> %a)
@@ -1398,28 +1338,13 @@ entry:
 }
 
 define i32 @uminv_v3i32(<3 x i32> %a) {
-; CHECK-SD-LABEL: uminv_v3i32:
-; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    mov w8, #-1 // =0xffffffff
-; CHECK-SD-NEXT:    mov v0.s[3], w8
-; CHECK-SD-NEXT:    uminv s0, v0.4s
-; CHECK-SD-NEXT:    fmov w0, s0
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: uminv_v3i32:
-; CHECK-GI:       // %bb.0: // %entry
-; CHECK-GI-NEXT:    mov s1, v0.s[1]
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    mov s2, v0.s[2]
-; CHECK-GI-NEXT:    fmov w9, s1
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fmov w9, s2
-; CHECK-GI-NEXT:    fcsel s0, s0, s1, lo
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fcsel s0, s0, s2, lo
-; CHECK-GI-NEXT:    fmov w0, s0
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: uminv_v3i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    mov w8, #-1 // =0xffffffff
+; CHECK-NEXT:    mov v0.s[3], w8
+; CHECK-NEXT:    uminv s0, v0.4s
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    ret
 entry:
   %arg1 = call i32 @llvm.vector.reduce.umin.v3i32(<3 x i32> %a)
   ret i32 %arg1
@@ -1697,17 +1622,10 @@ define i16 @umaxv_v3i16(<3 x i16> %a) {
 ; CHECK-GI-LABEL: umaxv_v3i16:
 ; CHECK-GI:       // %bb.0: // %entry
 ; CHECK-GI-NEXT:    // kill: def $d0 killed $d0 def $q0
-; CHECK-GI-NEXT:    mov h1, v0.h[1]
-; CHECK-GI-NEXT:    umov w8, v0.h[0]
-; CHECK-GI-NEXT:    umov w9, v0.h[0]
-; CHECK-GI-NEXT:    umov w10, v0.h[1]
-; CHECK-GI-NEXT:    umov w11, v0.h[2]
-; CHECK-GI-NEXT:    umov w13, v0.h[2]
-; CHECK-GI-NEXT:    fmov w12, s1
-; CHECK-GI-NEXT:    cmp w8, w12, uxth
-; CHECK-GI-NEXT:    csel w8, w9, w10, hi
-; CHECK-GI-NEXT:    cmp w11, w8, uxth
-; CHECK-GI-NEXT:    csel w0, w8, w13, lo
+; CHECK-GI-NEXT:    mov w8, #0 // =0x0
+; CHECK-GI-NEXT:    mov v0.h[3], w8
+; CHECK-GI-NEXT:    umaxv h0, v0.4h
+; CHECK-GI-NEXT:    fmov w0, s0
 ; CHECK-GI-NEXT:    ret
 entry:
   %arg1 = call i16 @llvm.vector.reduce.umax.v3i16(<3 x i16> %a)
@@ -1760,27 +1678,12 @@ entry:
 }
 
 define i32 @umaxv_v3i32(<3 x i32> %a) {
-; CHECK-SD-LABEL: umaxv_v3i32:
-; CHECK-SD:       // %bb.0: // %entry
-; CHECK-SD-NEXT:    mov v0.s[3], wzr
-; CHECK-SD-NEXT:    umaxv s0, v0.4s
-; CHECK-SD-NEXT:    fmov w0, s0
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: umaxv_v3i32:
-; CHECK-GI:       // %bb.0: // %entry
-; CHECK-GI-NEXT:    mov s1, v0.s[1]
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    mov s2, v0.s[2]
-; CHECK-GI-NEXT:    fmov w9, s1
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fmov w9, s2
-; CHECK-GI-NEXT:    fcsel s0, s0, s1, hi
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fcsel s0, s0, s2, hi
-; CHECK-GI-NEXT:    fmov w0, s0
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: umaxv_v3i32:
+; CHECK:       // %bb.0: // %entry
+; CHECK-NEXT:    mov v0.s[3], wzr
+; CHECK-NEXT:    umaxv s0, v0.4s
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    ret
 entry:
   %arg1 = call i32 @llvm.vector.reduce.umax.v3i32(<3 x i32> %a)
   ret i32 %arg1
diff --git a/llvm/test/CodeGen/AArch64/vecreduce-umax-legalization.ll b/llvm/test/CodeGen/AArch64/vecreduce-umax-legalization.ll
index 7f804fe48fd854..6d848e7b5c7c17 100644
--- a/llvm/test/CodeGen/AArch64/vecreduce-umax-legalization.ll
+++ b/llvm/test/CodeGen/AArch64/vecreduce-umax-legalization.ll
@@ -200,27 +200,12 @@ define i8 @test_v9i8(<9 x i8> %a) nounwind {
 }
 
 define i32 @test_v3i32(<3 x i32> %a) nounwind {
-; CHECK-SD-LABEL: test_v3i32:
-; CHECK-SD:       // %bb.0:
-; CHECK-SD-NEXT:    mov v0.s[3], wzr
-; CHECK-SD-NEXT:    umaxv s0, v0.4s
-; CHECK-SD-NEXT:    fmov w0, s0
-; CHECK-SD-NEXT:    ret
-;
-; CHECK-GI-LABEL: test_v3i32:
-; CHECK-GI:       // %bb.0:
-; CHECK-GI-NEXT:    mov s1, v0.s[1]
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    mov s2, v0.s[2]
-; CHECK-GI-NEXT:    fmov w9, s1
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fmov w9, s2
-; CHECK-GI-NEXT:    fcsel s0, s0, s1, hi
-; CHECK-GI-NEXT:    fmov w8, s0
-; CHECK-GI-NEXT:    cmp w8, w9
-; CHECK-GI-NEXT:    fcsel s0, s0, s2, hi
-; CHECK-GI-NEXT:    fmov w0, s0
-; CHECK-GI-NEXT:    ret
+; CHECK-LABEL: test_v3i32:
+; CHECK:       // %bb.0:
+; CHECK-NEXT:    mov v0.s[3], wzr
+; CHECK-NEXT:    umaxv s0, v0.4s
+; CHECK-NEXT:    fmov w0, s0
+; CHECK-NEXT:    ret
   %b = call i32 @llvm.vector.reduce.umax.v3i32(<3 x i32> %a)
   ret i32 %b
 }

@davemgreen davemgreen mentioned this pull request Feb 15, 2024
Closed
davemgreen
davemgreen reviewed Feb 15, 2024
View reviewed changes
llvm/lib/Target/AArch64/GISel/AArch64LegalizerInfo.cpp
@@ -1070,6 +1070,13 @@ AArch64LegalizerInfo::AArch64LegalizerInfo(const AArch64Subtarget &ST)
{s16, v8s16},
{s32, v2s32},
{s32, v4s32}})
.moreElementsIf(
Copy link
Collaborator

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I think this can happen for more than just odd numbers, if we have support in the legalizer. I think I would make it moreElementsToNextPow2 unless there is a big reason not to.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

As I noted in my commit message, unfortunately that causes regressions for odd-sized i8 vectors: https://gist.github.com/dc03-work/3d749a7be0dc893d86d2df0fbc31709a (except for the very last case... for some reason). I was noticing another test failure when I enabled it for even-sized vectors, however that seems to have gone away now.

@dc03-work
Copy link
Contributor Author

Merged with #81830 in #82740

@dc03-work dc03-work closed this Feb 23, 2024
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Reviewers

@davemgreen davemgreen davemgreen left review comments

@madhur13490 madhur13490 Awaiting requested review from madhur13490

Assignees
No one assigned
Labels
backend:AArch64
Projects
None yet
Milestone
No milestone
Development

Successfully merging this pull request may close these issues.

None yet
3 participants
@dc03-work @llvmbot @davemgreen