[PowerPC] Add prefix load pattern for fpext to v2f64

[amyk]

Summary

This patch adds a prefixed load pattern involving v2f32 fpext v2f64, where we
are dealing with a value with an offset that fits into a 34-bit signed immediate.
A reduced test case is also added to patch that tests the pattern, in which the pattern
is tested in the big endian CHECKs of the newly added test.

This is a continuation of D96075.

[lei]

LGTM
thx!

llvm/lib/Target/PowerPC/PPCInstrPrefix.td

let Predicates = [PrefixInstrs] in {
   def : Pat<(atomic_load_32 PDForm:$src), (PLWZ memri34:$src)>;
   def : Pat<(atomic_load_64 PDForm:$src), (PLD memri34:$src)>;
 
   // Atomic Store
   def : Pat<(atomic_store_8 PDForm:$dst, i32:$RS), (PSTB $RS, memri34:$dst)>;
   def : Pat<(atomic_store_16 PDForm:$dst, i32:$RS), (PSTH $RS, memri34:$dst)>;
   def : Pat<(atomic_store_32 PDForm:$dst, i32:$RS), (PSTW $RS, memri34:$dst)>;
   def : Pat<(atomic_store_64 PDForm:$dst, i64:$RS), (PSTD $RS, memri34:$dst)>;
+
+  // Prefixed fpext to v2f64
+  def : Pat<(v4f32 (PPCldvsxlh PDForm:$src)),
+            (SUBREG_TO_REG (i64 1), (PLFD PDForm:$src), sub_64)>;
 }
 
 def InsertEltShift {
   dag Sub32Left0 = (EXTRACT_SUBREG $rB, sub_32);
   dag Sub32Left1 = (RLWINM (EXTRACT_SUBREG $rB, sub_32), 1, 0, 30);
   dag Sub32Left2 = (RLWINM (EXTRACT_SUBREG $rB, sub_32), 2, 0, 29);
   dag Left3 = (RLWINM8 $rB, 3, 0, 28);
 }

llvm/test/CodeGen/PowerPC/reduce_scalarization.ll

 ; CHECK-P10-BE-NEXT:    xvadddp v2, vs0, v2
 ; CHECK-P10-BE-NEXT:    blr
 entry:
   %0 = load <2 x float>, <2 x float>* @G, align 8
   %1 = fpext <2 x float> %0 to <2 x double>
   %add = fadd <2 x double> %1, %a
   ret <2 x double> %add
 }
+
+%0 = type <{ i32, i8, [1 x i8], i16, i32, i32, i8, [1 x i8], i16, i32, float, float, double, double, ppc_fp128, { float, float }, { float, float }, { double, double }, { double, double }, { ppc_fp128, ppc_fp128 }, [89856 x i8] }>
+@Glob1 = external dso_local unnamed_addr global [25 x %0], align 16
+
+define dso_local i32 @test6() #0 {
+; CHECK-P10-LABEL: test6:
+; CHECK-P10:       # %bb.0: # %bb
+; CHECK-P10-NEXT:    plfd f0, Glob1@PCREL+562536(0), 1
+; CHECK-P10-NEXT:    xxlxor vs1, vs1, vs1
+; CHECK-P10-NEXT:    xxmrghw vs0, vs0, vs0
+; CHECK-P10-NEXT:    xvcvspdp vs0, vs0
+; CHECK-P10-NEXT:    xvcmpeqdp v2, vs1, vs0
+; CHECK-P10-NEXT:    xxswapd v3, v2
+; CHECK-P10-NEXT:    xxland vs0, v2, v3
+; CHECK-P10-NEXT:    mfvsrld r3, vs0
+; CHECK-P10-NEXT:    andi. r3, r3, 1
+; CHECK-P10-NEXT:    bc 4, gt, .LBB5_2
+; CHECK-P10-NEXT:  # %bb.1: # %bb8
+; CHECK-P10-NEXT:  .LBB5_2: # %bb7
+;
+; CHECK-P10-BE-LABEL: test6:
+; CHECK-P10-BE:       # %bb.0: # %bb
+; CHECK-P10-BE-NEXT:    addis r3, r2, Glob1@toc@ha
+; CHECK-P10-BE-NEXT:    xxlxor vs1, vs1, vs1
+; CHECK-P10-BE-NEXT:    addi r3, r3, Glob1@toc@l
+; CHECK-P10-BE-NEXT:    plfd f0, 562536(r3), 0
+; CHECK-P10-BE-NEXT:    xxmrghw vs0, vs0, vs0
+; CHECK-P10-BE-NEXT:    xvcvspdp vs0, vs0
+; CHECK-P10-BE-NEXT:    xvcmpeqdp v2, vs1, vs0
+; CHECK-P10-BE-NEXT:    xxswapd v3, v2
+; CHECK-P10-BE-NEXT:    xxland vs0, v2, v3
+; CHECK-P10-BE-NEXT:    mffprd r3, f0
+; CHECK-P10-BE-NEXT:    andi. r3, r3, 1
+; CHECK-P10-BE-NEXT:    bc 4, gt, .LBB5_2
+; CHECK-P10-BE-NEXT:  # %bb.1: # %bb8
+; CHECK-P10-BE-NEXT:  .LBB5_2: # %bb7
+bb:
+  br label %bb1
+
+bb1:                                              ; preds = %bb
+  %i = load <2 x float>, <2 x float>* bitcast (i8* getelementptr inbounds ([25 x %0], [25 x %0]* @Glob1, i64 0, i64 6, i32 20, i64 22392) to <2 x float>*), align 8
+  %i2 = fpext <2 x float> %i to <2 x double>
+  %i3 = fcmp contract oeq <2 x double> zeroinitializer, %i2
+  %i4 = shufflevector <2 x i1> %i3, <2 x i1> poison, <2 x i32> <i32 1, i32 undef>
+  %i5 = and <2 x i1> %i3, %i4
+  %i6 = extractelement <2 x i1> %i5, i32 0
+  br i1 %i6, label %bb8, label %bb7
+
+bb7:                                              ; preds = %bb1
+  unreachable
+
+bb8:                                              ; preds = %bb1
+  unreachable
+}