JIT: Remove unneeded unconditional jumps

src/coreclr/jit/codegen.h

@@ -1472,8 +1472,11 @@ XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX
 
 public:
     void instGen(instruction ins);
-
+#if defined(TARGET_XARCH)
+    void inst_JMP(emitJumpKind jmp, BasicBlock* tgtBlock, bool isRemovableJmpCandidate = false);
+#else
     void inst_JMP(emitJumpKind jmp, BasicBlock* tgtBlock);
+#endif
 
     void inst_SET(emitJumpKind condition, regNumber reg);
 

src/coreclr/jit/codegenarmarch.cpp

@@ -4301,6 +4301,18 @@ void CodeGen::inst_SETCC(GenCondition condition, var_types type, regNumber dstRe
 #endif
 }
 
+//------------------------------------------------------------------------
+// inst_JMP: Generate a jump instruction.
+//
+void CodeGen::inst_JMP(emitJumpKind jmp, BasicBlock* tgtBlock)
+{
+#if !FEATURE_FIXED_OUT_ARGS
+    assert((tgtBlock->bbTgtStkDepth * sizeof(int) == genStackLevel) || isFramePointerUsed());
+#endif // !FEATURE_FIXED_OUT_ARGS
+
+    GetEmitter()->emitIns_J(emitter::emitJumpKindToIns(jmp), tgtBlock);
+}
+
 //------------------------------------------------------------------------
 // genCodeForStoreBlk: Produce code for a GT_STORE_OBJ/GT_STORE_DYN_BLK/GT_STORE_BLK node.
 //

src/coreclr/jit/codegencommon.cpp

@@ -1944,11 +1944,12 @@ void CodeGen::genGenerateMachineCode()
 #endif // DEBUG
 
     /* We can now generate the function prolog and epilog */
-
     genGeneratePrologsAndEpilogs();
 
-    /* Bind jump distances */
+    // check to see if any jumps can be removed
+    GetEmitter()->emitRemoveJumpToNextInst();
 
+    /* Bind jump distances */
     GetEmitter()->emitJumpDistBind();
 
 #if FEATURE_LOOP_ALIGN

src/coreclr/jit/codegenlinear.cpp

@@ -753,7 +753,24 @@ void CodeGen::genCodeForBBlist()
                 break;
 
             case BBJ_ALWAYS:
-                inst_JMP(EJ_jmp, block->bbJumpDest);
+                inst_JMP(EJ_jmp, block->bbJumpDest
+#ifdef TARGET_AMD64
+                         // AMD64 requires an instruction after a call instruction for unwinding
+                         // inside an EH region so if the last instruction generated was a call instruction
+                         // do not allow this jump to be marked for possible later removal.
+                         //
+                         // If a block was selected to place an alignment instruction because it ended
+                         // with a jump, do not remove jumps from such blocks.
+                         ,
+                         /* isRemovableJmpCandidate */ !GetEmitter()->emitIsLastInsCall() && !block->hasAlign()
+#else
+#ifdef TARGET_XARCH
+                         ,
+                         /* isRemovableJmpCandidate */ !block->hasAlign()
+#endif
+
+#endif
+                             );
                 FALLTHROUGH;
 
             case BBJ_COND:

src/coreclr/jit/codegenxarch.cpp

@@ -1368,6 +1368,30 @@ void CodeGen::inst_SETCC(GenCondition condition, var_types type, regNumber dstRe
     }
 }
 
+//------------------------------------------------------------------------
+// inst_JMP: Generate a jump instruction.
+//
+void CodeGen::inst_JMP(emitJumpKind jmp, BasicBlock* tgtBlock, bool isRemovableJmpCandidate)
+{
+#if !FEATURE_FIXED_OUT_ARGS
+    // On the x86 we are pushing (and changing the stack level), but on x64 and other archs we have
+    // a fixed outgoing args area that we store into and we never change the stack level when calling methods.
+    //
+    // Thus only on x86 do we need to assert that the stack level at the target block matches the current stack level.
+    //
+    CLANG_FORMAT_COMMENT_ANCHOR;
+
+#ifdef UNIX_X86_ABI
+    // bbTgtStkDepth is a (pure) argument count (stack alignment padding should be excluded).
+    assert((tgtBlock->bbTgtStkDepth * sizeof(int) == (genStackLevel - curNestedAlignment)) || isFramePointerUsed());
+#else
+    assert((tgtBlock->bbTgtStkDepth * sizeof(int) == genStackLevel) || isFramePointerUsed());
+#endif
+#endif // !FEATURE_FIXED_OUT_ARGS
+
+    GetEmitter()->emitIns_J(emitter::emitJumpKindToIns(jmp), tgtBlock, 0, isRemovableJmpCandidate);
+}
+
 //------------------------------------------------------------------------
 // genCodeForReturnTrap: Produce code for a GT_RETURNTRAP node.
 //