i386.c (classify_argument): Pass MMX arguments in memory

	* i386.c (classify_argument): Pass MMX arguments in memory
	(ix86_expand_builtin): Expand proper address mode for cflush.
	* i386.md (movdqa): Fix typo.
	(sse2_cflush): Accept DImode addresses.

	* xmmintrin.h (_mm_sqrt_sd): Accept two arguments.
	(_mm_max_sd): Fix pasto.
	(_mm_storeh_pd, _mm_storel_pd): Fix.

	* i386.c (bdesc_comi): Fix to match specification.
	(ix86_expand_sse_comi): Emit the comparison properly.
	* i386.md (sse_comi, sse2_comi, sse_ucomi, sse2_ucomi):
	Do not use comparison operator.
	(vnmaskcmp): Fix template.

	* xmmintrin.h (_mm_cvtps_pi16): Fix.

From-SVN: r58321

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 9274dd3bd7ecdcea00e496e9944cfe470757bc82..a73f5dae512f3c039072d63c2cbf33f75e8f8229 100644 (file)
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,22 @@
+Sun Oct 20 00:31:31 CEST 2002  Jan Hubicka  <jh@suse.cz>
+
+       * i386.c (classify_argument): Pass MMX arguments in memory
+       (ix86_expand_builtin): Expand proper address mode for cflush.
+       * i386.md (movdqa): Fix typo.
+       (sse2_cflush): Accept DImode addresses.
+
+       * xmmintrin.h (_mm_sqrt_sd): Accept two arguments.
+       (_mm_max_sd): Fix pasto.
+       (_mm_storeh_pd, _mm_storel_pd): Fix.
+
+       * i386.c (bdesc_comi): Fix to match specification.
+       (ix86_expand_sse_comi): Emit the comparison properly.
+       * i386.md (sse_comi, sse2_comi, sse_ucomi, sse2_ucomi):
+       Do not use comparison operator.
+       (vnmaskcmp): Fix template.
+
+       * xmmintrin.h (_mm_cvtps_pi16): Fix.
+
 2002-10-19  Sebastian Pop  <s.pop@laposte.net>
 
        * dependence.c : Removed.

diff --git a/gcc/config/i386/i386.c b/gcc/config/i386/i386.c
index c96270241d1f0e1552b8b0dbf69207782e9fab65..11602573dd4fbdacc3ce5659473e1ebf2044064a 100644 (file)
--- a/gcc/config/i386/i386.c
+++ b/gcc/config/i386/i386.c
@@ -1922,8 +1922,7 @@ classify_argument (mode, type, classes, bit_offset)
     case V2SImode:
     case V4HImode:
     case V8QImode:
-      classes[0] = X86_64_SSE_CLASS;
-      return 1;
+      return 0;
     case BLKmode:
     case VOIDmode:
       return 0;
@@ -11755,30 +11754,30 @@ struct builtin_description
 
 static const struct builtin_description bdesc_comi[] =
 {
-  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comieq", IX86_BUILTIN_COMIEQSS, EQ, 0 },
-  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comilt", IX86_BUILTIN_COMILTSS, LT, 0 },
-  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comile", IX86_BUILTIN_COMILESS, LE, 0 },
-  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comigt", IX86_BUILTIN_COMIGTSS, LT, 1 },
-  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comige", IX86_BUILTIN_COMIGESS, LE, 1 },
-  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comineq", IX86_BUILTIN_COMINEQSS, NE, 0 },
-  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomieq", IX86_BUILTIN_UCOMIEQSS, EQ, 0 },
-  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomilt", IX86_BUILTIN_UCOMILTSS, LT, 0 },
-  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomile", IX86_BUILTIN_UCOMILESS, LE, 0 },
-  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomigt", IX86_BUILTIN_UCOMIGTSS, LT, 1 },
-  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomige", IX86_BUILTIN_UCOMIGESS, LE, 1 },
-  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomineq", IX86_BUILTIN_UCOMINEQSS, NE, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdeq", IX86_BUILTIN_COMIEQSD, EQ, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdlt", IX86_BUILTIN_COMILTSD, LT, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdle", IX86_BUILTIN_COMILESD, LE, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdgt", IX86_BUILTIN_COMIGTSD, LT, 1 },
-  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdge", IX86_BUILTIN_COMIGESD, LE, 1 },
-  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdneq", IX86_BUILTIN_COMINEQSD, NE, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdeq", IX86_BUILTIN_UCOMIEQSD, EQ, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdlt", IX86_BUILTIN_UCOMILTSD, LT, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdle", IX86_BUILTIN_UCOMILESD, LE, 0 },
-  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdgt", IX86_BUILTIN_UCOMIGTSD, LT, 1 },
-  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdge", IX86_BUILTIN_UCOMIGESD, LE, 1 },
-  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdneq", IX86_BUILTIN_UCOMINEQSD, NE, 0 },
+  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comieq", IX86_BUILTIN_COMIEQSS, UNEQ, 0 },
+  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comilt", IX86_BUILTIN_COMILTSS, UNLT, 0 },
+  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comile", IX86_BUILTIN_COMILESS, UNLE, 0 },
+  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comigt", IX86_BUILTIN_COMIGTSS, GT, 0 },
+  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comige", IX86_BUILTIN_COMIGESS, GE, 0 },
+  { MASK_SSE1, CODE_FOR_sse_comi, "__builtin_ia32_comineq", IX86_BUILTIN_COMINEQSS, LTGT, 0 },
+  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomieq", IX86_BUILTIN_UCOMIEQSS, UNEQ, 0 },
+  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomilt", IX86_BUILTIN_UCOMILTSS, UNLT, 0 },
+  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomile", IX86_BUILTIN_UCOMILESS, UNLE, 0 },
+  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomigt", IX86_BUILTIN_UCOMIGTSS, GT, 0 },
+  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomige", IX86_BUILTIN_UCOMIGESS, GE, 0 },
+  { MASK_SSE1, CODE_FOR_sse_ucomi, "__builtin_ia32_ucomineq", IX86_BUILTIN_UCOMINEQSS, LTGT, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdeq", IX86_BUILTIN_COMIEQSD, UNEQ, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdlt", IX86_BUILTIN_COMILTSD, UNLT, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdle", IX86_BUILTIN_COMILESD, UNLE, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdgt", IX86_BUILTIN_COMIGTSD, GT, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdge", IX86_BUILTIN_COMIGESD, GE, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_comi, "__builtin_ia32_comisdneq", IX86_BUILTIN_COMINEQSD, LTGT, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdeq", IX86_BUILTIN_UCOMIEQSD, UNEQ, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdlt", IX86_BUILTIN_UCOMILTSD, UNLT, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdle", IX86_BUILTIN_UCOMILESD, UNLE, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdgt", IX86_BUILTIN_UCOMIGTSD, GT, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdge", IX86_BUILTIN_UCOMIGESD, GE, 0 },
+  { MASK_SSE2, CODE_FOR_sse2_ucomi, "__builtin_ia32_ucomisdneq", IX86_BUILTIN_UCOMINEQSD, LTGT, 0 },
 };
 
 static const struct builtin_description bdesc_2arg[] =
@@ -12881,14 +12880,14 @@ ix86_expand_sse_comi (d, arglist, target)
     op1 = copy_to_mode_reg (mode1, op1);
 
   op2 = gen_rtx_fmt_ee (comparison, mode0, op0, op1);
-  pat = GEN_FCN (d->icode) (op0, op1, op2);
+  pat = GEN_FCN (d->icode) (op0, op1);
   if (! pat)
     return 0;
   emit_insn (pat);
   emit_insn (gen_rtx_SET (VOIDmode,
                          gen_rtx_STRICT_LOW_PART (VOIDmode, target),
                          gen_rtx_fmt_ee (comparison, QImode,
-                                         gen_rtx_REG (CCmode, FLAGS_REG),
+                                         SET_DEST (pat),
                                          const0_rtx)));
 
   return SUBREG_REG (target);
@@ -13349,9 +13348,8 @@ ix86_expand_builtin (exp, target, subtarget, mode, ignore)
        arg0 = TREE_VALUE (arglist);
        op0 = expand_expr (arg0, NULL_RTX, VOIDmode, 0);
        icode = CODE_FOR_sse2_clflush;
-       mode0 = insn_data[icode].operand[0].mode;
-       if (! (*insn_data[icode].operand[0].predicate) (op0, mode0))
-           op0 = copy_to_mode_reg (mode0, op0);
+       if (! (*insn_data[icode].operand[0].predicate) (op0, Pmode))
+           op0 = copy_to_mode_reg (Pmode, op0);
 
        emit_insn (gen_sse2_clflush (op0));
        return 0;

diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 095de745e40f7b8b36c27889dcfc2db69db38669..6825149b1e6f6c68b0619cef01bd425b85afd074 100644 (file)
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -17828,7 +17828,7 @@
        (match_operand:V2DI 1 "nonimmediate_operand" "xm,x"))]
   "TARGET_SSE"
   ;; @@@ let's try to use movaps here.
-  "movdga\t{%1, %0|%0, %1}"
+  "movdqa\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssemov")
    (set_attr "mode" "V4SF")])
 
@@ -18883,13 +18883,12 @@
 
 (define_insn "sse_comi"
   [(set (reg:CCFP 17)
-        (match_operator:CCFP 2 "sse_comparison_operator"
-                       [(vec_select:SF
-                         (match_operand:V4SF 0 "register_operand" "x")
-                         (parallel [(const_int 0)]))
-                        (vec_select:SF
-                         (match_operand:V4SF 1 "register_operand" "x")
-                         (parallel [(const_int 0)]))]))]
+        (compare:CCFP (vec_select:SF
+                      (match_operand:V4SF 0 "register_operand" "x")
+                      (parallel [(const_int 0)]))
+                     (vec_select:SF
+                      (match_operand:V4SF 1 "register_operand" "x")
+                      (parallel [(const_int 0)]))))]
   "TARGET_SSE"
   "comiss\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssecmp")
@@ -18897,13 +18896,12 @@
 
 (define_insn "sse_ucomi"
   [(set (reg:CCFPU 17)
-       (match_operator:CCFPU 2 "sse_comparison_operator"
-                       [(vec_select:SF
-                         (match_operand:V4SF 0 "register_operand" "x")
-                         (parallel [(const_int 0)]))
-                        (vec_select:SF
-                         (match_operand:V4SF 1 "register_operand" "x")
-                         (parallel [(const_int 0)]))]))]
+       (compare:CCFPU (vec_select:SF
+                       (match_operand:V4SF 0 "register_operand" "x")
+                       (parallel [(const_int 0)]))
+                      (vec_select:SF
+                       (match_operand:V4SF 1 "register_operand" "x")
+                       (parallel [(const_int 0)]))))]
   "TARGET_SSE"
   "ucomiss\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssecmp")
@@ -20409,7 +20407,12 @@
                              [(match_operand:V2DF 1 "register_operand" "0")
                               (match_operand:V2DF 2 "nonimmediate_operand" "x")])))]
   "TARGET_SSE2"
-  "cmpn%D3pd\t{%2, %0|%0, %2}"
+{
+  if (GET_CODE (operands[3]) == UNORDERED)
+    return "cmpordps\t{%2, %0|%0, %2}";
+  else
+    return "cmpn%D3pd\t{%2, %0|%0, %2}";
+}
   [(set_attr "type" "ssecmp")
    (set_attr "mode" "V2DF")])
 
@@ -20436,19 +20439,23 @@
         (subreg:V2DI (match_dup 1) 0)
         (const_int 1)))]
   "TARGET_SSE2"
-  "cmp%D3sd\t{%2, %0|%0, %2}"
+{
+  if (GET_CODE (operands[3]) == UNORDERED)
+    return "cmpordsd\t{%2, %0|%0, %2}";
+  else
+    return "cmpn%D3sd\t{%2, %0|%0, %2}";
+}
   [(set_attr "type" "ssecmp")
    (set_attr "mode" "DF")])
 
 (define_insn "sse2_comi"
   [(set (reg:CCFP 17)
-        (match_operator:CCFP 2 "sse_comparison_operator"
-                       [(vec_select:DF
-                         (match_operand:V2DF 0 "register_operand" "x")
-                         (parallel [(const_int 0)]))
-                        (vec_select:DF
-                         (match_operand:V2DF 1 "register_operand" "x")
-                         (parallel [(const_int 0)]))]))]
+        (compare:CCFP (vec_select:DF
+                      (match_operand:V2DF 0 "register_operand" "x")
+                      (parallel [(const_int 0)]))
+                     (vec_select:DF
+                      (match_operand:V2DF 1 "register_operand" "x")
+                      (parallel [(const_int 0)]))))]
   "TARGET_SSE2"
   "comisd\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssecmp")
@@ -20456,13 +20463,12 @@
 
 (define_insn "sse2_ucomi"
   [(set (reg:CCFPU 17)
-       (match_operator:CCFPU 2 "sse_comparison_operator"
-                       [(vec_select:DF
-                         (match_operand:V2DF 0 "register_operand" "x")
-                         (parallel [(const_int 0)]))
-                        (vec_select:DF
-                         (match_operand:V2DF 1 "register_operand" "x")
-                         (parallel [(const_int 0)]))]))]
+       (compare:CCFPU (vec_select:DF
+                        (match_operand:V2DF 0 "register_operand" "x")
+                        (parallel [(const_int 0)]))
+                       (vec_select:DF
+                        (match_operand:V2DF 1 "register_operand" "x")
+                        (parallel [(const_int 0)]))))]
   "TARGET_SSE2"
   "ucomisd\t{%1, %0|%0, %1}"
   [(set_attr "type" "ssecmp")
@@ -21630,7 +21636,7 @@
    (set_attr "mode" "V2DF")])
 
 (define_insn "sse2_clflush"
-  [(unspec_volatile [(match_operand:SI 0 "address_operand" "p")]
+  [(unspec_volatile [(match_operand 0 "address_operand" "p")]
                    UNSPECV_CLFLUSH)]
   "TARGET_SSE2"
   "clflush %0"

diff --git a/gcc/config/i386/xmmintrin.h b/gcc/config/i386/xmmintrin.h
index ed464430c900ad881bc61fbb77120bf5e49bb0c6..42278b7d2e56c5d4fe66d7c3d02d3b6666ddf893 100644 (file)
--- a/gcc/config/i386/xmmintrin.h
+++ b/gcc/config/i386/xmmintrin.h
@@ -606,7 +606,7 @@ _mm_cvtps_pi16(__m128 __A)
   __v4sf __losf = __builtin_ia32_movhlps (__hisf, __hisf);
   __v2si __hisi = __builtin_ia32_cvtps2pi (__hisf);
   __v2si __losi = __builtin_ia32_cvtps2pi (__losf);
-  return (__m64) __builtin_ia32_packssdw (__losi, __hisi);
+  return (__m64) __builtin_ia32_packssdw (__hisi, __losi);
 }
 
 /* Convert the four SPFP values in A to four signed 8-bit integers.  */
@@ -1293,10 +1293,12 @@ _mm_sqrt_pd (__m128d __A)
   return (__m128d)__builtin_ia32_sqrtpd ((__v2df)__A);
 }
 
+/* Return pair {sqrt (A[0), B[1]}.  */
 static __inline __m128d
-_mm_sqrt_sd (__m128d __A)
+_mm_sqrt_sd (__m128d __A, __m128d __B)
 {
-  return (__m128d)__builtin_ia32_sqrtsd ((__v2df)__A);
+  __v2df __tmp = __builtin_ia32_movsd ((__v2df)__A, (__v2df)__B);
+  return (__m128d)__builtin_ia32_sqrtsd ((__v2df)__tmp);
 }
 
 static __inline __m128d
@@ -1320,7 +1322,7 @@ _mm_max_pd (__m128d __A, __m128d __B)
 static __inline __m128d
 _mm_max_sd (__m128d __A, __m128d __B)
 {
-  return (__m128d)__builtin_ia32_minsd ((__v2df)__A, (__v2df)__B);
+  return (__m128d)__builtin_ia32_maxsd ((__v2df)__A, (__v2df)__B);
 }
 
 static __inline __m128d
@@ -1696,10 +1698,9 @@ _mm_loadh_pd (__m128d __A, double *__B)
 }
 
 static __inline void
-_mm_storeh_pd (__m128d *__A, double __B)
+_mm_storeh_pd (double *__A, __m128d __B)
 {
-  __v2df __tmp = __builtin_ia32_loadsd (&__B);
-  __builtin_ia32_storehpd ((__v2si *)__A, __tmp);
+  __builtin_ia32_storehpd ((__v2si *)__A, (__v2df)__B);
 }
 
 static __inline __m128d
@@ -1709,10 +1710,9 @@ _mm_loadl_pd (__m128d __A, double *__B)
 }
 
 static __inline void
-_mm_storel_pd (__m128d *__A, double __B)
+_mm_storel_pd (double *__A, __m128d __B)
 {
-  __v2df __tmp = __builtin_ia32_loadsd (&__B);
-  __builtin_ia32_storelpd ((__v2si *)__A, __tmp);
+  __builtin_ia32_storelpd ((__v2si *)__A, (__v2df)__B);
 }
 
 static __inline int