[SelectionDAG] Use int64_t to store the integer power of llvm.powi

https://llvm.org/docs/LangRef.html#llvm-powi-intrinsic
The max length of the integer power of `llvm.powi` intrinsic is 32, and
the value can be negative. If we use `int32_t` to store this value, `-Val`
will underflow when it is `INT32_MIN`

The issue was reported in D149033.

Author: KanRobert

llvm/lib/CodeGen/SelectionDAG/SelectionDAGBuilder.cpp

@@ -5451,7 +5451,7 @@ static SDValue ExpandPowI(const SDLoc &DL, SDValue LHS, SDValue RHS,
   // it's beneficial on the target, otherwise we end up lowering to a call to
   // __powidf2 (for example).
   if (ConstantSDNode *RHSC = dyn_cast<ConstantSDNode>(RHS)) {
-    int Val = RHSC->getSExtValue();
+    int64_t Val = RHSC->getSExtValue();
 
     // powi(x, 0) -> 1.0
     if (Val == 0)

llvm/test/CodeGen/X86/powi-int32min.ll

@@ -0,0 +1,17 @@
+; RUN: llc -mtriple=x86_64-unknown-unknown < %s | FileCheck %s
+
+define float @test_powi(ptr %p) nounwind {
+; CHECK-LABEL: test_powi:
+; CHECK:       # %bb.0: # %bb
+; CHECK-NEXT:        movss {{.*#+}} xmm1 = mem[0],zero,zero,zero
+; CHECK-COUNT-31:    mulss %xmm1, %xmm1
+; CHECK-NEXT:        movss {{.*#+}} xmm0 = mem[0],zero,zero,zero
+; CHECK-NEXT:        divss %xmm1, %xmm0
+; CHECK-NEXT:        retq
+bb:
+  %load = load float, ptr %p, align 4
+  %call = call contract float @llvm.powi.f32.i32(float %load, i32 -2147483648)
+  ret float %call
+}
+
+declare float @llvm.powi.f32.i32(float, i32)