i386.md (abs/neg splitter): Fix calculation of sign bit for TFmodes

	* i386.md (abs/neg splitter): Fix calculation of sign bit for TFmodes
	(pushqi2_rex64, pushhi2_rex64): Add.

From-SVN: r41599

2 files changed, 182 insertions, 7 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index 57205d1..05ebf9f 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,3 +1,8 @@
+Thu Apr 26 19:20:28 CEST 2001  Jan Hubicka  <jh@suse.cz>
+
+	* i386.md (abs/neg splitter): Fix calculation of sign bit for TFmodes
+	(pushqi2_rex64, pushhi2_rex64): Add.
+
 2001-04-26  Andrew Haley  <aph@cambridge.redhat.com>
 
 	* except.c (expand_eh_region_end_cleanup): Force pending stack
diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 3991e87..2233c14 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -1826,6 +1826,15 @@
   [(set_attr "type" "push")
    (set_attr "mode" "HI")])
 
+;; For 64BIT abi we always round up to 8 bytes.
+(define_insn "*pushhi2_rex64"
+  [(set (match_operand:HI 0 "push_operand" "=X")
+	(match_operand:HI 1 "nonmemory_no_elim_operand" "ri"))]
+  "TARGET_64BIT"
+  "push{q}\\t%q1"
+  [(set_attr "type" "push")
+   (set_attr "mode" "QI")])
+
 (define_insn "*movhi_1"
   [(set (match_operand:HI 0 "nonimmediate_operand" "=*a,r,r,*a,r,m")
 	(match_operand:HI 1 "general_operand" "i,r,rn,rm,rm,rn"))]
@@ -1985,6 +1994,15 @@
   [(set_attr "type" "push")
    (set_attr "mode" "HI")])
 
+;; For 64BIT abi we always round up to 8 bytes.
+(define_insn "*pushqi2_rex64"
+  [(set (match_operand:QI 0 "push_operand" "=X")
+	(match_operand:QI 1 "nonmemory_no_elim_operand" "ri"))]
+  "TARGET_64BIT"
+  "push{q}\\t%q1"
+  [(set_attr "type" "push")
+   (set_attr "mode" "QI")])
+
 ;; Situation is quite tricky about when to choose full sized (SImode) move
 ;; over QImode moves.  For Q_REG -> Q_REG move we use full size only for
 ;; partial register dependency machines (such as AMD Athlon), where QImode
@@ -9308,7 +9326,82 @@
 		   (neg:SF (match_operand:SF 1 "nonimmediate_operand" "")))
 	      (clobber (reg:CC 17))])]
   "TARGET_80387"
-  "ix86_expand_unary_operator (NEG, SFmode, operands); DONE;")
+  "if (TARGET_SSE)
+     {
+       /* In case operand is in memory,  we will not use SSE.  */
+       if (memory_operand (operands[0], VOIDmode)
+	   && rtx_equal_p (operands[0], operands[1]))
+	 emit_insn (gen_negsf2_memory (operands[0], operands[1]));
+       else
+	{
+	  /* Using SSE is tricky, since we need bitwise negation of -0
+	     in register.  */
+	  rtx reg = gen_reg_rtx (SFmode);
+	  emit_move_insn (reg,
+			  gen_lowpart (SFmode,
+				       trunc_int_for_mode (0x80000000,
+							   SImode)));
+	  emit_insn (gen_negsf2_ifs (operands[0], operands[1], reg));
+	}
+       DONE;
+     }
+   ix86_expand_unary_operator (NEG, SFmode, operands); DONE;")
+
+(define_insn "negsf2_memory"
+  [(set (match_operand:SF 0 "memory_operand" "=m")
+	(neg:SF (match_operand:SF 1 "memory_operand" "0")))
+   (clobber (reg:CC 17))]
+  "ix86_unary_operator_ok (NEG, SFmode, operands)"
+  "#")
+
+(define_insn "negsf2_ifs"
+  [(set (match_operand:SF 0 "nonimmediate_operand" "=x#fr,x#fr,f#xr,r#xf")
+	(neg:SF (match_operand:SF 1 "nonimmediate_operand" "0,x#fr,0,0")))
+   (use (match_operand:SF 2 "nonmemory_operand" "x#x,0#x,*X#x,*X#x"))
+   (clobber (reg:CC 17))]
+  "TARGET_SSE"
+  "#")
+
+(define_split
+  [(set (match_operand:SF 0 "memory_operand" "")
+	(neg:SF (match_operand:SF 1 "memory_operand" "")))
+   (use (match_operand:SF 2 "" ""))
+   (clobber (reg:CC 17))]
+  ""
+  [(parallel [(set (match_dup 0)
+		   (neg:SF (match_dup 1)))
+	      (clobber (reg:CC 17))])])
+
+(define_split
+  [(set (match_operand:SF 0 "register_operand" "")
+	(neg:SF (match_operand:SF 1 "register_operand" "")))
+   (use (match_operand:SF 2 "" ""))
+   (clobber (reg:CC 17))]
+  "reload_completed && !SSE_REG_P (operands[0])"
+  [(parallel [(set (match_dup 0)
+		   (neg:SF (match_dup 1)))
+	      (clobber (reg:CC 17))])])
+
+(define_split
+  [(set (match_operand:SF 0 "register_operand" "")
+	(neg:SF (match_operand:SF 1 "register_operand" "")))
+   (use (match_operand:SF 2 "register_operand" ""))
+   (clobber (reg:CC 17))]
+  "reload_completed && SSE_REG_P (operands[0])"
+  [(set (subreg:TI (match_dup 0) 0)
+	(xor:TI (subreg:TI (match_dup 1) 0)
+		(subreg:TI (match_dup 2) 0)))]
+  "
+{
+  if (operands_match_p (operands[0], operands[2]))
+    {
+      rtx tmp;
+      tmp = operands[1];
+      operands[1] = operands[2];
+      operands[2] = tmp;
+    }
+}")
+
 
 ;; Keep 'f' and 'r' in separate alternatives to avoid reload problems
 ;; because of secondary memory needed to reload from class FLOAT_INT_REGS
@@ -9317,7 +9410,8 @@
   [(set (match_operand:SF 0 "nonimmediate_operand" "=f#r,rm#f")
 	(neg:SF (match_operand:SF 1 "nonimmediate_operand" "0,0")))
    (clobber (reg:CC 17))]
-  "TARGET_80387 && ix86_unary_operator_ok (NEG, SFmode, operands)"
+  "TARGET_80387 && !TARGET_SSE
+   && ix86_unary_operator_ok (NEG, SFmode, operands)"
   "#")
 
 (define_split
@@ -9350,8 +9444,8 @@
 {
   int size = GET_MODE_SIZE (GET_MODE (operands[1]));
 
-  /* XFmode's size is 12, but only 10 bytes are used.  */
-  if (size == 12)
+  /* XFmode's size is 12, TFmode 16, but only 10 bytes are used.  */
+  if (size >= 12)
     size = 10;
   operands[0] = gen_rtx_MEM (QImode, XEXP (operands[0], 0));
   operands[0] = adj_offsettable_operand (operands[0], size - 1);
@@ -9363,7 +9457,83 @@
 		   (neg:DF (match_operand:DF 1 "nonimmediate_operand" "")))
 	      (clobber (reg:CC 17))])]
   "TARGET_80387"
-  "ix86_expand_unary_operator (NEG, DFmode, operands); DONE;")
+  "if (TARGET_SSE)
+     {
+       /* In case operand is in memory,  we will not use SSE.  */
+       if (memory_operand (operands[0], VOIDmode)
+	   && rtx_equal_p (operands[0], operands[1]))
+	 emit_insn (gen_negdf2_memory (operands[0], operands[1]));
+       else
+	{
+	  /* Using SSE is tricky, since we need bitwise negation of -0
+	     in register.  */
+	  rtx reg = gen_reg_rtx (DFmode);
+#if HOST_BITS_PER_WIDE_INT >= 64
+	  rtx imm = GEN_INT (0x80000000);
+#else
+	  rtx imm = immed_double_const (0, 0x80000000, DImode);
+#endif
+	  emit_move_insn (reg, gen_lowpart (DFmode, imm));
+	  emit_insn (gen_negdf2_ifs (operands[0], operands[1], reg));
+	}
+       DONE;
+     }
+   ix86_expand_unary_operator (NEG, DFmode, operands); DONE;")
+
+(define_insn "negdf2_memory"
+  [(set (match_operand:DF 0 "memory_operand" "=m")
+	(neg:DF (match_operand:DF 1 "memory_operand" "0")))
+   (clobber (reg:CC 17))]
+  "ix86_unary_operator_ok (NEG, DFmode, operands)"
+  "#")
+
+(define_insn "negdf2_ifs"
+  [(set (match_operand:DF 0 "nonimmediate_operand" "=x#fr,x#fr,f#xr,r#xf")
+	(neg:DF (match_operand:DF 1 "nonimmediate_operand" "0,x#fr,0,0")))
+   (use (match_operand:DF 2 "nonmemory_operand" "x#x,0#x,*X#x,*X#x"))
+   (clobber (reg:CC 17))]
+  "TARGET_SSE"
+  "#")
+
+(define_split
+  [(set (match_operand:DF 0 "memory_operand" "")
+	(neg:DF (match_operand:DF 1 "memory_operand" "")))
+   (use (match_operand:DF 2 "" ""))
+   (clobber (reg:CC 17))]
+  ""
+  [(parallel [(set (match_dup 0)
+		   (neg:DF (match_dup 1)))
+	      (clobber (reg:CC 17))])])
+
+(define_split
+  [(set (match_operand:DF 0 "register_operand" "")
+	(neg:DF (match_operand:DF 1 "register_operand" "")))
+   (use (match_operand:DF 2 "" ""))
+   (clobber (reg:CC 17))]
+  "reload_completed && !SSE_REG_P (operands[0])"
+  [(parallel [(set (match_dup 0)
+		   (neg:DF (match_dup 1)))
+	      (clobber (reg:CC 17))])])
+
+(define_split
+  [(set (match_operand:DF 0 "register_operand" "")
+	(neg:DF (match_operand:DF 1 "register_operand" "")))
+   (use (match_operand:DF 2 "register_operand" ""))
+   (clobber (reg:CC 17))]
+  "reload_completed && SSE_REG_P (operands[0])"
+  [(set (subreg:TI (match_dup 0) 0)
+	(xor:TI (subreg:TI (match_dup 1) 0)
+		(subreg:TI (match_dup 2) 0)))]
+  "
+{
+  if (operands_match_p (operands[0], operands[2]))
+    {
+      rtx tmp;
+      tmp = operands[1];
+      operands[1] = operands[2];
+      operands[2] = tmp;
+    }
+}")
 
 ;; Keep 'f' and 'r' in separate alternatives to avoid reload problems
 ;; because of secondary memory needed to reload from class FLOAT_INT_REGS
@@ -9666,8 +9836,8 @@
 {
   int size = GET_MODE_SIZE (GET_MODE (operands[1]));
 
-  /* XFmode's size is 12, but only 10 bytes are used.  */
-  if (size == 12)
+  /* XFmode's size is 12, TFmode 16, but only 10 bytes are used.  */
+  if (size >= 12)
     size = 10;
   operands[0] = gen_rtx_MEM (QImode, XEXP (operands[0], 0));
   operands[0] = adj_offsettable_operand (operands[0], size - 1);