[RISCV] Add basic ISel patterns for Xqcilo instructions (llvm#135901)

This patch adds basic instruction selection patterns for generating the
48 bit load/store instructions that are a part of the Qualcomm uC Xqcilo
vendor extension.

Author: svs-quic

llvm/lib/Target/RISCV/RISCVInstrInfoXqci.td

@@ -159,6 +159,11 @@ def bare_simm32_lsb0 : Operand<OtherVT> {
   let OperandType = "OPERAND_PCREL";
 }
 
+def AddLike: PatFrags<(ops node:$A, node:$B),
+                      [(add node:$A, node:$B), (or node:$A, node:$B)], [{
+    return CurDAG->isBaseWithConstantOffset(SDValue(N, 0));
+}]>;
+
 //===----------------------------------------------------------------------===//
 // Instruction Formats
 //===----------------------------------------------------------------------===//
@@ -1239,6 +1244,14 @@ class PatGprNoX0Simm32NoSimm26<SDPatternOperator OpNode, RVInst48 Inst>
     : Pat<(i32 (OpNode (i32 GPRNoX0:$rs1), simm32_nosimm26:$imm)),
           (Inst GPRNoX0:$rs1, simm32_nosimm26:$imm)>;
 
+class QC48LdPat<PatFrag LoadOp, RVInst48 Inst>
+    : Pat<(i32 (LoadOp (AddLike (i32 GPR:$rs1), simm26_nosimm12:$imm26))),
+          (Inst GPR:$rs1, simm26_nosimm12:$imm26)>;
+
+class QC48StPat<PatFrag StoreOp, RVInst48 Inst>
+    : Pat<(StoreOp (i32 GPR:$rs2), (AddLike (i32 GPR:$rs1), simm26_nosimm12:$imm26)),
+          (Inst GPR:$rs2, GPR:$rs1, simm26_nosimm12:$imm26)>;
+
 /// Simple arithmetic operations
 
 let Predicates = [HasVendorXqcilia, IsRV32] in {
@@ -1253,5 +1266,19 @@ def : PatGprNoX0Simm26NoSimm12<or, QC_E_ORI>;
 def : PatGprNoX0Simm26NoSimm12<xor, QC_E_XORI>;
 } // Predicates = [HasVendorXqcilia, IsRV32]
 
+let Predicates = [HasVendorXqcilo, IsRV32], AddedComplexity = 2 in {
+  def : QC48LdPat<sextloadi8, QC_E_LB>;
+  def : QC48LdPat<extloadi8, QC_E_LBU>; // Prefer unsigned due to no c.lb in Zcb.
+  def : QC48LdPat<sextloadi16, QC_E_LH>;
+  def : QC48LdPat<extloadi16, QC_E_LH>;
+  def : QC48LdPat<load, QC_E_LW>;
+  def : QC48LdPat<zextloadi8, QC_E_LBU>;
+  def : QC48LdPat<zextloadi16, QC_E_LHU>;
+
+  def : QC48StPat<truncstorei8, QC_E_SB>;
+  def : QC48StPat<truncstorei16, QC_E_SH>;
+  def : QC48StPat<store, QC_E_SW>;
+} // Predicates = [HasVendorXqcilo, IsRV32], AddedComplexity = 2
+
 let Predicates = [HasVendorXqciint, IsRV32] in
 def : Pat<(riscv_mileaveret_glue), (QC_C_MILEAVERET)>;

llvm/test/CodeGen/RISCV/xqcilo.ll

@@ -0,0 +1,143 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc -mtriple=riscv32 -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s -check-prefixes=RV32I
+; RUN: llc -mtriple=riscv32 -mattr=+experimental-xqcilo -verify-machineinstrs < %s \
+; RUN:   | FileCheck %s -check-prefixes=RV32IXQCILO
+
+define i32 @lb_ri(i8* %a) {
+; RV32I-LABEL: lb_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a1, 2
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    lb a0, 1808(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: lb_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.lb a0, 10000(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i8, i8* %a, i32 10000
+  %2 = load i8, i8* %1
+  %3 = sext i8 %2 to i32
+  ret i32 %3
+}
+
+define i32 @lbu_ri(i8* %a) {
+; RV32I-LABEL: lbu_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a1, 1048574
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    lbu a0, 192(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: lbu_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.lbu a0, -8000(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i8, i8* %a, i32 -8000
+  %2 = load i8, i8* %1
+  %3 = zext i8 %2 to i32
+  ret i32 %3
+}
+
+define i32 @lh_ri(i16* %a) {
+; RV32I-LABEL: lh_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a1, 11
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    lhu a0, -612(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: lh_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.lhu a0, 44444(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i16, i16* %a, i32 22222
+  %2 = load i16, i16* %1
+  %3 = zext i16 %2 to i32
+  ret i32 %3
+}
+
+define i32 @lhu_ri(i16* %a) {
+; RV32I-LABEL: lhu_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a1, 1048570
+; RV32I-NEXT:    add a0, a0, a1
+; RV32I-NEXT:    lhu a0, 120(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: lhu_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.lhu a0, -24456(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i16, i16* %a, i32 -12228
+  %2 = load i16, i16* %1
+  %3 = zext i16 %2 to i32
+  ret i32 %3
+}
+
+define i32 @lw_ri(i32* %a) {
+; RV32I-LABEL: lw_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi a0, a0, 2047
+; RV32I-NEXT:    lw a0, 1953(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: lw_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.lw a0, 4000(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i32, i32* %a, i32 1000
+  %2 = load i32, i32* %1
+  ret i32 %2
+}
+
+define void @sb_ri(i8* %a, i8 %b) {
+; RV32I-LABEL: sb_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a2, 2
+; RV32I-NEXT:    add a0, a0, a2
+; RV32I-NEXT:    sb a1, 1808(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: sb_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.sb a1, 10000(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i8, i8* %a, i32 10000
+  store i8 %b, i8* %1
+  ret void
+}
+
+define void @sh_ri(i16* %a, i16 %b) {
+; RV32I-LABEL: sh_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    lui a2, 11
+; RV32I-NEXT:    add a0, a0, a2
+; RV32I-NEXT:    sh a1, -612(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: sh_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.sh a1, 44444(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i16, i16* %a, i32 22222
+  store i16 %b, i16* %1
+  ret void
+}
+
+define void @sw_ri(i32* %a, i32 %b) {
+; RV32I-LABEL: sw_ri:
+; RV32I:       # %bb.0:
+; RV32I-NEXT:    addi a0, a0, 2047
+; RV32I-NEXT:    sw a1, 1953(a0)
+; RV32I-NEXT:    ret
+;
+; RV32IXQCILO-LABEL: sw_ri:
+; RV32IXQCILO:       # %bb.0:
+; RV32IXQCILO-NEXT:    qc.e.sw a1, 4000(a0)
+; RV32IXQCILO-NEXT:    ret
+  %1 = getelementptr i32, i32* %a, i32 1000
+  store i32 %b, i32* %1
+  ret void
+}