fix confusion of MOVHLPS and MOVLPS

DigitalMars · Mar 23, 2021 · b894e9d · b894e9d
1 parent 3632fbc
commit b894e9d
Show file tree

Hide file tree

Showing 4 changed files with 15 additions and 5 deletions.
diff --git a/dm/src/dmc/cgreg.d b/dm/src/dmc/cgreg.d
@@ -87,7 +87,7 @@ void cgreg_init()
     {   uint sz;
         Symbol *s = globsym[i];
 
-        //printf("considering candidate '%s' for register\n",s.Sident);
+        //printf("considering candidate '%s' for register\n", s.Sident.ptr);
 
         if (s.Srange)
             s.Srange = vec_realloc(s.Srange,dfo.length);

diff --git a/dm/src/dmc/cgxmm.d b/dm/src/dmc/cgxmm.d
@@ -2,7 +2,7 @@
  * Compiler implementation of the
  * $(LINK2 http://www.dlang.org, D programming language).
  *
- * Copyright:   Copyright (C) 2011-2020 by The D Language Foundation, All Rights Reserved
+ * Copyright:   Copyright (C) 2011-2021 by The D Language Foundation, All Rights Reserved
  * Authors:     $(LINK2 http://www.digitalmars.com, Walter Bright)
  * License:     $(LINK2 http://www.boost.org/LICENSE_1_0.txt, Boost License 1.0)
  * Source:      $(LINK2 /~https://github.com/dlang/dmd/blob/master/src/dmd/backend/cgxmm.d, backend/cgxmm.d)
@@ -1180,12 +1180,21 @@ static if (0)
         codelem(cdb,op1,&retregs,false); // eval left leaf
         const reg = findreg(retregs);
 
-        if ((op2.Eoper == OPind && !op2.Ecount) || op2.Eoper == OPvar)
+        /* MOVHLPS and LODLPS have the same opcode. They are distinguished
+         * by MOVHLPS has a second operand of size 128, LODLPS has 64
+         *  https://www.felixcloutier.com/x86/movlps
+         *  https://www.felixcloutier.com/x86/movhlps
+         * MOVHLPS must be an XMM operand, LODLPS must be a memory operand
+         */
+        const isMOVHLPS = op == MOVHLPS && tysize(ty2) == 16;
+
+        if (((op2.Eoper == OPind && !op2.Ecount) || op2.Eoper == OPvar) && !isMOVHLPS)
         {
             getlvalue(cdb,&cs, op2, RMload | retregs);     // get addressing mode
         }
         else
         {
+            // load op2 into XMM register
             regm_t rretregs = XMMREGS & ~retregs;
             scodelem(cdb, op2, &rretregs, retregs, true);
             const rreg = findreg(rretregs) - XMM0;

diff --git a/dm/src/dmc/cod3.d b/dm/src/dmc/cod3.d
@@ -739,6 +739,7 @@ void cgreg_dst_regs(reg_t* dst_integer_reg, reg_t* dst_float_reg)
 
 void cgreg_set_priorities(tym_t ty, const(reg_t)** pseq, const(reg_t)** pseqmsw)
 {
+    //printf("cgreg_set_priorities %x\n", ty);
     const sz = tysize(ty);
 
     if (tyxmmreg(ty))

diff --git a/dm/src/dmc/xmm.d b/dm/src/dmc/xmm.d
@@ -2,7 +2,7 @@
  * Compiler implementation of the
  * $(LINK2 http://www.dlang.org, D programming language).
  *
- * Copyright:   Copyright (C) ?-2020 by The D Language Foundation, All Rights Reserved
+ * Copyright:   Copyright (C) ?-2021 by The D Language Foundation, All Rights Reserved
  * Authors:     $(LINK2 http://www.digitalmars.com, Walter Bright)
  * License:     $(LINK2 http://www.boost.org/LICENSE_1_0.txt, Boost License 1.0)
  * Source:      $(LINK2 /~https://github.com/dlang/dmd/blob/master/src/dmd/backend/xmm.d, backend/_xmm.d)
@@ -74,14 +74,14 @@ enum
     LODDQU   = 0xF30F6F,        // MOVDQU xmm1, xmm2/mem128  F3 0F 6F /r
     STODQU   = 0xF30F7F,        // MOVDQU xmm1/mem128, xmm2  F3 0F 7F /r
     MOVDQ2Q  = 0xF20FD6,        // MOVDQ2Q mmx, xmm          F2 0F D6 /r
-    MOVHLPS  = 0x0F12,          // MOVHLPS xmm1, xmm2        0F 12 /r
     LODHPD   = 0x660F16,        // MOVHPD xmm, mem64         66 0F 16 /r
     STOHPD   = 0x660F17,        // MOVHPD mem64, xmm         66 0F 17 /r
     LODHPS   = 0x0F16,          // MOVHPS xmm, mem64         0F 16 /r
     STOHPS   = 0x0F17,          // MOVHPS mem64, xmm         0F 17 /r
     MOVLHPS  = 0x0F16,          // MOVLHPS xmm1, xmm2        0F 16 /r
     LODLPD   = 0x660F12,        // MOVLPD xmm, mem64         66 0F 12 /r
     STOLPD   = 0x660F13,        // MOVLPD mem64, xmm         66 0F 13 /r
+    MOVHLPS  = 0x0F12,          // MOVHLPS xmm1, xmm2        0F 12 /r
     LODLPS   = 0x0F12,          // MOVLPS xmm, mem64         0F 12 /r
     STOLPS   = 0x0F13,          // MOVLPS mem64, xmm         0F 13 /r
     MOVMSKPD = 0x660F50,        // MOVMSKPD reg32, xmm 66 0F 50 /r