[llvm][RISCV] Support P Extension CodeGen

llvm/lib/Target/RISCV/RISCVISelLowering.cpp

@@ -525,6 +525,7 @@ RISCVTargetLowering::RISCVTargetLowering(const TargetMachine &TM,
     setOperationAction(ISD::SADDSAT, VTs, Legal);
     setOperationAction(ISD::USUBSAT, VTs, Legal);
     setOperationAction(ISD::SSUBSAT, VTs, Legal);
+    setOperationAction(ISD::SSHLSAT, VTs, Legal);
     setOperationAction({ISD::AVGFLOORS, ISD::AVGFLOORU}, VTs, Legal);
     setOperationAction({ISD::ABDS, ISD::ABDU}, VTs, Legal);
     setOperationAction(ISD::SPLAT_VECTOR, VTs, Legal);

llvm/lib/Target/RISCV/RISCVInstrInfoP.td

@@ -1520,6 +1520,17 @@ let Predicates = [HasStdExtP] in {
   def: Pat<(XLenVecI16VT (abds GPR:$rs1, GPR:$rs2)), (PDIF_H GPR:$rs1, GPR:$rs2)>;
   def: Pat<(XLenVecI16VT (abdu GPR:$rs1, GPR:$rs2)), (PDIFU_H GPR:$rs1, GPR:$rs2)>;
 
+  // 8-bit logical shift left patterns
+  def: Pat<(XLenVecI8VT (shl GPR:$rs1, (XLenVecI8VT (splat_vector uimm3:$shamt)))),
+           (PSLLI_B GPR:$rs1, uimm3:$shamt)>;
+
+  // 16-bit logical shift left patterns
+  def: Pat<(XLenVecI16VT (shl GPR:$rs1, (XLenVecI16VT (splat_vector uimm4:$shamt)))),
+           (PSLLI_H GPR:$rs1, uimm4:$shamt)>;
+
+  // 16-bit signed saturation shift left patterns
+  def: Pat<(XLenVecI16VT (sshlsat GPR:$rs1, (XLenVecI16VT (splat_vector uimm4:$shamt)))),
+           (PSSLAI_H GPR:$rs1, uimm4:$shamt)>;
 
   // 8-bit PLI SD node pattern
   def: Pat<(XLenVecI8VT (splat_vector simm8_unsigned:$imm8)), (PLI_B simm8_unsigned:$imm8)>;
@@ -1570,6 +1581,14 @@ let Predicates = [HasStdExtP, IsRV64] in {
   // splat pattern
   def: Pat<(v2i32 (splat_vector (XLenVT GPR:$rs2))), (PADD_WS (XLenVT X0), GPR:$rs2)>;
 
+  // 32-bit logical shift left patterns
+  def: Pat<(v2i32 (shl GPR:$rs1, (v2i32 (splat_vector uimm5:$shamt)))),
+           (PSLLI_W GPR:$rs1, uimm5:$shamt)>;
+
+  // 32-bit signed saturation shift left patterns
+  def: Pat<(v2i32 (sshlsat GPR:$rs1, (v2i32 (splat_vector uimm5:$shamt)))),
+           (PSSLAI_W GPR:$rs1, uimm5:$shamt)>;
+
   // Load/Store patterns
   def : StPat<store, SD, GPR, v8i8>;
   def : StPat<store, SD, GPR, v4i16>;

llvm/test/CodeGen/RISCV/rvp-ext-rv32.ll

@@ -564,6 +564,48 @@ define void @test_build_vector_i16(ptr %ret_ptr, i16 %a, i16 %b) {
   ret void
 }
 
+; Test logical shift left immediate for v2i16
+define void @test_pslli_h(ptr %ret_ptr, ptr %a_ptr) {
+; CHECK-LABEL: test_pslli_h:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lw a1, 0(a1)
+; CHECK-NEXT:    pslli.h a1, a1, 2
+; CHECK-NEXT:    sw a1, 0(a0)
+; CHECK-NEXT:    ret
+  %a = load <2 x i16>, ptr %a_ptr
+  %res = shl <2 x i16> %a, splat(i16 2)
+  store <2 x i16> %res, ptr %ret_ptr
+  ret void
+}
+
+; Test logical shift left immediate for v4i8
+define void @test_pslli_b(ptr %ret_ptr, ptr %a_ptr) {
+; CHECK-LABEL: test_pslli_b:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lw a1, 0(a1)
+; CHECK-NEXT:    pslli.b a1, a1, 2
+; CHECK-NEXT:    sw a1, 0(a0)
+; CHECK-NEXT:    ret
+  %a = load <4 x i8>, ptr %a_ptr
+  %res = shl <4 x i8> %a, splat(i8 2)
+  store <4 x i8> %res, ptr %ret_ptr
+  ret void
+}
+
+; Test arithmetic saturation shift left immediate for v2i16
+define void @test_psslai_h(ptr %ret_ptr, ptr %a_ptr) {
+; CHECK-LABEL: test_psslai_h:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    lw a1, 0(a1)
+; CHECK-NEXT:    psslai.h a1, a1, 2
+; CHECK-NEXT:    sw a1, 0(a0)
+; CHECK-NEXT:    ret
+  %a = load <2 x i16>, ptr %a_ptr
+  %res = call <2 x i16> @llvm.sshl.sat.v2i16(<2 x i16> %a, <2 x i16> splat(i16 2))
+  store <2 x i16> %res, ptr %ret_ptr
+  ret void
+}
+
 ; Intrinsic declarations
 declare <2 x i16> @llvm.sadd.sat.v2i16(<2 x i16>, <2 x i16>)
 declare <2 x i16> @llvm.uadd.sat.v2i16(<2 x i16>, <2 x i16>)

llvm/test/CodeGen/RISCV/rvp-ext-rv64.ll

@@ -738,6 +738,61 @@ define void @test_build_vector_i32(ptr %ret_ptr, i32 %a, i32 %b) {
   ret void
 }
 
+; Test logical shift left immediate for v4i16
+define void @test_pslli_h(ptr %ret_ptr, ptr %a_ptr) {
+; CHECK-LABEL: test_pslli_h:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ld a1, 0(a1)
+; CHECK-NEXT:    pslli.h a1, a1, 2
+; CHECK-NEXT:    sd a1, 0(a0)
+; CHECK-NEXT:    ret
+  %a = load <4 x i16>, ptr %a_ptr
+  %res = shl <4 x i16> %a, splat(i16 2)
+  store <4 x i16> %res, ptr %ret_ptr
+  ret void
+}
+
+; Test logical shift left immediate for v8i8
+define void @test_pslli_b(ptr %ret_ptr, ptr %a_ptr) {
+; CHECK-LABEL: test_pslli_b:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ld a1, 0(a1)
+; CHECK-NEXT:    pslli.b a1, a1, 2
+; CHECK-NEXT:    sd a1, 0(a0)
+; CHECK-NEXT:    ret
+  %a = load <8 x i8>, ptr %a_ptr
+  %res = shl <8 x i8> %a, splat(i8 2)
+  store <8 x i8> %res, ptr %ret_ptr
+  ret void
+}
+
+; Test logical shift left immediate for v2i32
+define void @test_pslli_w(ptr %ret_ptr, ptr %a_ptr) {
+; CHECK-LABEL: test_pslli_w:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ld a1, 0(a1)
+; CHECK-NEXT:    pslli.w a1, a1, 2
+; CHECK-NEXT:    sd a1, 0(a0)
+; CHECK-NEXT:    ret
+  %a = load <2 x i32>, ptr %a_ptr
+  %res = shl <2 x i32> %a, splat(i32 2)
+  store <2 x i32> %res, ptr %ret_ptr
+  ret void
+}
+
+; Test arithmetic saturation shift left immediate for v2i32
+define void @test_psslai_w(ptr %ret_ptr, ptr %a_ptr) {
+; CHECK-LABEL: test_psslai_w:
+; CHECK:       # %bb.0:
+; CHECK-NEXT:    ld a1, 0(a1)
+; CHECK-NEXT:    psslai.w a1, a1, 2
+; CHECK-NEXT:    sd a1, 0(a0)
+; CHECK-NEXT:    ret
+  %a = load <2 x i32>, ptr %a_ptr
+  %res = call <2 x i32> @llvm.sshl.sat.v2i32(<2 x i32> %a, <2 x i32> splat(i32 2))
+  store <2 x i32> %res, ptr %ret_ptr
+  ret void
+}
 ; Intrinsic declarations
 declare <4 x i16> @llvm.sadd.sat.v4i16(<4 x i16>, <4 x i16>)
 declare <4 x i16> @llvm.uadd.sat.v4i16(<4 x i16>, <4 x i16>)