[codegen] Use ldm instead of 2 ldrs before a TailCall

Combines 2 ldrs into a single ldm (without writeback since the
instruction uses fp as base and as a target). Shrinks the builtin
instruction size on ARM by 2932 bytes.

Change-Id: Id74e1e158a9d5db49caa2927e88df2a350adafab
Reviewed-on: https://chromium-review.googlesource.com/c/v8/v8/+/2011103Reviewed-by: Ross McIlroy <rmcilroy@chromium.org>
Commit-Queue: Dan Elphick <delphick@chromium.org>
Cr-Commit-Position: refs/heads/master@{#65882}

src/compiler/backend/arm/code-generator-arm.cc

@@ -509,8 +509,7 @@ void CodeGenerator::AssembleDeconstructFrame() {
 
 void CodeGenerator::AssemblePrepareTailCall() {
   if (frame_access_state()->has_frame()) {
-    __ ldr(lr, MemOperand(fp, StandardFrameConstants::kCallerPCOffset));
-    __ ldr(fp, MemOperand(fp, StandardFrameConstants::kCallerFPOffset));
+    __ ldm(ia, fp, lr.bit() | fp.bit());
   }
   frame_access_state()->SetFrameAccessToSP();
 }