[llvm] [RISCV] Lower (abd[s|u] a, b) to (abs (sub a, b)) for i32/i64 element when `Zvabd` exists. (PR #180494)
via llvm-commits
llvm-commits at lists.llvm.org
Mon Feb 9 02:09:18 PST 2026
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-risc-v
Author: Pengcheng Wang (wangpc-pp)
<details>
<summary>Changes</summary>
This saves one instruction for these cases.
---
Full diff: https://github.com/llvm/llvm-project/pull/180494.diff
3 Files Affected:
- (modified) llvm/lib/Target/RISCV/RISCVISelLowering.cpp (+10-7)
- (modified) llvm/test/CodeGen/RISCV/rvv/abd.ll (+22-27)
- (modified) llvm/test/CodeGen/RISCV/rvv/fixed-vectors-abd.ll (+28-42)
``````````diff
diff --git a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
index 9b88bc5c39ce4..b6373a187bc1b 100644
--- a/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
+++ b/llvm/lib/Target/RISCV/RISCVISelLowering.cpp
@@ -8839,17 +8839,20 @@ SDValue RISCVTargetLowering::LowerOperation(SDValue Op,
return lowerToScalableOp(Op, DAG);
case ISD::ABDS:
case ISD::ABDU: {
- EVT VT = Op->getValueType(0);
- // Only SEW=8/16 are supported in Zvabd.
- if (Subtarget.hasStdExtZvabd() && VT.isVector() &&
- (VT.getVectorElementType() == MVT::i8 ||
- VT.getVectorElementType() == MVT::i16))
- return lowerToScalableOp(Op, DAG);
-
SDLoc dl(Op);
+ EVT VT = Op->getValueType(0);
SDValue LHS = DAG.getFreeze(Op->getOperand(0));
SDValue RHS = DAG.getFreeze(Op->getOperand(1));
bool IsSigned = Op->getOpcode() == ISD::ABDS;
+ if (Subtarget.hasStdExtZvabd() && VT.isVector()) {
+ // Only SEW=8/16 are supported in Zvabd.
+ if (VT.getVectorElementType() == MVT::i8 ||
+ VT.getVectorElementType() == MVT::i16)
+ return lowerToScalableOp(Op, DAG);
+ // For SEW=32/64, we lower it to (abs (sub lhs, rhs)).
+ return DAG.getNode(ISD::ABS, dl, VT,
+ DAG.getNode(ISD::SUB, dl, VT, LHS, RHS));
+ }
// abds(lhs, rhs) -> sub(smax(lhs,rhs), smin(lhs,rhs))
// abdu(lhs, rhs) -> sub(umax(lhs,rhs), umin(lhs,rhs))
diff --git a/llvm/test/CodeGen/RISCV/rvv/abd.ll b/llvm/test/CodeGen/RISCV/rvv/abd.ll
index c451559a29a69..b39a07e2b0de9 100644
--- a/llvm/test/CodeGen/RISCV/rvv/abd.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/abd.ll
@@ -112,9 +112,8 @@ define <vscale x 4 x i32> @sabd_s(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b)
; ZVABD-LABEL: sabd_s:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetvli a0, zero, e32, m2, ta, ma
-; ZVABD-NEXT: vmin.vv v12, v8, v10
-; ZVABD-NEXT: vmax.vv v8, v8, v10
-; ZVABD-NEXT: vsub.vv v8, v8, v12
+; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.sext = sext <vscale x 4 x i32> %a to <vscale x 4 x i64>
%b.sext = sext <vscale x 4 x i32> %b to <vscale x 4 x i64>
@@ -161,9 +160,8 @@ define <vscale x 2 x i64> @sabd_d(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b)
; ZVABD-LABEL: sabd_d:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetvli a0, zero, e64, m2, ta, ma
-; ZVABD-NEXT: vmin.vv v12, v8, v10
-; ZVABD-NEXT: vmax.vv v8, v8, v10
-; ZVABD-NEXT: vsub.vv v8, v8, v12
+; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.sext = sext <vscale x 2 x i64> %a to <vscale x 2 x i128>
%b.sext = sext <vscale x 2 x i64> %b to <vscale x 2 x i128>
@@ -187,9 +185,8 @@ define <vscale x 2 x i64> @sabd_d_promoted_ops(<vscale x 2 x i32> %a, <vscale x
; ZVABD-LABEL: sabd_d_promoted_ops:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetvli a0, zero, e32, m1, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v10, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v10, v8
; ZVABD-NEXT: vsetvli zero, zero, e64, m2, ta, ma
; ZVABD-NEXT: vzext.vf2 v8, v10
; ZVABD-NEXT: ret
@@ -308,9 +305,8 @@ define <vscale x 4 x i32> @uabd_s(<vscale x 4 x i32> %a, <vscale x 4 x i32> %b)
; ZVABD-LABEL: uabd_s:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetvli a0, zero, e32, m2, ta, ma
-; ZVABD-NEXT: vminu.vv v12, v8, v10
-; ZVABD-NEXT: vmaxu.vv v8, v8, v10
-; ZVABD-NEXT: vsub.vv v8, v8, v12
+; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.zext = zext <vscale x 4 x i32> %a to <vscale x 4 x i64>
%b.zext = zext <vscale x 4 x i32> %b to <vscale x 4 x i64>
@@ -357,9 +353,8 @@ define <vscale x 2 x i64> @uabd_d(<vscale x 2 x i64> %a, <vscale x 2 x i64> %b)
; ZVABD-LABEL: uabd_d:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetvli a0, zero, e64, m2, ta, ma
-; ZVABD-NEXT: vminu.vv v12, v8, v10
-; ZVABD-NEXT: vmaxu.vv v8, v8, v10
-; ZVABD-NEXT: vsub.vv v8, v8, v12
+; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.zext = zext <vscale x 2 x i64> %a to <vscale x 2 x i128>
%b.zext = zext <vscale x 2 x i64> %b to <vscale x 2 x i128>
@@ -383,9 +378,8 @@ define <vscale x 2 x i64> @uabd_d_promoted_ops(<vscale x 2 x i32> %a, <vscale x
; ZVABD-LABEL: uabd_d_promoted_ops:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetvli a0, zero, e32, m1, ta, ma
-; ZVABD-NEXT: vminu.vv v10, v8, v9
-; ZVABD-NEXT: vmaxu.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v10, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v10, v8
; ZVABD-NEXT: vsetvli zero, zero, e64, m2, ta, ma
; ZVABD-NEXT: vzext.vf2 v8, v10
; ZVABD-NEXT: ret
@@ -410,11 +404,11 @@ define <vscale x 4 x i32> @uabd_non_matching_extension(<vscale x 4 x i32> %a, <v
;
; ZVABD-LABEL: uabd_non_matching_extension:
; ZVABD: # %bb.0:
-; ZVABD-NEXT: vsetvli a0, zero, e32, m2, ta, ma
-; ZVABD-NEXT: vzext.vf4 v12, v10
-; ZVABD-NEXT: vminu.vv v10, v8, v12
-; ZVABD-NEXT: vmaxu.vv v8, v8, v12
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsetvli a0, zero, e16, m1, ta, ma
+; ZVABD-NEXT: vzext.vf2 v11, v10
+; ZVABD-NEXT: vwsubu.wv v8, v8, v11
+; ZVABD-NEXT: vsetvli zero, zero, e32, m2, ta, ma
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.zext = zext <vscale x 4 x i32> %a to <vscale x 4 x i64>
%b.zext = zext <vscale x 4 x i8> %b to <vscale x 4 x i64>
@@ -470,10 +464,11 @@ define <vscale x 4 x i32> @uabd_non_matching_promotion(<vscale x 4 x i8> %a, <vs
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetvli a0, zero, e32, m2, ta, ma
; ZVABD-NEXT: vzext.vf4 v10, v8
-; ZVABD-NEXT: vsext.vf4 v12, v9
-; ZVABD-NEXT: vmin.vv v8, v10, v12
-; ZVABD-NEXT: vmax.vv v10, v10, v12
-; ZVABD-NEXT: vsub.vv v8, v10, v8
+; ZVABD-NEXT: vsetvli zero, zero, e16, m1, ta, ma
+; ZVABD-NEXT: vsext.vf2 v8, v9
+; ZVABD-NEXT: vwsub.wv v10, v10, v8
+; ZVABD-NEXT: vsetvli zero, zero, e32, m2, ta, ma
+; ZVABD-NEXT: vabs.v v8, v10
; ZVABD-NEXT: ret
%a.zext = zext <vscale x 4 x i8> %a to <vscale x 4 x i32>
%b.zext = sext <vscale x 4 x i8> %b to <vscale x 4 x i32>
diff --git a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-abd.ll b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-abd.ll
index 6bfac12fa3b99..9225c9514c9d6 100644
--- a/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-abd.ll
+++ b/llvm/test/CodeGen/RISCV/rvv/fixed-vectors-abd.ll
@@ -187,9 +187,8 @@ define <2 x i32> @sabd_2s(<2 x i32> %a, <2 x i32> %b) {
; ZVABD-LABEL: sabd_2s:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.sext = sext <2 x i32> %a to <2 x i64>
%b.sext = sext <2 x i32> %b to <2 x i64>
@@ -238,9 +237,8 @@ define <4 x i32> @sabd_4s(<4 x i32> %a, <4 x i32> %b) {
; ZVABD-LABEL: sabd_4s:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 4, e32, m1, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.sext = sext <4 x i32> %a to <4 x i64>
%b.sext = sext <4 x i32> %b to <4 x i64>
@@ -288,9 +286,8 @@ define <2 x i64> @sabd_2d(<2 x i64> %a, <2 x i64> %b) {
; ZVABD-LABEL: sabd_2d:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e64, m1, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.sext = sext <2 x i64> %a to <2 x i128>
%b.sext = sext <2 x i64> %b to <2 x i128>
@@ -315,9 +312,8 @@ define <2 x i64> @sabd_2d_promoted_ops(<2 x i32> %a, <2 x i32> %b) {
; ZVABD-LABEL: sabd_2d_promoted_ops:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v9, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v9, v8
; ZVABD-NEXT: vsetvli zero, zero, e64, m1, ta, ma
; ZVABD-NEXT: vzext.vf2 v8, v9
; ZVABD-NEXT: ret
@@ -489,9 +485,8 @@ define <2 x i32> @uabd_2s(<2 x i32> %a, <2 x i32> %b) {
; ZVABD-LABEL: uabd_2s:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
-; ZVABD-NEXT: vminu.vv v10, v8, v9
-; ZVABD-NEXT: vmaxu.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.zext = zext <2 x i32> %a to <2 x i64>
%b.zext = zext <2 x i32> %b to <2 x i64>
@@ -540,9 +535,8 @@ define <4 x i32> @uabd_4s(<4 x i32> %a, <4 x i32> %b) {
; ZVABD-LABEL: uabd_4s:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 4, e32, m1, ta, ma
-; ZVABD-NEXT: vminu.vv v10, v8, v9
-; ZVABD-NEXT: vmaxu.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.zext = zext <4 x i32> %a to <4 x i64>
%b.zext = zext <4 x i32> %b to <4 x i64>
@@ -590,9 +584,8 @@ define <2 x i64> @uabd_2d(<2 x i64> %a, <2 x i64> %b) {
; ZVABD-LABEL: uabd_2d:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e64, m1, ta, ma
-; ZVABD-NEXT: vminu.vv v10, v8, v9
-; ZVABD-NEXT: vmaxu.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a.zext = zext <2 x i64> %a to <2 x i128>
%b.zext = zext <2 x i64> %b to <2 x i128>
@@ -617,9 +610,8 @@ define <2 x i64> @uabd_2d_promoted_ops(<2 x i32> %a, <2 x i32> %b) {
; ZVABD-LABEL: uabd_2d_promoted_ops:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e32, mf2, ta, ma
-; ZVABD-NEXT: vminu.vv v10, v8, v9
-; ZVABD-NEXT: vmaxu.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v9, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v9, v8
; ZVABD-NEXT: vsetvli zero, zero, e64, m1, ta, ma
; ZVABD-NEXT: vzext.vf2 v8, v9
; ZVABD-NEXT: ret
@@ -767,9 +759,8 @@ define <4 x i32> @sabd_v4i32_nsw(<4 x i32> %a, <4 x i32> %b) {
; ZVABD-LABEL: sabd_v4i32_nsw:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 4, e32, m1, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%sub = sub nsw <4 x i32> %a, %b
%abs = call <4 x i32> @llvm.abs.v4i32(<4 x i32> %sub, i1 true)
@@ -789,9 +780,8 @@ define <2 x i64> @sabd_v2i64_nsw(<2 x i64> %a, <2 x i64> %b) {
; ZVABD-LABEL: sabd_v2i64_nsw:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e64, m1, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%sub = sub nsw <2 x i64> %a, %b
%abs = call <2 x i64> @llvm.abs.v2i64(<2 x i64> %sub, i1 true)
@@ -853,9 +843,8 @@ define <4 x i32> @smaxmin_v4i32(<4 x i32> %0, <4 x i32> %1) {
; ZVABD-LABEL: smaxmin_v4i32:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 4, e32, m1, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a = tail call <4 x i32> @llvm.smax.v4i32(<4 x i32> %0, <4 x i32> %1)
%b = tail call <4 x i32> @llvm.smin.v4i32(<4 x i32> %0, <4 x i32> %1)
@@ -876,9 +865,8 @@ define <2 x i64> @smaxmin_v2i64(<2 x i64> %0, <2 x i64> %1) {
; ZVABD-LABEL: smaxmin_v2i64:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e64, m1, ta, ma
-; ZVABD-NEXT: vmin.vv v10, v8, v9
-; ZVABD-NEXT: vmax.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a = tail call <2 x i64> @llvm.smax.v2i64(<2 x i64> %0, <2 x i64> %1)
%b = tail call <2 x i64> @llvm.smin.v2i64(<2 x i64> %0, <2 x i64> %1)
@@ -941,9 +929,8 @@ define <4 x i32> @umaxmin_v4i32(<4 x i32> %0, <4 x i32> %1) {
; ZVABD-LABEL: umaxmin_v4i32:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 4, e32, m1, ta, ma
-; ZVABD-NEXT: vminu.vv v10, v8, v9
-; ZVABD-NEXT: vmaxu.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a = tail call <4 x i32> @llvm.umax.v4i32(<4 x i32> %0, <4 x i32> %1)
%b = tail call <4 x i32> @llvm.umin.v4i32(<4 x i32> %0, <4 x i32> %1)
@@ -964,9 +951,8 @@ define <2 x i64> @umaxmin_v2i64(<2 x i64> %0, <2 x i64> %1) {
; ZVABD-LABEL: umaxmin_v2i64:
; ZVABD: # %bb.0:
; ZVABD-NEXT: vsetivli zero, 2, e64, m1, ta, ma
-; ZVABD-NEXT: vminu.vv v10, v8, v9
-; ZVABD-NEXT: vmaxu.vv v8, v8, v9
-; ZVABD-NEXT: vsub.vv v8, v8, v10
+; ZVABD-NEXT: vsub.vv v8, v8, v9
+; ZVABD-NEXT: vabs.v v8, v8
; ZVABD-NEXT: ret
%a = tail call <2 x i64> @llvm.umax.v2i64(<2 x i64> %0, <2 x i64> %1)
%b = tail call <2 x i64> @llvm.umin.v2i64(<2 x i64> %0, <2 x i64> %1)
``````````
</details>
https://github.com/llvm/llvm-project/pull/180494
More information about the llvm-commits
mailing list