[PATCH, i386]: Fix PR 80799: x86-32 generates MMX without EMMS

[Uros Bizjak <ubizjak@gmail.com>]

[-- Attachment #1: Type: text/plain, Size: 904 bytes --]

Hello!

Attached patch fixes unwanted MMX reg generation on x86-32 by also
allowing target to allocate SSE registers for MMX moves. The
difference between x86-32 and x86-64 was historic, and is not needed
any more. The patch also adds one missing inter-unit move handling to
vec_concatv2di.

2017-05-18  Uros Bizjak  <ubizjak@gmail.com>

    PR target/80799
    * config/i386/mmx.md (*mov<mode>_internal): Enable
    alternatives 11, 12, 13 and 14 also for 32bit targets.
    Remove alternatives 15, 16, 17 and 18.
    * config/i386/sse.md (vec_concatv2di): Change
    alternative (!x, *y) to (x, ?!*Yn).

testsuite/ChangeLog:

2017-05-18  Uros Bizjak  <ubizjak@gmail.com>

    PR target/80799
    * g++.dg/other/i386-11.C: New test.

Patch was bootstrapped and regression tested on x86_64-linux-gnu {,-m32}.

Committed to mainline SVN, will be backported to release branches in a
couple of days.

Uros.

[-- Attachment #2: p.diff.txt --]
[-- Type: text/plain, Size: 4270 bytes --]

Index: config/i386/mmx.md
===================================================================
--- config/i386/mmx.md	(revision 248239)
+++ config/i386/mmx.md	(working copy)
@@ -78,9 +78,9 @@
 
 (define_insn "*mov<mode>_internal"
   [(set (match_operand:MMXMODE 0 "nonimmediate_operand"
-    "=r ,o ,r,r ,m ,?!y,!y,?!y,m  ,r   ,?!Ym,v,v,v,m,*x,*x,*x,m ,r ,Yi,!Ym,*Yi")
+    "=r ,o ,r,r ,m ,?!y,!y,?!y,m  ,r   ,?!Ym,v,v,v,m,r ,Yi,!Ym,*Yi")
 	(match_operand:MMXMODE 1 "vector_move_operand"
-    "rCo,rC,C,rm,rC,C  ,!y,m  ,?!y,?!Yn,r   ,C,v,m,v,C ,*x,m ,*x,Yj,r ,*Yj,!Yn"))]
+    "rCo,rC,C,rm,rC,C  ,!y,m  ,?!y,?!Yn,r   ,C,v,m,v,Yj,r ,*Yj,!Yn"))]
   "TARGET_MMX
    && !(MEM_P (operands[0]) && MEM_P (operands[1]))"
 {
@@ -146,7 +146,7 @@
   [(set (attr "isa")
      (cond [(eq_attr "alternative" "0,1")
 	      (const_string "nox64")
-	    (eq_attr "alternative" "2,3,4,9,10,11,12,13,14,19,20")
+	    (eq_attr "alternative" "2,3,4,9,10,15,16")
 	      (const_string "x64")
 	   ]
 	   (const_string "*")))
@@ -159,14 +159,14 @@
 	      (const_string "mmx")
 	    (eq_attr "alternative" "6,7,8,9,10")
 	      (const_string "mmxmov")
-	    (eq_attr "alternative" "11,15")
+	    (eq_attr "alternative" "11")
 	      (const_string "sselog1")
-	    (eq_attr "alternative" "21,22")
+	    (eq_attr "alternative" "17,18")
 	      (const_string "ssecvt")
 	   ]
 	   (const_string "ssemov")))
    (set (attr "prefix_rex")
-     (if_then_else (eq_attr "alternative" "9,10,19,20")
+     (if_then_else (eq_attr "alternative" "9,10,15,16")
        (const_string "1")
        (const_string "*")))
    (set (attr "prefix")
@@ -181,7 +181,7 @@
    (set (attr "mode")
      (cond [(eq_attr "alternative" "2")
 	      (const_string "SI")
-	    (eq_attr "alternative" "11,12,15,16")
+	    (eq_attr "alternative" "11,12")
 	      (cond [(ior (match_operand 0 "ext_sse_reg_operand")
 			  (match_operand 1 "ext_sse_reg_operand"))
 			(const_string "XI")
@@ -197,7 +197,7 @@
 		    ]
 		    (const_string "TI"))
 
-	    (and (eq_attr "alternative" "13,14,17,18")
+	    (and (eq_attr "alternative" "13,14")
 	    	 (ior (match_test "<MODE>mode == V2SFmode")
 		      (not (match_test "TARGET_SSE2"))))
 	      (const_string "V2SF")
Index: config/i386/sse.md
===================================================================
--- config/i386/sse.md	(revision 248239)
+++ config/i386/sse.md	(working copy)
@@ -13863,10 +13863,10 @@
 ;; movd instead of movq is required to handle broken assemblers.
 (define_insn "vec_concatv2di"
   [(set (match_operand:V2DI 0 "register_operand"
-	  "=Yr,*x,x ,v ,Yi,v ,!x,x,v ,x,x,v")
+	  "=Yr,*x,x ,v ,Yi,v ,x    ,x,v ,x,x,v")
 	(vec_concat:V2DI
 	  (match_operand:DI 1 "nonimmediate_operand"
-	  "  0, 0,x ,Yv,r ,vm,*y,0,Yv,0,0,v")
+	  "  0, 0,x ,Yv,r ,vm,?!*Yn,0,Yv,0,0,v")
 	  (match_operand:DI 2 "vector_move_operand"
 	  "*rm,rm,rm,rm,C ,C ,C ,x,Yv,x,m,m")))]
   "TARGET_SSE"
Index: testsuite/g++.dg/other/i386-11.C
===================================================================
--- testsuite/g++.dg/other/i386-11.C	(nonexistent)
+++ testsuite/g++.dg/other/i386-11.C	(working copy)
@@ -0,0 +1,57 @@
+// PR target/80799
+// { dg-do compile { target i?86-*-* x86_64-*-* } }
+// { dg-require-effective-target c++11 }
+// { dg-options "-O2 -msse2" }
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+
+class alignas(16) GSVector4i
+{
+public:
+    __m128i m;
+
+	explicit GSVector4i(__m128i m)
+	{
+		this->m = m;
+	}
+
+	static void storel(void* p, const GSVector4i& v)
+	{
+		_mm_storel_epi64((__m128i*)p, v.m);
+	}
+
+	static GSVector4i loadl(const void* p)
+	{
+		return GSVector4i(_mm_loadl_epi64((__m128i*)p));
+	}
+
+	bool eq(const GSVector4i& v) const
+	{
+		return _mm_movemask_epi8(_mm_cmpeq_epi32(m, v.m)) == 0xffff;
+	}
+};
+
+
+union GIFRegTRXPOS
+{
+	unsigned long long u64;
+	void operator = (const GSVector4i& v) {GSVector4i::storel(this, v);}
+	bool operator != (const union GIFRegTRXPOS& r) const {return !((GSVector4i)r).eq(*this);}
+	operator GSVector4i() const {return GSVector4i::loadl(this);}
+};
+
+extern void dummy_call();
+extern GIFRegTRXPOS TRXPOS;
+
+void GIFRegHandlerTRXPOS(const GIFRegTRXPOS&  p)
+{
+	if(p != TRXPOS)
+	{
+		dummy_call();
+	}
+
+	TRXPOS = (GSVector4i)p;
+}
+
+// { dg-final { scan-assembler-not "%mm" } }