[WebAssembly] Add libcall signatures for roundeven

[tlively]

Summary

Since clang started emitting roundeven intrinsics in a7d6593a0a17, they would
cause a crash in the WebAssembly backend because it did not know the roundeven
library function signatures. Fix the crash by adding the signatures.

[tlively]

This fixes a regression reported at https://github.com/simd-everywhere/simde/issues/1001.

llvm/lib/Target/WebAssembly/WebAssemblyRuntimeLibcallSignatures.cpp

RuntimeLibcallSignatureTable() : Table(RTLIB::UNKNOWN_LIBCALL, unsupported) {
     Table[RTLIB::RINT_F64] = f64_func_f64;
     Table[RTLIB::RINT_F128] = i64_i64_func_i64_i64;
     Table[RTLIB::NEARBYINT_F32] = f32_func_f32;
     Table[RTLIB::NEARBYINT_F64] = f64_func_f64;
     Table[RTLIB::NEARBYINT_F128] = i64_i64_func_i64_i64;
     Table[RTLIB::ROUND_F32] = f32_func_f32;
     Table[RTLIB::ROUND_F64] = f64_func_f64;
     Table[RTLIB::ROUND_F128] = i64_i64_func_i64_i64;
+    Table[RTLIB::ROUNDEVEN_F32] = f32_func_f32;
+    Table[RTLIB::ROUNDEVEN_F64] = f64_func_f64;
+    Table[RTLIB::ROUNDEVEN_F128] = i64_i64_func_i64_i64;
     Table[RTLIB::LROUND_F32] = iPTR_func_f32;
     Table[RTLIB::LROUND_F64] = iPTR_func_f64;
     Table[RTLIB::LROUND_F128] = iPTR_func_i64_i64;
     Table[RTLIB::LLROUND_F32] = i64_func_f32;
     Table[RTLIB::LLROUND_F64] = i64_func_f64;
     Table[RTLIB::LLROUND_F128] = i64_func_i64_i64;
     Table[RTLIB::LRINT_F32] = iPTR_func_f32;
     Table[RTLIB::LRINT_F64] = iPTR_func_f64;

llvm/test/CodeGen/WebAssembly/libcalls.ll

 declare fp128 @llvm.powi.f128.i32(fp128, i32)
 
 declare double @llvm.cos.f64(double)
 declare double @llvm.log10.f64(double)
 declare double @llvm.pow.f64(double, double)
 declare double @llvm.powi.f64.i32(double, i32)
 declare double @llvm.log.f64(double)
 declare double @llvm.exp.f64(double)
+declare double @llvm.roundeven.f64(double)
 declare i32 @llvm.lround(double)
 
-
-
 define fp128 @fp128libcalls(fp128 %x, fp128 %y, i32 %z) {
   ; compiler-rt call
 ; CHECK-LABEL: fp128libcalls:
 ; CHECK:         .functype fp128libcalls (i32, i64, i64, i64, i64, i32) -> ()
 ; CHECK-NEXT:    .local i32
 ; CHECK-NEXT:  # %bb.0:
 ; CHECK-NEXT:    global.get $push28=, __stack_pointer
 ; CHECK-NEXT:    i32.const $push29=, 144
; CHECK-NEXT:    return
   %c = urem i128 %b, %y
   ret i128 %c
 }
 
 define i32 @f64libcalls(double %x, double %y, i32 %z) {
 ; CHECK-LABEL: f64libcalls:
 ; CHECK:         .functype f64libcalls (f64, f64, i32) -> (i32)
 ; CHECK-NEXT:  # %bb.0:
-; CHECK-NEXT:    local.get $push8=, 0
-; CHECK-NEXT:    call $push0=, cos, $pop8
+; CHECK-NEXT:    local.get $push9=, 0
+; CHECK-NEXT:    call $push0=, cos, $pop9
 ; CHECK-NEXT:    call $push1=, log10, $pop0
-; CHECK-NEXT:    local.get $push9=, 1
-; CHECK-NEXT:    call $push2=, pow, $pop1, $pop9
-; CHECK-NEXT:    local.get $push10=, 2
-; CHECK-NEXT:    call $push3=, __powidf2, $pop2, $pop10
+; CHECK-NEXT:    local.get $push10=, 1
+; CHECK-NEXT:    call $push2=, pow, $pop1, $pop10
+; CHECK-NEXT:    local.get $push11=, 2
+; CHECK-NEXT:    call $push3=, __powidf2, $pop2, $pop11
 ; CHECK-NEXT:    call $push4=, log, $pop3
 ; CHECK-NEXT:    call $push5=, exp, $pop4
 ; CHECK-NEXT:    call $push6=, cbrt, $pop5
-; CHECK-NEXT:    call $push7=, lround, $pop6
-; CHECK-NEXT:    return $pop7
+; CHECK-NEXT:    call $push7=, roundeven, $pop6
+; CHECK-NEXT:    call $push8=, lround, $pop7
+; CHECK-NEXT:    return $pop8
  %a = call double @llvm.cos.f64(double %x)
  %b = call double @llvm.log10.f64(double %a)
  %c = call double @llvm.pow.f64(double %b, double %y)
  %d = call double @llvm.powi.f64.i32(double %c, i32 %z)
  %e = call double @llvm.log.f64(double %d)
  %f = call double @llvm.exp.f64(double %e)
  %g = call fast double @llvm.pow.f64(double %f, double 0x3FD5555555555555)
- %h = call i32 @llvm.lround(double %g)
- ret i32 %h
+ %h = call double @llvm.roundeven.f64(double %g)
+ %i = call i32 @llvm.lround(double %h)
+ ret i32 %i
 }
 
 ; fcmp ord and unord (RTLIB::O_F32 / RTLIB::UO_F32 etc) are a special case (see
 ; comment in WebAssemblyRunimeLibcallSignatures.cpp) so check them separately.
 ; no libcalls are needed for f32 and f64
 
 define i1 @unordd(double %x, double %y) {
 ; CHECK-LABEL: unordd: