compiler: work around an ARM backend bug in LLVM

Because of a bug in the ARM backend of LLVM, the cmpxchg instruction is
lowered using ldrexd/strexd instructions which don't exist on Cortex-M
cores. This leads to an "undefined instruction" exception at runtime.
Therefore, this patch works around this by lowering directly to a call
to the __sync_val_compare_and_swap_8 function, which is what the backend
should be doing.

For details, see: https://reviews.llvm.org/D95891

To test this patch, you can run the code on a Cortex-M3 or higher
microcontroller, for example:

    tinygo flash -target=pca10040 ./testdata/atomic.go

Before this patch, this would trigger an error. With this patch, the
behavior is correct. The error (without this patch) could look like
this:

    fatal error: undefined instruction with sp=0x200007cc pc=nil

compiler/atomic.go

@@ -1,6 +1,8 @@
 package compiler
 
 import (
+	"strings"
+
 	"golang.org/x/tools/go/ssa"
 	"tinygo.org/x/go-llvm"
 )
@@ -35,6 +37,31 @@ func (b *builder) createAtomicOp(call *ssa.CallCommon) (llvm.Value, bool) {
 		ptr := b.getValue(call.Args[0])
 		old := b.getValue(call.Args[1])
 		newVal := b.getValue(call.Args[2])
+		if strings.HasSuffix(name, "64") {
+			arch := strings.Split(b.Triple, "-")[0]
+			if strings.HasPrefix(arch, "arm") && strings.HasSuffix(arch, "m") {
+				// Work around a bug in LLVM, at least LLVM 11:
+				// https://reviews.llvm.org/D95891
+				// Check for armv6m, armv7, armv7em, and perhaps others.
+				// See also: https://gcc.gnu.org/onlinedocs/gcc/_005f_005fsync-Builtins.html
+				compareAndSwap := b.mod.NamedFunction("__sync_val_compare_and_swap_8")
+				if compareAndSwap.IsNil() {
+					// Declare the function if it isn't already declared.
+					i64Type := b.ctx.Int64Type()
+					fnType := llvm.FunctionType(i64Type, []llvm.Type{llvm.PointerType(i64Type, 0), i64Type, i64Type}, false)
+					compareAndSwap = llvm.AddFunction(b.mod, "__sync_val_compare_and_swap_8", fnType)
+				}
+				actualOldValue := b.CreateCall(compareAndSwap, []llvm.Value{ptr, old, newVal}, "")
+				// The __sync_val_compare_and_swap_8 function returns the old
+				// value. However, we shouldn't return the old value, we should
+				// return whether the compare/exchange was successful. This is
+				// easily done by comparing the returned (actual) old value with
+				// the expected old value passed to
+				// __sync_val_compare_and_swap_8.
+				swapped := b.CreateICmp(llvm.IntEQ, old, actualOldValue, "")
+				return swapped, true
+			}
+		}
 		tuple := b.CreateAtomicCmpXchg(ptr, old, newVal, llvm.AtomicOrderingSequentiallyConsistent, llvm.AtomicOrderingSequentiallyConsistent, true)
 		swapped := b.CreateExtractValue(tuple, 1, "")
 		return swapped, true