--- old/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp	2018-07-09 12:24:32.979117724 +0300
+++ new/src/hotspot/cpu/aarch64/interp_masm_aarch64.cpp	2018-07-09 12:24:32.719101199 +0300
@@ -1758,7 +1758,7 @@
       br(Assembler::EQ, do_profile);
       get_method(tmp);
       ldrh(rscratch1, Address(tmp, Method::intrinsic_id_offset_in_bytes()));
-      cmp(rscratch1, vmIntrinsics::_compiledLambdaForm);
+      cmp_imm12(rscratch1, vmIntrinsics::_compiledLambdaForm);
       br(Assembler::NE, profile_continue);
 
       bind(do_profile);
--- old/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp	2018-07-09 12:24:33.767167812 +0300
+++ new/src/hotspot/cpu/aarch64/macroAssembler_aarch64.cpp	2018-07-09 12:24:33.511151539 +0300
@@ -4357,7 +4357,7 @@
     cmp(cnt1, 8);             // Use Linear Scan if cnt1 < 8 || cnt1 >= 256
     br(LT, LINEARSEARCH);
     dup(v0, T16B, cnt1); // done in separate FPU pipeline. Almost no penalty
-    cmp(cnt1, 256);
+    cmp_imm12(cnt1, 256);
     lsr(tmp1, cnt2, 2);
     ccmp(cnt1, tmp1, 0b0000, LT); // Source must be 4 * pattern for BM
     br(GE, LINEARSTUB);
@@ -4463,7 +4463,7 @@
     BIND(BCLOOP);
       (this->*str1_load_1chr)(ch1, Address(post(tmp3, str1_chr_size)));
       if (!str1_isL) {
-        cmp(ch1, ASIZE);
+        cmp_imm12(ch1, ASIZE);
         br(HS, BCSKIP);
       }
       strb(ch2, Address(sp, ch1));
@@ -4527,7 +4527,7 @@
         } else {
           mov(result_tmp, 1);
         }
-        cmp(skipch, ASIZE);
+        cmp_imm12(skipch, ASIZE);
         br(HS, BMADV);
       }
       ldrb(result_tmp, Address(sp, skipch)); // load skip distance
--- old/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp	2018-07-09 12:24:34.455211542 +0300
+++ new/src/hotspot/cpu/aarch64/macroAssembler_aarch64.hpp	2018-07-09 12:24:34.235197559 +0300
@@ -180,8 +180,12 @@
 
   template<class T>
   inline void cmpw(Register Rd, T imm)  { subsw(zr, Rd, imm); }
-  // imm is limited to 12 bits.
-  inline void cmp(Register Rd, unsigned imm)  { subs(zr, Rd, imm); }
+
+  inline void cmp(Register Rd, unsigned char imm8)  { subs(zr, Rd, imm8); }
+  // imm is limited to 12 bits: 0<x<4096 or 4096<=x<2^24
+  inline void cmp_imm12(Register Rd, unsigned imm12)  { subs(zr, Rd, imm12); }
+
+  inline void cmp(Register Rd, Register Rn, unsigned imm32)  { subs(Rd, Rn, imm32); }
 
   inline void cmnw(Register Rd, unsigned imm) { addsw(zr, Rd, imm); }
   inline void cmn(Register Rd, unsigned imm) { adds(zr, Rd, imm); }
--- old/src/hotspot/cpu/aarch64/macroAssembler_aarch64_log.cpp	2018-07-09 12:24:35.207259341 +0300
+++ new/src/hotspot/cpu/aarch64/macroAssembler_aarch64_log.cpp	2018-07-09 12:24:34.963243832 +0300
@@ -292,7 +292,7 @@
     // vtmp1 = AS_DOUBLE_BITS(0x77F0 << 48 | mantissa(X)) == mx
     fmovd(vtmp1, tmp4);
     subw(tmp2, tmp2, 16);
-    cmp(tmp2, 0x8000);
+    cmp_imm12(tmp2, 0x8000);
     br(GE, SMALL_VALUE);
   bind(MAIN);
     fmovs(tmp3, vtmp5);                        // int intB0 = AS_INT_BITS(B);
--- old/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp	2018-07-09 12:24:35.887302564 +0300
+++ new/src/hotspot/cpu/aarch64/stubGenerator_aarch64.cpp	2018-07-09 12:24:35.627286037 +0300
@@ -3851,9 +3851,7 @@
     __ sub(cnt1, cnt1, 8 * wordSize);
     __ eor(tmp7, tmp7, tmp8);
     __ ldp(tmp2, tmp4, Address(__ post(a2, 2 * wordSize)));
-    // tmp6 is not used. MacroAssembler::subs is used here (rather than
-    // cmp) because subs allows an unlimited range of immediate operand.
-    __ subs(tmp6, cnt1, loopThreshold);
+    __ cmp(tmp6, cnt1, loopThreshold);
     __ orr(tmp5, tmp5, tmp7);
     __ cbnz(tmp5, NOT_EQUAL);
     __ br(__ GE, LOOP);
@@ -3879,7 +3877,7 @@
     __ ld1(v0, v1, v2, v3, __ T2D, Address(__ post(a1, 4 * 2 * wordSize)));
     __ sub(cnt1, cnt1, 8 * wordSize);
     __ ld1(v4, v5, v6, v7, __ T2D, Address(__ post(a2, 4 * 2 * wordSize)));
-    __ subs(tmp1, cnt1, loopThreshold);
+    __ cmp(tmp1, cnt1, loopThreshold);
     __ eor(v0, __ T16B, v0, v4);
     __ eor(v1, __ T16B, v1, v5);
     __ eor(v2, __ T16B, v2, v6);
@@ -3938,7 +3936,7 @@
     }
     if (UseSIMDForArrayEquals) {
       if (SoftwarePrefetchHintDistance >= 0) {
-        __ subs(tmp1, cnt1, prefetchLoopThreshold);
+        __ cmp(tmp1, cnt1, prefetchLoopThreshold);
         __ br(__ LE, NO_PREFETCH_LARGE_LOOP);
         generate_large_array_equals_loop_simd(prefetchLoopThreshold,
             /* prfm = */ true, NOT_EQUAL);
@@ -3951,7 +3949,7 @@
     } else {
       __ push(spilled_regs, sp);
       if (SoftwarePrefetchHintDistance >= 0) {
-        __ subs(tmp1, cnt1, prefetchLoopThreshold);
+        __ cmp(tmp1, cnt1, prefetchLoopThreshold);
         __ br(__ LE, NO_PREFETCH_LARGE_LOOP);
         generate_large_array_equals_loop_nonsimd(prefetchLoopThreshold,
             /* prfm = */ true, NOT_EQUAL);
@@ -4106,7 +4104,7 @@
     __ ldr(tmp3, Address(__ post(cnt1, 8)));
 
     if (SoftwarePrefetchHintDistance >= 0) {
-      __ cmp(cnt2, prefetchLoopExitCondition);
+      __ cmp(rscratch2, cnt2, prefetchLoopExitCondition);
       __ br(__ LT, SMALL_LOOP);
       __ bind(LARGE_LOOP_PREFETCH);
         __ prfm(Address(tmp2, SoftwarePrefetchHintDistance));
@@ -4123,7 +4121,7 @@
           __ subs(tmp4, tmp4, 1);
           __ br(__ GT, LARGE_LOOP_PREFETCH_REPEAT2);
           __ sub(cnt2, cnt2, 64);
-          __ cmp(cnt2, prefetchLoopExitCondition);
+          __ cmp(rscratch2, cnt2, prefetchLoopExitCondition); 
           __ br(__ GE, LARGE_LOOP_PREFETCH);
     }
     __ cbz(cnt2, LOAD_LAST); // no characters left except last load
@@ -4240,7 +4238,7 @@
         compare_string_16_bytes_same(DIFF, DIFF2);
         __ sub(cnt2, cnt2, isLL ? 64 : 32);
         compare_string_16_bytes_same(DIFF, DIFF2);
-        __ cmp(cnt2, largeLoopExitCondition);
+        __ cmp(rscratch2, cnt2, largeLoopExitCondition);
         compare_string_16_bytes_same(DIFF, DIFF2);
         __ br(__ GT, LARGE_LOOP_PREFETCH);
         __ cbz(cnt2, LAST_CHECK_AND_LENGTH_DIFF); // no more chars left?
@@ -4659,7 +4657,7 @@
     __ zip1(v2, __ T16B, v2, v0);
     __ st1(v1, v2, __ T16B, __ post(dst, 32));
     __ ld1(v3, v4, v5, v6, __ T16B, Address(__ post(src, 64)));
-    __ cmp(octetCounter, large_loop_threshold);
+    __ cmp(rscratch1, octetCounter, large_loop_threshold);
     __ br(__ LE, LOOP_START);
     __ b(LOOP_PRFM_START);
     __ bind(LOOP_PRFM);
@@ -4667,7 +4665,7 @@
     __ bind(LOOP_PRFM_START);
       __ prfm(Address(src, SoftwarePrefetchHintDistance));
       __ sub(octetCounter, octetCounter, 8);
-      __ cmp(octetCounter, large_loop_threshold);
+      __ cmp(rscratch1, octetCounter, large_loop_threshold);
       inflate_and_store_2_fp_registers(true, v3, v4);
       inflate_and_store_2_fp_registers(true, v5, v6);
       __ br(__ GT, LOOP_PRFM);