8279508: Auto-vectorize Math.round API

src/hotspot/cpu/x86/c2_MacroAssembler_x86.cpp

@@ -4142,8 +4142,7 @@ void C2_MacroAssembler::vector_round_double_evex(XMMRegister dst, XMMRegister sr
   mov64(scratch, 4602678819172646912L);
   evpbroadcastq(xtmp1, scratch, vec_enc);
   vaddpd(xtmp1, src , xtmp1, vec_enc);
-  vrndscalepd(dst, xtmp1, 0x4, vec_enc);
-  evcvtpd2qq(dst, dst, vec_enc);
+  evcvtpd2qq(dst, xtmp1, vec_enc);
   vector_cast_double_special_cases_evex(dst, src, xtmp1, xtmp2, ktmp1, ktmp2, scratch, double_sign_flip, vec_enc);
   ldmxcsr(mxcsr_std);
 }
@@ -4158,8 +4157,7 @@ void C2_MacroAssembler::vector_round_float_evex(XMMRegister dst, XMMRegister src
   movl(scratch, 1056964608);
   evpbroadcastd(xtmp1, scratch, vec_enc);
   vaddps(xtmp1, src , xtmp1, vec_enc);
-  vrndscaleps(dst, xtmp1, 0x4, vec_enc);
-  vcvtps2dq(dst, dst, vec_enc);
+  vcvtps2dq(dst, xtmp1, vec_enc);
   vector_cast_float_special_cases_evex(dst, src, xtmp1, xtmp2, ktmp1, ktmp2, scratch, float_sign_flip, vec_enc);
   ldmxcsr(mxcsr_std);
 }
@@ -4175,8 +4173,7 @@ void C2_MacroAssembler::vector_round_float_avx(XMMRegister dst, XMMRegister src,
   movq(xtmp1, scratch);
   vpbroadcastd(xtmp1, xtmp1, vec_enc);
   vaddps(xtmp1, src , xtmp1, vec_enc);
-  vroundps(dst, xtmp1, 0x4, vec_enc);
-  vcvtps2dq(dst, dst, vec_enc);
+  vcvtps2dq(dst, xtmp1, vec_enc);
   vector_cast_float_special_cases_avx(dst, src, xtmp1, xtmp2, xtmp3, xtmp4, scratch, float_sign_flip, vec_enc);
   ldmxcsr(mxcsr_std);
 }

src/hotspot/cpu/x86/macroAssembler_x86.cpp

@@ -8954,7 +8954,7 @@ void MacroAssembler::convert_f2l(Register dst, XMMRegister src) {
 void MacroAssembler::round_float(Register dst, XMMRegister src, Register rtmp, Register rcx) {
   // Following code is exactly mimicking the functionality of java.lang.Math.round(float) method.
   Label L_special_case, L_block1, L_exit;
-  movl(rtmp, 0x7F800000);
+  movl(rtmp, 0x7f800000);
   movdl(dst, src);
   andl(dst, rtmp);
   sarl(dst, 0x17);