[AArch64] Add ISel support for partial reductions to use SVE2.1 udot/sdot #158310
Conversation
|
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 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. |
|
@llvm/pr-subscribers-backend-aarch64 Author: Damian Heaton (dheaton-arm) ChangesThis allows dot products with scalable 8xi16 vectors (and fixed-length vectors which are converted into a scalable vector) accumulating into a 4xi32 vector to lower into a single instruction ( Full diff: https://github.com/llvm/llvm-project/pull/158310.diff 2 Files Affected:
diff --git a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
index 05a3eab638eaa..88dfe4342ccae 100644
--- a/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
+++ b/llvm/lib/Target/AArch64/AArch64SVEInstrInfo.td
@@ -4237,6 +4237,13 @@ defm UDOT_ZZZ_HtoS : sve2p1_two_way_dot_vv<"udot", 0b1, int_aarch64_sve_udot_x2
defm SDOT_ZZZI_HtoS : sve2p1_two_way_dot_vvi<"sdot", 0b0, int_aarch64_sve_sdot_lane_x2>;
defm UDOT_ZZZI_HtoS : sve2p1_two_way_dot_vvi<"udot", 0b1, int_aarch64_sve_udot_lane_x2>;
+let Predicates = [HasSVE2p1_or_SME2] in {
+ def : Pat<(nxv4i32 (partial_reduce_umla nxv4i32:$Acc, nxv8i16:$MulLHS, nxv8i16:$MulRHS)),
+ (UDOT_ZZZ_HtoS $Acc, $MulLHS, $MulRHS)>;
+ def : Pat<(nxv4i32 (partial_reduce_smla nxv4i32:$Acc, nxv8i16:$MulLHS, nxv8i16:$MulRHS)),
+ (SDOT_ZZZ_HtoS $Acc, $MulLHS, $MulRHS)>;
+} // End HasSVE_or_SME
+
defm SQCVTN_Z2Z_StoH : sve2p1_multi_vec_extract_narrow<"sqcvtn", 0b00, int_aarch64_sve_sqcvtn_x2>;
defm UQCVTN_Z2Z_StoH : sve2p1_multi_vec_extract_narrow<"uqcvtn", 0b01, int_aarch64_sve_uqcvtn_x2>;
defm SQCVTUN_Z2Z_StoH : sve2p1_multi_vec_extract_narrow<"sqcvtun", 0b10, int_aarch64_sve_sqcvtun_x2>;
diff --git a/llvm/test/CodeGen/AArch64/sve2p1-dots-partial-reduction.ll b/llvm/test/CodeGen/AArch64/sve2p1-dots-partial-reduction.ll
new file mode 100644
index 0000000000000..8abd3a86ff1f7
--- /dev/null
+++ b/llvm/test/CodeGen/AArch64/sve2p1-dots-partial-reduction.ll
@@ -0,0 +1,62 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=aarch64-linux-gnu -mattr=+sve2p1 < %s | FileCheck %s
+
+define <vscale x 4 x i32> @udot(<vscale x 4 x i32> %acc, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: udot:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: udot z0.s, z1.h, z2.h
+; CHECK-NEXT: ret
+entry:
+ %a.wide = zext <vscale x 8 x i16> %a to <vscale x 8 x i32>
+ %b.wide = zext <vscale x 8 x i16> %b to <vscale x 8 x i32>
+ %mult = mul nuw nsw <vscale x 8 x i32> %a.wide, %b.wide
+ %partial.reduce = tail call <vscale x 4 x i32> @llvm.experimental.vector.partial.reduce.add(<vscale x 4 x i32> %acc, <vscale x 8 x i32> %mult)
+ ret <vscale x 4 x i32> %partial.reduce
+}
+
+define <vscale x 4 x i32> @sdot(<vscale x 4 x i32> %acc, <vscale x 8 x i16> %a, <vscale x 8 x i16> %b) {
+; CHECK-LABEL: sdot:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: sdot z0.s, z1.h, z2.h
+; CHECK-NEXT: ret
+entry:
+ %a.wide = sext <vscale x 8 x i16> %a to <vscale x 8 x i32>
+ %b.wide = sext <vscale x 8 x i16> %b to <vscale x 8 x i32>
+ %mult = mul nuw nsw <vscale x 8 x i32> %a.wide, %b.wide
+ %partial.reduce = tail call <vscale x 4 x i32> @llvm.experimental.vector.partial.reduce.add(<vscale x 4 x i32> %acc, <vscale x 8 x i32> %mult)
+ ret <vscale x 4 x i32> %partial.reduce
+}
+
+define <4 x i32> @fixed_udot_s_h(<4 x i32> %acc, <8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: fixed_udot_s_h:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: // kill: def $q2 killed $q2 def $z2
+; CHECK-NEXT: // kill: def $q1 killed $q1 def $z1
+; CHECK-NEXT: udot z0.s, z1.h, z2.h
+; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z0
+; CHECK-NEXT: ret
+entry:
+ %a.wide = zext <8 x i16> %a to <8 x i32>
+ %b.wide = zext <8 x i16> %b to <8 x i32>
+ %mult = mul nuw nsw <8 x i32> %a.wide, %b.wide
+ %partial.reduce = tail call <4 x i32> @llvm.experimental.vector.partial.reduce.add(<4 x i32> %acc, <8 x i32> %mult)
+ ret <4 x i32> %partial.reduce
+}
+
+define <4 x i32> @fixed_sdot_s_h(<4 x i32> %acc, <8 x i16> %a, <8 x i16> %b) {
+; CHECK-LABEL: fixed_sdot_s_h:
+; CHECK: // %bb.0: // %entry
+; CHECK-NEXT: // kill: def $q0 killed $q0 def $z0
+; CHECK-NEXT: // kill: def $q2 killed $q2 def $z2
+; CHECK-NEXT: // kill: def $q1 killed $q1 def $z1
+; CHECK-NEXT: sdot z0.s, z1.h, z2.h
+; CHECK-NEXT: // kill: def $q0 killed $q0 killed $z0
+; CHECK-NEXT: ret
+entry:
+ %a.wide = sext <8 x i16> %a to <8 x i32>
+ %b.wide = sext <8 x i16> %b to <8 x i32>
+ %mult = mul nuw nsw <8 x i32> %a.wide, %b.wide
+ %partial.reduce = tail call <4 x i32> @llvm.experimental.vector.partial.reduce.add(<4 x i32> %acc, <8 x i32> %mult)
+ ret <4 x i32> %partial.reduce
+}
|
…sdot This allows dot products with scalable 8xi16 vectors (and fixed-length vectors which are converted into a scalable vector) accumulating into a 4xi32 vector to lower into a single instruction (`udot`/`sdot`), rather than a sequence of `umlalb`s and `umlalt`s`.
dheaton-arm
force-pushed
the
sve2p1-dots-partial-reduction
branch
from
35e983a to
b8aa1d0
Compare
|
@sdesmalen-arm Could you merge for me? I don't have write access. |
|
@dheaton-arm Congratulations on having your first Pull Request (PR) merged into the LLVM Project! Your changes will be combined with recent changes from other authors, then tested by our build bots. If there is a problem with a build, you may receive a report in an email or a comment on this PR. Please check whether problems have been caused by your change specifically, as the builds can include changes from many authors. It is not uncommon for your change to be included in a build that fails due to someone else's changes, or infrastructure issues. How to do this, and the rest of the post-merge process, is covered in detail here. If your change does cause a problem, it may be reverted, or you can revert it yourself. This is a normal part of LLVM development. You can fix your changes and open a new PR to merge them again. If you don't get any reports, no action is required from you. Your changes are working as expected, well done! |
3 participants
This allows dot products with scalable 8xi16 vectors (and fixed-length vectors which are converted into a scalable vector) accumulating into a 4xi32 vector to lower into a single instruction (
udot/sdot), rather than a sequence ofumlalbs andumlalts`.