caio/clang-p2996
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

[RISCV] Don't commute with shift if XAndesPerf is enabled (#142920)

Browse Source 
More nds.lea.{h,w,d} are generated, similar to sh{1,2,3}add
This commit is contained in:
Jim Lin
2025-06-06 11:08:23 +08:00
committed by GitHub
parent 05c12b2284
commit f8df24015a
3 changed files with 176 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
llvm/lib/Target/RISCV/RISCVISelLowering.cpp
View File

@@ -20413,8 +20413,8 @@ bool RISCVTargetLowering::isDesirableToCommuteWithShift(
auto *C2 = dyn_cast<ConstantSDNode>(N->getOperand(1));
// Bail if we might break a sh{1,2,3}add pattern.
if (Subtarget.hasStdExtZba() && C2 && C2->getZExtValue() >= 1 &&
C2->getZExtValue() <= 3 && N->hasOneUse() &&
if ((Subtarget.hasStdExtZba() || Subtarget.hasVendorXAndesPerf()) && C2 &&
C2->getZExtValue() >= 1 && C2->getZExtValue() <= 3 && N->hasOneUse() &&
N->user_begin()->getOpcode() == ISD::ADD &&
!isUsedByLdSt(*N->user_begin(), nullptr) &&
!isa<ConstantSDNode>(N->user_begin()->getOperand(1)))

97
llvm/test/CodeGen/RISCV/add_sext_shl_constant.ll
View File

@@ -1,6 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 3
; RUN: llc -mtriple=riscv64 < %s | FileCheck -check-prefixes=RV64,NO-ZBA %s
; RUN: llc -mtriple=riscv64 -mattr=+zba < %s | FileCheck -check-prefixes=RV64,ZBA %s
; RUN: llc -mtriple=riscv64 -mattr=+xandesperf < %s | FileCheck -check-prefixes=RV64,XANDESPERF %s
define void @add_sext_shl_moreOneUse_add(ptr %array1, i32 %a, i32 %b) {
; NO-ZBA-LABEL: add_sext_shl_moreOneUse_add:
@@ -23,6 +24,16 @@ define void @add_sext_shl_moreOneUse_add(ptr %array1, i32 %a, i32 %b) {
; ZBA-NEXT: sw a2, 24(a0)
; ZBA-NEXT: sw a3, 140(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_sext_shl_moreOneUse_add:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a3, a1, 5
; XANDESPERF-NEXT: sext.w a1, a1
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: sw a2, 20(a0)
; XANDESPERF-NEXT: sw a2, 24(a0)
; XANDESPERF-NEXT: sw a3, 140(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 5
%idxprom = sext i32 %add to i64
@@ -66,6 +77,19 @@ define void @add_sext_shl_moreOneUse_addexceedsign12(ptr %array1, i32 %a, i32 %b
; ZBA-NEXT: sw a3, 4(a0)
; ZBA-NEXT: sw a2, 120(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_sext_shl_moreOneUse_addexceedsign12:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a3, a1, 2047
; XANDESPERF-NEXT: lui a4, 2
; XANDESPERF-NEXT: sext.w a1, a1
; XANDESPERF-NEXT: addi a3, a3, 1
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: add a0, a0, a4
; XANDESPERF-NEXT: sw a2, 0(a0)
; XANDESPERF-NEXT: sw a3, 4(a0)
; XANDESPERF-NEXT: sw a2, 120(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 2048
%idxprom = sext i32 %add to i64
@@ -101,6 +125,16 @@ define void @add_sext_shl_moreOneUse_sext(ptr %array1, i32 %a, i32 %b) {
; ZBA-NEXT: sw a2, 24(a0)
; ZBA-NEXT: sd a3, 140(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_sext_shl_moreOneUse_sext:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: sext.w a1, a1
; XANDESPERF-NEXT: addi a3, a1, 5
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: sw a2, 20(a0)
; XANDESPERF-NEXT: sw a2, 24(a0)
; XANDESPERF-NEXT: sd a3, 140(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 5
%idxprom = sext i32 %add to i64
@@ -147,6 +181,20 @@ define void @add_sext_shl_moreOneUse_add_inSelect(ptr %array1, i32 signext %a,
; ZBA-NEXT: sw a5, 24(a0)
; ZBA-NEXT: sw a4, 140(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_sext_shl_moreOneUse_add_inSelect:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a4, a1, 5
; XANDESPERF-NEXT: mv a5, a4
; XANDESPERF-NEXT: bgtz a3, .LBB3_2
; XANDESPERF-NEXT: # %bb.1: # %entry
; XANDESPERF-NEXT: mv a5, a2
; XANDESPERF-NEXT: .LBB3_2: # %entry
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: sw a5, 20(a0)
; XANDESPERF-NEXT: sw a5, 24(a0)
; XANDESPERF-NEXT: sw a4, 140(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 5
%cmp = icmp sgt i32 %x, 0
@@ -200,6 +248,23 @@ define void @add_sext_shl_moreOneUse_add_inSelect_addexceedsign12(ptr %array1, i
; ZBA-NEXT: sw a1, 4(a0)
; ZBA-NEXT: sw a4, 120(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_sext_shl_moreOneUse_add_inSelect_addexceedsign12:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a4, a1, 2047
; XANDESPERF-NEXT: lui a5, 2
; XANDESPERF-NEXT: addi a4, a4, 1
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: add a0, a0, a5
; XANDESPERF-NEXT: mv a1, a4
; XANDESPERF-NEXT: bgtz a3, .LBB4_2
; XANDESPERF-NEXT: # %bb.1: # %entry
; XANDESPERF-NEXT: mv a1, a2
; XANDESPERF-NEXT: .LBB4_2: # %entry
; XANDESPERF-NEXT: sw a1, 0(a0)
; XANDESPERF-NEXT: sw a1, 4(a0)
; XANDESPERF-NEXT: sw a4, 120(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 2048
%cmp = icmp sgt i32 %x, 0
@@ -245,6 +310,20 @@ define void @add_shl_moreOneUse_inSelect(ptr %array1, i64 %a, i64 %b, i64 %x) {
; ZBA-NEXT: sd a5, 48(a0)
; ZBA-NEXT: sd a4, 280(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_inSelect:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a4, a1, 5
; XANDESPERF-NEXT: mv a5, a4
; XANDESPERF-NEXT: bgtz a3, .LBB5_2
; XANDESPERF-NEXT: # %bb.1: # %entry
; XANDESPERF-NEXT: mv a5, a2
; XANDESPERF-NEXT: .LBB5_2: # %entry
; XANDESPERF-NEXT: nds.lea.d a0, a0, a1
; XANDESPERF-NEXT: sd a5, 40(a0)
; XANDESPERF-NEXT: sd a5, 48(a0)
; XANDESPERF-NEXT: sd a4, 280(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i64 %a, 5
%cmp = icmp sgt i64 %x, 0
@@ -274,6 +353,12 @@ define i64 @add_shl_moreOneUse_sh1add(i64 %x) {
; ZBA-NEXT: ori a0, a0, 1
; ZBA-NEXT: sh1add a0, a0, a0
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_sh1add:
; XANDESPERF: # %bb.0:
; XANDESPERF-NEXT: ori a0, a0, 1
; XANDESPERF-NEXT: nds.lea.h a0, a0, a0
; XANDESPERF-NEXT: ret
%or = or i64 %x, 1
%mul = shl i64 %or, 1
%add = add i64 %mul, %or
@@ -294,6 +379,12 @@ define i64 @add_shl_moreOneUse_sh2add(i64 %x) {
; ZBA-NEXT: ori a0, a0, 1
; ZBA-NEXT: sh2add a0, a0, a0
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_sh2add:
; XANDESPERF: # %bb.0:
; XANDESPERF-NEXT: ori a0, a0, 1
; XANDESPERF-NEXT: nds.lea.w a0, a0, a0
; XANDESPERF-NEXT: ret
%or = or i64 %x, 1
%mul = shl i64 %or, 2
%add = add i64 %mul, %or
@@ -314,6 +405,12 @@ define i64 @add_shl_moreOneUse_sh3add(i64 %x) {
; ZBA-NEXT: ori a0, a0, 1
; ZBA-NEXT: sh3add a0, a0, a0
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_sh3add:
; XANDESPERF: # %bb.0:
; XANDESPERF-NEXT: ori a0, a0, 1
; XANDESPERF-NEXT: nds.lea.d a0, a0, a0
; XANDESPERF-NEXT: ret
%or = or i64 %x, 1
%mul = shl i64 %or, 3
%add = add i64 %mul, %or

77
llvm/test/CodeGen/RISCV/add_shl_constant.ll
View File

@@ -1,6 +1,7 @@
; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
; RUN: llc -mtriple=riscv32 < %s | FileCheck -check-prefixes=RV32,NO-ZBA %s
; RUN: llc -mtriple=riscv32 -mattr=+zba < %s | FileCheck -check-prefixes=RV32,ZBA %s
; RUN: llc -mtriple=riscv32 -mattr=+xandesperf < %s | FileCheck -check-prefixes=RV32,XANDESPERF %s
define i32 @add_shl_oneUse(i32 %x, i32 %y) nounwind {
; NO-ZBA-LABEL: add_shl_oneUse:
@@ -15,6 +16,12 @@ define i32 @add_shl_oneUse(i32 %x, i32 %y) nounwind {
; ZBA-NEXT: addi a0, a0, 123
; ZBA-NEXT: sh3add a0, a0, a1
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_oneUse:
; XANDESPERF: # %bb.0:
; XANDESPERF-NEXT: addi a0, a0, 123
; XANDESPERF-NEXT: nds.lea.d a0, a1, a0
; XANDESPERF-NEXT: ret
%add.0 = add i32 %x, 123
%shl = shl i32 %add.0, 3
%add.1 = add i32 %shl, %y
@@ -40,6 +47,15 @@ define void @add_shl_moreOneUse_inStore(ptr %array1, i32 %a, i32 %b) {
; ZBA-NEXT: sw a2, 24(a0)
; ZBA-NEXT: sw a3, 140(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_inStore:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a3, a1, 5
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: sw a2, 20(a0)
; XANDESPERF-NEXT: sw a2, 24(a0)
; XANDESPERF-NEXT: sw a3, 140(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 5
%arrayidx = getelementptr inbounds i32, ptr %array1, i32 %add
@@ -77,6 +93,18 @@ define void @add_shl_moreOneUse_inStore_addexceedsign12(ptr %array1, i32 %a, i32
; ZBA-NEXT: sw a3, 4(a0)
; ZBA-NEXT: sw a2, 120(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_inStore_addexceedsign12:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a3, a1, 2047
; XANDESPERF-NEXT: lui a4, 2
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: addi a3, a3, 1
; XANDESPERF-NEXT: add a0, a0, a4
; XANDESPERF-NEXT: sw a2, 0(a0)
; XANDESPERF-NEXT: sw a3, 4(a0)
; XANDESPERF-NEXT: sw a2, 120(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 2048
%arrayidx = getelementptr inbounds i32, ptr %array1, i32 %add
@@ -118,6 +146,20 @@ define void @add_shl_moreOneUse_inSelect(ptr %array1, i32 %a, i32 %b, i32 %x) {
; ZBA-NEXT: sw a5, 24(a0)
; ZBA-NEXT: sw a4, 140(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_inSelect:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a4, a1, 5
; XANDESPERF-NEXT: mv a5, a4
; XANDESPERF-NEXT: bgtz a3, .LBB3_2
; XANDESPERF-NEXT: # %bb.1: # %entry
; XANDESPERF-NEXT: mv a5, a2
; XANDESPERF-NEXT: .LBB3_2: # %entry
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: sw a5, 20(a0)
; XANDESPERF-NEXT: sw a5, 24(a0)
; XANDESPERF-NEXT: sw a4, 140(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 5
%cmp = icmp sgt i32 %x, 0
@@ -167,6 +209,23 @@ define void @add_shl_moreOneUse_inSelect_addexceedsign12(ptr %array1, i32 %a, i3
; ZBA-NEXT: sw a5, 4(a0)
; ZBA-NEXT: sw a4, 120(a0)
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_inSelect_addexceedsign12:
; XANDESPERF: # %bb.0: # %entry
; XANDESPERF-NEXT: addi a4, a1, 2047
; XANDESPERF-NEXT: addi a4, a4, 1
; XANDESPERF-NEXT: mv a5, a4
; XANDESPERF-NEXT: bgtz a3, .LBB4_2
; XANDESPERF-NEXT: # %bb.1: # %entry
; XANDESPERF-NEXT: mv a5, a2
; XANDESPERF-NEXT: .LBB4_2: # %entry
; XANDESPERF-NEXT: lui a2, 2
; XANDESPERF-NEXT: nds.lea.w a0, a0, a1
; XANDESPERF-NEXT: add a0, a0, a2
; XANDESPERF-NEXT: sw a5, 0(a0)
; XANDESPERF-NEXT: sw a5, 4(a0)
; XANDESPERF-NEXT: sw a4, 120(a0)
; XANDESPERF-NEXT: ret
entry:
%add = add nsw i32 %a, 2048
%cmp = icmp sgt i32 %x, 0
@@ -195,6 +254,12 @@ define i32 @add_shl_moreOneUse_sh1add(i32 %x) {
; ZBA-NEXT: ori a0, a0, 1
; ZBA-NEXT: sh1add a0, a0, a0
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_sh1add:
; XANDESPERF: # %bb.0:
; XANDESPERF-NEXT: ori a0, a0, 1
; XANDESPERF-NEXT: nds.lea.h a0, a0, a0
; XANDESPERF-NEXT: ret
%or = or i32 %x, 1
%mul = shl i32 %or, 1
%add = add i32 %mul, %or
@@ -215,6 +280,12 @@ define i32 @add_shl_moreOneUse_sh2add(i32 %x) {
; ZBA-NEXT: ori a0, a0, 1
; ZBA-NEXT: sh2add a0, a0, a0
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_sh2add:
; XANDESPERF: # %bb.0:
; XANDESPERF-NEXT: ori a0, a0, 1
; XANDESPERF-NEXT: nds.lea.w a0, a0, a0
; XANDESPERF-NEXT: ret
%or = or i32 %x, 1
%mul = shl i32 %or, 2
%add = add i32 %mul, %or
@@ -235,6 +306,12 @@ define i32 @add_shl_moreOneUse_sh3add(i32 %x) {
; ZBA-NEXT: ori a0, a0, 1
; ZBA-NEXT: sh3add a0, a0, a0
; ZBA-NEXT: ret
;
; XANDESPERF-LABEL: add_shl_moreOneUse_sh3add:
; XANDESPERF: # %bb.0:
; XANDESPERF-NEXT: ori a0, a0, 1
; XANDESPERF-NEXT: nds.lea.d a0, a0, a0
; XANDESPERF-NEXT: ret
%or = or i32 %x, 1
%mul = shl i32 %or, 3
%add = add i32 %mul, %or