[llvm] [LoongArch] Set isReMaterializable on LU{12,32,52}I.D/ADDI.D and {X}ORI instructions (PR #94552)
via llvm-commits
llvm-commits at lists.llvm.org
Wed Jun 5 18:54:17 PDT 2024
https://github.com/heiher created https://github.com/llvm/llvm-project/pull/94552
None
>From 675206365e9d938b5662f477fd8cec05f65333e1 Mon Sep 17 00:00:00 2001
From: WANG Rui <wangrui at loongson.cn>
Date: Fri, 31 May 2024 15:19:42 +0800
Subject: [PATCH] [LoongArch] Set isReMaterializable on LU{12,32,52}I.D/ADDI.D
and {X}ORI instructions
---
.../Target/LoongArch/LoongArchInstrInfo.td | 10 ++-
llvm/test/CodeGen/LoongArch/vector-fp-imm.ll | 73 +++++++++----------
2 files changed, 43 insertions(+), 40 deletions(-)
diff --git a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td
index 66bd74e068b95..6b6641aef9030 100644
--- a/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td
+++ b/llvm/lib/Target/LoongArch/LoongArchInstrInfo.td
@@ -734,6 +734,7 @@ def ADD_W : ALU_3R<0x00100000>;
def SUB_W : ALU_3R<0x00110000>;
def ADDI_W : ALU_2RI12<0x02800000, simm12_addlike>;
def ALSL_W : ALU_3RI2<0x00040000, uimm2_plus1>;
+let isReMaterializable = 1 in
def LU12I_W : ALU_1RI20<0x14000000, simm20_lu12iw>;
def SLT : ALU_3R<0x00120000>;
def SLTU : ALU_3R<0x00128000>;
@@ -749,8 +750,10 @@ def XOR : ALU_3R<0x00158000>;
def ANDN : ALU_3R<0x00168000>;
def ORN : ALU_3R<0x00160000>;
def ANDI : ALU_2RI12<0x03400000, uimm12>;
+let isReMaterializable = 1 in {
def ORI : ALU_2RI12<0x03800000, uimm12_ori>;
def XORI : ALU_2RI12<0x03c00000, uimm12>;
+}
def MUL_W : ALU_3R<0x001c0000>;
def MULH_W : ALU_3R<0x001c8000>;
def MULH_WU : ALU_3R<0x001d0000>;
@@ -852,16 +855,21 @@ let Predicates = [IsLA64] in {
// Arithmetic Operation Instructions for 64-bits
def ADD_D : ALU_3R<0x00108000>;
def SUB_D : ALU_3R<0x00118000>;
+// ADDI_D isn't always rematerializable, but isReMaterializable will be used as
+// a hint which is verified in isReallyTriviallyReMaterializable.
+let isReMaterializable = 1 in
def ADDI_D : ALU_2RI12<0x02c00000, simm12_addlike>;
def ADDU16I_D : ALU_2RI16<0x10000000, simm16>;
def ALSL_WU : ALU_3RI2<0x00060000, uimm2_plus1>;
def ALSL_D : ALU_3RI2<0x002c0000, uimm2_plus1>;
let Constraints = "$rd = $dst" in {
-let hasSideEffects = 0, mayLoad = 0, mayStore = 0 in
+let hasSideEffects = 0, mayLoad = 0, mayStore = 0,
+ isReMaterializable = 1 in
def LU32I_D : Fmt1RI20<0x16000000, (outs GPR:$dst),
(ins GPR:$rd, simm20_lu32id:$imm20),
"$rd, $imm20">;
}
+let isReMaterializable = 1 in
def LU52I_D : ALU_2RI12<0x03000000, simm12_lu52id>;
def PCADDU18I : ALU_1RI20<0x1e000000, simm20_pcaddu18i>;
def MUL_D : ALU_3R<0x001d8000>;
diff --git a/llvm/test/CodeGen/LoongArch/vector-fp-imm.ll b/llvm/test/CodeGen/LoongArch/vector-fp-imm.ll
index 18d17751a7719..0a401ebe5f6b2 100644
--- a/llvm/test/CodeGen/LoongArch/vector-fp-imm.ll
+++ b/llvm/test/CodeGen/LoongArch/vector-fp-imm.ll
@@ -742,12 +742,10 @@ define void @test_d8(ptr %P, ptr %S) nounwind {
; LA32F-NEXT: st.w $a2, $sp, 32 # 4-byte Folded Spill
; LA32F-NEXT: ld.w $a2, $a0, 52
; LA32F-NEXT: st.w $a2, $sp, 28 # 4-byte Folded Spill
-; LA32F-NEXT: ld.w $a2, $a0, 40
-; LA32F-NEXT: st.w $a2, $sp, 16 # 4-byte Folded Spill
+; LA32F-NEXT: ld.w $s8, $a0, 40
; LA32F-NEXT: ld.w $a2, $a0, 44
-; LA32F-NEXT: st.w $a2, $sp, 12 # 4-byte Folded Spill
-; LA32F-NEXT: ld.w $a2, $a0, 32
-; LA32F-NEXT: st.w $a2, $sp, 0 # 4-byte Folded Spill
+; LA32F-NEXT: st.w $a2, $sp, 16 # 4-byte Folded Spill
+; LA32F-NEXT: ld.w $s3, $a0, 32
; LA32F-NEXT: ld.w $s4, $a0, 36
; LA32F-NEXT: ld.w $s5, $a0, 24
; LA32F-NEXT: ld.w $s6, $a0, 28
@@ -756,80 +754,77 @@ define void @test_d8(ptr %P, ptr %S) nounwind {
; LA32F-NEXT: ld.w $s7, $a0, 8
; LA32F-NEXT: ld.w $s0, $a0, 12
; LA32F-NEXT: ld.w $a2, $a0, 0
-; LA32F-NEXT: ld.w $a3, $a0, 4
+; LA32F-NEXT: ld.w $a4, $a0, 4
; LA32F-NEXT: move $fp, $a1
-; LA32F-NEXT: lu12i.w $s8, 261888
+; LA32F-NEXT: lu12i.w $a3, 261888
; LA32F-NEXT: move $a0, $a2
-; LA32F-NEXT: move $a1, $a3
+; LA32F-NEXT: move $a1, $a4
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s8
; LA32F-NEXT: bl %plt(__adddf3)
; LA32F-NEXT: st.w $a0, $sp, 40 # 4-byte Folded Spill
; LA32F-NEXT: st.w $a1, $sp, 36 # 4-byte Folded Spill
-; LA32F-NEXT: lu12i.w $s3, 262144
+; LA32F-NEXT: lu12i.w $a3, 262144
; LA32F-NEXT: move $a0, $s7
; LA32F-NEXT: move $a1, $s0
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s3
+; LA32F-NEXT: move $s0, $a3
; LA32F-NEXT: bl %plt(__adddf3)
; LA32F-NEXT: st.w $a0, $sp, 24 # 4-byte Folded Spill
; LA32F-NEXT: st.w $a1, $sp, 20 # 4-byte Folded Spill
-; LA32F-NEXT: lu12i.w $s0, 262272
+; LA32F-NEXT: lu12i.w $s7, 262272
; LA32F-NEXT: move $a0, $s1
; LA32F-NEXT: move $a1, $s2
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s0
+; LA32F-NEXT: move $a3, $s7
; LA32F-NEXT: bl %plt(__adddf3)
-; LA32F-NEXT: st.w $a0, $sp, 8 # 4-byte Folded Spill
-; LA32F-NEXT: st.w $a1, $sp, 4 # 4-byte Folded Spill
-; LA32F-NEXT: lu12i.w $s7, 262400
+; LA32F-NEXT: st.w $a0, $sp, 12 # 4-byte Folded Spill
+; LA32F-NEXT: move $s2, $a1
+; LA32F-NEXT: lu12i.w $a3, 262400
; LA32F-NEXT: move $a0, $s5
; LA32F-NEXT: move $a1, $s6
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s7
; LA32F-NEXT: bl %plt(__adddf3)
; LA32F-NEXT: move $s5, $a0
; LA32F-NEXT: move $s6, $a1
-; LA32F-NEXT: ld.w $a0, $sp, 0 # 4-byte Folded Reload
+; LA32F-NEXT: move $a0, $s3
; LA32F-NEXT: move $a1, $s4
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s8
+; LA32F-NEXT: lu12i.w $a3, 261888
; LA32F-NEXT: bl %plt(__adddf3)
-; LA32F-NEXT: move $s4, $a0
-; LA32F-NEXT: move $s8, $a1
-; LA32F-NEXT: ld.w $a0, $sp, 16 # 4-byte Folded Reload
-; LA32F-NEXT: ld.w $a1, $sp, 12 # 4-byte Folded Reload
+; LA32F-NEXT: move $s3, $a0
+; LA32F-NEXT: move $s4, $a1
+; LA32F-NEXT: move $a0, $s8
+; LA32F-NEXT: ld.w $a1, $sp, 16 # 4-byte Folded Reload
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s3
+; LA32F-NEXT: move $a3, $s0
; LA32F-NEXT: bl %plt(__adddf3)
-; LA32F-NEXT: move $s3, $a0
-; LA32F-NEXT: move $s1, $a1
+; LA32F-NEXT: move $s8, $a0
+; LA32F-NEXT: move $s0, $a1
; LA32F-NEXT: ld.w $a0, $sp, 32 # 4-byte Folded Reload
; LA32F-NEXT: ld.w $a1, $sp, 28 # 4-byte Folded Reload
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s0
+; LA32F-NEXT: move $a3, $s7
; LA32F-NEXT: bl %plt(__adddf3)
-; LA32F-NEXT: move $s0, $a0
-; LA32F-NEXT: move $s2, $a1
+; LA32F-NEXT: move $s7, $a0
+; LA32F-NEXT: move $s1, $a1
; LA32F-NEXT: ld.w $a0, $sp, 48 # 4-byte Folded Reload
; LA32F-NEXT: ld.w $a1, $sp, 44 # 4-byte Folded Reload
; LA32F-NEXT: move $a2, $zero
-; LA32F-NEXT: move $a3, $s7
+; LA32F-NEXT: lu12i.w $a3, 262400
; LA32F-NEXT: bl %plt(__adddf3)
; LA32F-NEXT: st.w $a0, $fp, 56
; LA32F-NEXT: st.w $a1, $fp, 60
-; LA32F-NEXT: st.w $s0, $fp, 48
-; LA32F-NEXT: st.w $s2, $fp, 52
-; LA32F-NEXT: st.w $s3, $fp, 40
-; LA32F-NEXT: st.w $s1, $fp, 44
-; LA32F-NEXT: st.w $s4, $fp, 32
-; LA32F-NEXT: st.w $s8, $fp, 36
+; LA32F-NEXT: st.w $s7, $fp, 48
+; LA32F-NEXT: st.w $s1, $fp, 52
+; LA32F-NEXT: st.w $s8, $fp, 40
+; LA32F-NEXT: st.w $s0, $fp, 44
+; LA32F-NEXT: st.w $s3, $fp, 32
+; LA32F-NEXT: st.w $s4, $fp, 36
; LA32F-NEXT: st.w $s5, $fp, 24
; LA32F-NEXT: st.w $s6, $fp, 28
-; LA32F-NEXT: ld.w $a0, $sp, 8 # 4-byte Folded Reload
+; LA32F-NEXT: ld.w $a0, $sp, 12 # 4-byte Folded Reload
; LA32F-NEXT: st.w $a0, $fp, 16
-; LA32F-NEXT: ld.w $a0, $sp, 4 # 4-byte Folded Reload
-; LA32F-NEXT: st.w $a0, $fp, 20
+; LA32F-NEXT: st.w $s2, $fp, 20
; LA32F-NEXT: ld.w $a0, $sp, 24 # 4-byte Folded Reload
; LA32F-NEXT: st.w $a0, $fp, 8
; LA32F-NEXT: ld.w $a0, $sp, 20 # 4-byte Folded Reload
More information about the llvm-commits
mailing list