[NVPTX] Add missing type suffixes for barrier.cta.red #172945
Merged
Conversation
This file contains hidden or 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
Member
|
@llvm/pr-subscribers-backend-nvptx Author: Alex MacLean (AlexMaclean) ChangesFull diff: https://github.com/llvm/llvm-project/pull/172945.diff 2 Files Affected:
diff --git a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
index b145e1d53f46c..0303fd15a001a 100644
--- a/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
+++ b/llvm/lib/Target/NVPTX/NVPTXIntrinsics.td
@@ -151,19 +151,19 @@ defm BARRIER_CTA_SYNC_ALL : BARRIER_ALL<"barrier.sync", int_nvvm_barrier_cta_syn
defm BARRIER_CTA_SYNC : BARRIER_COUNT<"barrier.sync", int_nvvm_barrier_cta_sync_count, [hasPTX<60>]>;
defm BARRIER_CTA_ARRIVE : BARRIER_COUNT<"barrier.arrive", int_nvvm_barrier_cta_arrive_count, [hasPTX<60>]>;
-defm BARRIER_CTA_RED_POPC_ALIGNED_ALL : BARRIER_RED_ALL<"bar.red.popc", int_nvvm_barrier_cta_red_popc_aligned_all, I32RT>;
-defm BARRIER_CTA_RED_AND_ALIGNED_ALL : BARRIER_RED_ALL<"bar.red.and", int_nvvm_barrier_cta_red_and_aligned_all, I1RT>;
-defm BARRIER_CTA_RED_OR_ALIGNED_ALL : BARRIER_RED_ALL<"bar.red.or", int_nvvm_barrier_cta_red_or_aligned_all, I1RT>;
-defm BARRIER_CTA_RED_POPC_ALIGNED : BARRIER_RED_COUNT<"bar.red.popc", int_nvvm_barrier_cta_red_popc_aligned_count, I32RT>;
-defm BARRIER_CTA_RED_AND_ALIGNED : BARRIER_RED_COUNT<"bar.red.and", int_nvvm_barrier_cta_red_and_aligned_count, I1RT>;
-defm BARRIER_CTA_RED_OR_ALIGNED : BARRIER_RED_COUNT<"bar.red.or", int_nvvm_barrier_cta_red_or_aligned_count, I1RT>;
-
-defm BARRIER_CTA_RED_POPC_ALL : BARRIER_RED_ALL<"barrier.red.popc", int_nvvm_barrier_cta_red_popc_all, I32RT, [hasPTX<60>]>;
-defm BARRIER_CTA_RED_AND_ALL : BARRIER_RED_ALL<"barrier.red.and", int_nvvm_barrier_cta_red_and_all, I1RT, [hasPTX<60>]>;
-defm BARRIER_CTA_RED_OR_ALL : BARRIER_RED_ALL<"barrier.red.or", int_nvvm_barrier_cta_red_or_all, I1RT, [hasPTX<60>]>;
-defm BARRIER_CTA_RED_POPC_COUNT : BARRIER_RED_COUNT<"barrier.red.popc", int_nvvm_barrier_cta_red_popc_count, I32RT, [hasPTX<60>]>;
-defm BARRIER_CTA_RED_AND_COUNT : BARRIER_RED_COUNT<"barrier.red.and", int_nvvm_barrier_cta_red_and_count, I1RT, [hasPTX<60>]>;
-defm BARRIER_CTA_RED_OR_COUNT : BARRIER_RED_COUNT<"barrier.red.or", int_nvvm_barrier_cta_red_or_count, I1RT, [hasPTX<60>]>;
+defm BARRIER_CTA_RED_POPC_ALIGNED_ALL : BARRIER_RED_ALL<"bar.red.popc.u32", int_nvvm_barrier_cta_red_popc_aligned_all, I32RT>;
+defm BARRIER_CTA_RED_AND_ALIGNED_ALL : BARRIER_RED_ALL<"bar.red.and.pred", int_nvvm_barrier_cta_red_and_aligned_all, I1RT>;
+defm BARRIER_CTA_RED_OR_ALIGNED_ALL : BARRIER_RED_ALL<"bar.red.or.pred", int_nvvm_barrier_cta_red_or_aligned_all, I1RT>;
+defm BARRIER_CTA_RED_POPC_ALIGNED : BARRIER_RED_COUNT<"bar.red.popc.u32", int_nvvm_barrier_cta_red_popc_aligned_count, I32RT>;
+defm BARRIER_CTA_RED_AND_ALIGNED : BARRIER_RED_COUNT<"bar.red.and.pred", int_nvvm_barrier_cta_red_and_aligned_count, I1RT>;
+defm BARRIER_CTA_RED_OR_ALIGNED : BARRIER_RED_COUNT<"bar.red.or.pred", int_nvvm_barrier_cta_red_or_aligned_count, I1RT>;
+
+defm BARRIER_CTA_RED_POPC_ALL : BARRIER_RED_ALL<"barrier.red.popc.u32", int_nvvm_barrier_cta_red_popc_all, I32RT, [hasPTX<60>]>;
+defm BARRIER_CTA_RED_AND_ALL : BARRIER_RED_ALL<"barrier.red.and.pred", int_nvvm_barrier_cta_red_and_all, I1RT, [hasPTX<60>]>;
+defm BARRIER_CTA_RED_OR_ALL : BARRIER_RED_ALL<"barrier.red.or.pred", int_nvvm_barrier_cta_red_or_all, I1RT, [hasPTX<60>]>;
+defm BARRIER_CTA_RED_POPC_COUNT : BARRIER_RED_COUNT<"barrier.red.popc.u32", int_nvvm_barrier_cta_red_popc_count, I32RT, [hasPTX<60>]>;
+defm BARRIER_CTA_RED_AND_COUNT : BARRIER_RED_COUNT<"barrier.red.and.pred", int_nvvm_barrier_cta_red_and_count, I1RT, [hasPTX<60>]>;
+defm BARRIER_CTA_RED_OR_COUNT : BARRIER_RED_COUNT<"barrier.red.or.pred", int_nvvm_barrier_cta_red_or_count, I1RT, [hasPTX<60>]>;
class INT_BARRIER_CLUSTER<string variant, Intrinsic Intr,
list<Predicate> Preds = [hasPTX<78>, hasSM<90>]>:
diff --git a/llvm/test/CodeGen/NVPTX/barrier.ll b/llvm/test/CodeGen/NVPTX/barrier.ll
index c785f09fcf87e..957a3ab59b591 100644
--- a/llvm/test/CodeGen/NVPTX/barrier.ll
+++ b/llvm/test/CodeGen/NVPTX/barrier.ll
@@ -147,10 +147,10 @@ define void @barrier_cta_red_popc_all(i32 %id, i1 %pred) {
; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_popc_all_param_0];
-; CHECK-NEXT: bar.red.popc %r2, %r1, %p1;
-; CHECK-NEXT: bar.red.popc %r3, 3, %p1;
-; CHECK-NEXT: barrier.red.popc %r4, %r1, %p1;
-; CHECK-NEXT: barrier.red.popc %r5, 3, %p1;
+; CHECK-NEXT: bar.red.popc.u32 %r2, %r1, %p1;
+; CHECK-NEXT: bar.red.popc.u32 %r3, 3, %p1;
+; CHECK-NEXT: barrier.red.popc.u32 %r4, %r1, %p1;
+; CHECK-NEXT: barrier.red.popc.u32 %r5, 3, %p1;
; CHECK-NEXT: ret;
%v1 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.all(i32 %id, i1 %pred)
%v2 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.all(i32 3, i1 %pred)
@@ -172,24 +172,24 @@ define void @barrier_cta_red_popc_count(i32 %id, i32 %cnt, i1 %pred) {
; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_popc_count_param_0];
; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_red_popc_count_param_1];
-; CHECK-NEXT: bar.red.popc %r3, %r1, %r2, %p1;
-; CHECK-NEXT: bar.red.popc %r4, 3, %r2, %p1;
-; CHECK-NEXT: barrier.red.popc %r5, %r1, %r2, %p1;
-; CHECK-NEXT: barrier.red.popc %r6, 3, %r2, %p1;
-; CHECK-NEXT: bar.red.popc %r7, %r1, 10, %p1;
-; CHECK-NEXT: bar.red.popc %r8, 3, 11, %p1;
-; CHECK-NEXT: barrier.red.popc %r9, %r1, 12, %p1;
-; CHECK-NEXT: barrier.red.popc %r10, 3, 13, %p1;
+; CHECK-NEXT: bar.red.popc.u32 %r3, %r1, %r2, %p1;
+; CHECK-NEXT: bar.red.popc.u32 %r4, 3, %r2, %p1;
+; CHECK-NEXT: barrier.red.popc.u32 %r5, %r1, %r2, %p1;
+; CHECK-NEXT: barrier.red.popc.u32 %r6, 3, %r2, %p1;
+; CHECK-NEXT: bar.red.popc.u32 %r7, %r1, 64, %p1;
+; CHECK-NEXT: bar.red.popc.u32 %r8, 3, 64, %p1;
+; CHECK-NEXT: barrier.red.popc.u32 %r9, %r1, 64, %p1;
+; CHECK-NEXT: barrier.red.popc.u32 %r10, 3, 64, %p1;
; CHECK-NEXT: ret;
%v1 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 %id, i32 %cnt, i1 %pred)
%v2 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 3, i32 %cnt, i1 %pred)
%v3 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 %id, i32 %cnt, i1 %pred)
%v4 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 3, i32 %cnt, i1 %pred)
- %v5 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 %id, i32 10, i1 %pred)
- %v6 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 3, i32 11, i1 %pred)
- %v7 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 %id, i32 12, i1 %pred)
- %v8 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 3, i32 13, i1 %pred)
+ %v5 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 %id, i32 64, i1 %pred)
+ %v6 = call i32 @llvm.nvvm.barrier.cta.red.popc.aligned.count(i32 3, i32 64, i1 %pred)
+ %v7 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 %id, i32 64, i1 %pred)
+ %v8 = call i32 @llvm.nvvm.barrier.cta.red.popc.count(i32 3, i32 64, i1 %pred)
ret void
}
@@ -205,10 +205,10 @@ define void @barrier_cta_red_and_all(i32 %id, i1 %pred) {
; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_and_all_param_0];
-; CHECK-NEXT: bar.red.and %p2, %r1, %p1;
-; CHECK-NEXT: bar.red.and %p3, 3, %p1;
-; CHECK-NEXT: barrier.red.and %p4, %r1, %p1;
-; CHECK-NEXT: barrier.red.and %p5, 3, %p1;
+; CHECK-NEXT: bar.red.and.pred %p2, %r1, %p1;
+; CHECK-NEXT: bar.red.and.pred %p3, 3, %p1;
+; CHECK-NEXT: barrier.red.and.pred %p4, %r1, %p1;
+; CHECK-NEXT: barrier.red.and.pred %p5, 3, %p1;
; CHECK-NEXT: ret;
%v1 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.all(i32 %id, i1 %pred)
%v2 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.all(i32 3, i1 %pred)
@@ -230,24 +230,24 @@ define void @barrier_cta_red_and_count(i32 %id, i32 %cnt, i1 %pred) {
; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_and_count_param_0];
; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_red_and_count_param_1];
-; CHECK-NEXT: bar.red.and %p2, %r1, %r2, %p1;
-; CHECK-NEXT: bar.red.and %p3, 3, %r2, %p1;
-; CHECK-NEXT: barrier.red.and %p4, %r1, %r2, %p1;
-; CHECK-NEXT: barrier.red.and %p5, 3, %r2, %p1;
-; CHECK-NEXT: bar.red.and %p6, %r1, 10, %p1;
-; CHECK-NEXT: bar.red.and %p7, 3, 11, %p1;
-; CHECK-NEXT: barrier.red.and %p8, %r1, 12, %p1;
-; CHECK-NEXT: barrier.red.and %p9, 3, 13, %p1;
+; CHECK-NEXT: bar.red.and.pred %p2, %r1, %r2, %p1;
+; CHECK-NEXT: bar.red.and.pred %p3, 3, %r2, %p1;
+; CHECK-NEXT: barrier.red.and.pred %p4, %r1, %r2, %p1;
+; CHECK-NEXT: barrier.red.and.pred %p5, 3, %r2, %p1;
+; CHECK-NEXT: bar.red.and.pred %p6, %r1, 64, %p1;
+; CHECK-NEXT: bar.red.and.pred %p7, 3, 32, %p1;
+; CHECK-NEXT: barrier.red.and.pred %p8, %r1, 64, %p1;
+; CHECK-NEXT: barrier.red.and.pred %p9, 3, 64, %p1;
; CHECK-NEXT: ret;
%v1 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 %id, i32 %cnt, i1 %pred)
%v2 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 3, i32 %cnt, i1 %pred)
%v3 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 %id, i32 %cnt, i1 %pred)
%v4 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 3, i32 %cnt, i1 %pred)
- %v5 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 %id, i32 10, i1 %pred)
- %v6 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 3, i32 11, i1 %pred)
- %v7 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 %id, i32 12, i1 %pred)
- %v8 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 3, i32 13, i1 %pred)
+ %v5 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 %id, i32 64, i1 %pred)
+ %v6 = call i1 @llvm.nvvm.barrier.cta.red.and.aligned.count(i32 3, i32 32, i1 %pred)
+ %v7 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 %id, i32 64, i1 %pred)
+ %v8 = call i1 @llvm.nvvm.barrier.cta.red.and.count(i32 3, i32 64, i1 %pred)
ret void
}
@@ -263,10 +263,10 @@ define void @barrier_cta_red_or_all(i32 %id, i1 %pred) {
; CHECK-NEXT: and.b16 %rs2, %rs1, 1;
; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_or_all_param_0];
-; CHECK-NEXT: bar.red.or %p2, %r1, %p1;
-; CHECK-NEXT: bar.red.or %p3, 3, %p1;
-; CHECK-NEXT: barrier.red.or %p4, %r1, %p1;
-; CHECK-NEXT: barrier.red.or %p5, 3, %p1;
+; CHECK-NEXT: bar.red.or.pred %p2, %r1, %p1;
+; CHECK-NEXT: bar.red.or.pred %p3, 3, %p1;
+; CHECK-NEXT: barrier.red.or.pred %p4, %r1, %p1;
+; CHECK-NEXT: barrier.red.or.pred %p5, 3, %p1;
; CHECK-NEXT: ret;
%v1 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.all(i32 %id, i1 %pred)
%v2 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.all(i32 3, i1 %pred)
@@ -288,23 +288,23 @@ define void @barrier_cta_red_or_count(i32 %id, i32 %cnt, i1 %pred) {
; CHECK-NEXT: setp.ne.b16 %p1, %rs2, 0;
; CHECK-NEXT: ld.param.b32 %r1, [barrier_cta_red_or_count_param_0];
; CHECK-NEXT: ld.param.b32 %r2, [barrier_cta_red_or_count_param_1];
-; CHECK-NEXT: bar.red.or %p2, %r1, %r2, %p1;
-; CHECK-NEXT: bar.red.or %p3, 3, %r2, %p1;
-; CHECK-NEXT: barrier.red.or %p4, %r1, %r2, %p1;
-; CHECK-NEXT: barrier.red.or %p5, 3, %r2, %p1;
-; CHECK-NEXT: bar.red.or %p6, %r1, 10, %p1;
-; CHECK-NEXT: bar.red.or %p7, 3, 11, %p1;
-; CHECK-NEXT: barrier.red.or %p8, %r1, 12, %p1;
-; CHECK-NEXT: barrier.red.or %p9, 3, 13, %p1;
+; CHECK-NEXT: bar.red.or.pred %p2, %r1, %r2, %p1;
+; CHECK-NEXT: bar.red.or.pred %p3, 3, %r2, %p1;
+; CHECK-NEXT: barrier.red.or.pred %p4, %r1, %r2, %p1;
+; CHECK-NEXT: barrier.red.or.pred %p5, 3, %r2, %p1;
+; CHECK-NEXT: bar.red.or.pred %p6, %r1, 64, %p1;
+; CHECK-NEXT: bar.red.or.pred %p7, 3, 32, %p1;
+; CHECK-NEXT: barrier.red.or.pred %p8, %r1, 64, %p1;
+; CHECK-NEXT: barrier.red.or.pred %p9, 3, 64, %p1;
; CHECK-NEXT: ret;
%v1 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 %id, i32 %cnt, i1 %pred)
%v2 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 3, i32 %cnt, i1 %pred)
%v3 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 %id, i32 %cnt, i1 %pred)
%v4 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 3, i32 %cnt, i1 %pred)
- %v5 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 %id, i32 10, i1 %pred)
- %v6 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 3, i32 11, i1 %pred)
- %v7 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 %id, i32 12, i1 %pred)
- %v8 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 3, i32 13, i1 %pred)
+ %v5 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 %id, i32 64, i1 %pred)
+ %v6 = call i1 @llvm.nvvm.barrier.cta.red.or.aligned.count(i32 3, i32 32, i1 %pred)
+ %v7 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 %id, i32 64, i1 %pred)
+ %v8 = call i1 @llvm.nvvm.barrier.cta.red.or.count(i32 3, i32 64, i1 %pred)
ret void
}
|
AlexMaclean
requested review from
Artem-B,
Wolfram70,
durga4github and
schwarzschild-radius
schwarzschild-radius
approved these changes
Dec 19, 2025
Contributor
schwarzschild-radius
left a comment
schwarzschild-radius
left a comment
There was a problem hiding this comment.
LGTM, Thanks!
durga4github
approved these changes
Dec 19, 2025
Collaborator
|
LLVM Buildbot has detected a new failure on builder Full details are available at: https://lab.llvm.org/buildbot/#/builders/51/builds/28987 Here is the relevant piece of the build log for the reference |
mahesh-attarde
pushed a commit
to mahesh-attarde/llvm-project
that referenced
this pull request
Dec 19, 2025
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
6 participants
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Fixes an oversight in the PTX instruction names used in #172541.