reg-stack.c (subst_stack_regs_pat): Handle <UNSPEC_FIST_FLOOR> and <UNSPEC_FIST_CEIL> case.

	* reg-stack.c (subst_stack_regs_pat): Handle <UNSPEC_FIST_FLOOR> and
	<UNSPEC_FIST_CEIL> case.

	* config/i386/i386.md (UNSPEC_FIST_FLOOR, UNSPEC_FIST_CEIL): New.
	(*fist<mode>2_floor_1, fistdi2_floor, fistdi2_floor_with_temp)
	(fist<mode>2_floor, fist<mode>2_floor_with_temp): New isns patterns
	to implement lfloor and llfloor built-ins as x87 intrinsic function.
	(fistdi2_floor, fist<mode>2_floor splitters): New splitters.
	(lfloor<mode>2): New expanders.
	(*fist<mode>2_ceil_1, fistdi2_ceil, fistdi2_ceil_with_temp)
	(fist<mode>2_ceil, fist<mode>2_ceil_with_temp): New isns patterns
	to implement lceil and llceil built-ins as x87 intrinsic function.
	(fistdi2_ceil, fist<mode>2_ceil splitters): New splitters.
	(lceil<mode>2): New expanders.

From-SVN: r98132

3 files changed, 337 insertions, 0 deletions

diff --git a/gcc/ChangeLog b/gcc/ChangeLog
index cdbbf33..5dc6889 100644
--- a/gcc/ChangeLog
+++ b/gcc/ChangeLog
@@ -1,5 +1,22 @@
 2005-04-14  Uros Bizjak  <uros@kss-loka.si>
 
+	* reg-stack.c (subst_stack_regs_pat): Handle <UNSPEC_FIST_FLOOR> and
+	<UNSPEC_FIST_CEIL> case.
+
+	* config/i386/i386.md (UNSPEC_FIST_FLOOR, UNSPEC_FIST_CEIL): New.
+	(*fist<mode>2_floor_1, fistdi2_floor, fistdi2_floor_with_temp)
+	(fist<mode>2_floor, fist<mode>2_floor_with_temp): New isns patterns
+	to implement lfloor and llfloor built-ins as x87 intrinsic function.
+	(fistdi2_floor, fist<mode>2_floor splitters): New splitters.
+	(lfloor<mode>2): New expanders.
+	(*fist<mode>2_ceil_1, fistdi2_ceil, fistdi2_ceil_with_temp)
+	(fist<mode>2_ceil, fist<mode>2_ceil_with_temp): New isns patterns
+	to implement lceil and llceil built-ins as x87 intrinsic function.
+	(fistdi2_ceil, fist<mode>2_ceil splitters): New splitters.
+	(lceil<mode>2): New expanders.
+
+2005-04-14  Uros Bizjak  <uros@kss-loka.si>
+
 	* convert.c (convert_to_integer): Convert (long int)trunc{,f,l},
 	and (long long int)ceil{,f,l} into FIX_TRUNC_EXPR.
 
diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md
index 098e326..e65d9c7 100644
--- a/gcc/config/i386/i386.md
+++ b/gcc/config/i386/i386.md
@@ -122,6 +122,8 @@
    (UNSPEC_FRNDINT_CEIL 	71)
    (UNSPEC_FRNDINT_TRUNC	72)
    (UNSPEC_FRNDINT_MASK_PM	73)
+   (UNSPEC_FIST_FLOOR		74)
+   (UNSPEC_FIST_CEIL 		75)
 
    ; x87 Double output FP
    (UNSPEC_SINCOS_COS		80)
@@ -16435,6 +16437,163 @@
   DONE;
 })
 
+(define_insn_and_split "*fist<mode>2_floor_1"
+  [(set (match_operand:X87MODEI 0 "nonimmediate_operand" "=m,?r")
+	(unspec:X87MODEI [(match_operand:XF 1 "register_operand" "f,f")]
+	 UNSPEC_FIST_FLOOR))
+   (clobber (reg:CC FLAGS_REG))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations
+   && !(reload_completed || reload_in_progress)"
+  "#"
+  "&& 1"
+  [(const_int 0)]
+{
+  ix86_optimize_mode_switching = 1;
+  operands[2] = assign_386_stack_local (HImode, 1);
+  operands[3] = assign_386_stack_local (HImode, 2);
+  if (memory_operand (operands[0], VOIDmode))
+    emit_insn (gen_fist<mode>2_floor (operands[0], operands[1],
+				      operands[2], operands[3]));
+  else
+    {
+      operands[4] = assign_386_stack_local (<MODE>mode, 0);
+      emit_insn (gen_fist<mode>2_floor_with_temp (operands[0], operands[1],
+						  operands[2], operands[3],
+						  operands[4]));
+    }
+  DONE;
+}
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "floor")
+   (set_attr "mode" "<MODE>")])
+
+(define_insn "fistdi2_floor"
+  [(set (match_operand:DI 0 "memory_operand" "=m")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "f")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" "m"))
+   (use (match_operand:HI 3 "memory_operand" "m"))
+   (clobber (match_scratch:XF 4 "=&1f"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "* return output_fix_trunc (insn, operands, 0);"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "floor")
+   (set_attr "mode" "DI")])
+
+(define_insn "fistdi2_floor_with_temp"
+  [(set (match_operand:DI 0 "nonimmediate_operand" "=m,?r")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "f,f")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" "m,m"))
+   (use (match_operand:HI 3 "memory_operand" "m,m"))
+   (clobber (match_operand:DI 4 "memory_operand" "=m,m"))
+   (clobber (match_scratch:XF 5 "=&1f,&1f"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "#"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "floor")
+   (set_attr "mode" "DI")])
+
+(define_split 
+  [(set (match_operand:DI 0 "register_operand" "")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:DI 4 "memory_operand" ""))
+   (clobber (match_scratch 5 ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 4) (unspec:DI [(match_dup 1)] UNSPEC_FIST_FLOOR))
+	      (use (match_dup 2))
+	      (use (match_dup 3))
+	      (clobber (match_dup 5))])
+   (set (match_dup 0) (match_dup 4))]
+  "")
+
+(define_split 
+  [(set (match_operand:DI 0 "memory_operand" "")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:DI 4 "memory_operand" ""))
+   (clobber (match_scratch 5 ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 0) (unspec:DI [(match_dup 1)] UNSPEC_FIST_FLOOR))
+	      (use (match_dup 2))
+	      (use (match_dup 3))
+	      (clobber (match_dup 5))])]
+  "")
+
+(define_insn "fist<mode>2_floor"
+  [(set (match_operand:X87MODEI12 0 "memory_operand" "=m")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "f")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" "m"))
+   (use (match_operand:HI 3 "memory_operand" "m"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "* return output_fix_trunc (insn, operands, 0);"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "floor")
+   (set_attr "mode" "<MODE>")])
+
+(define_insn "fist<mode>2_floor_with_temp"
+  [(set (match_operand:X87MODEI12 0 "nonimmediate_operand" "=m,?r")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "f,f")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" "m,m"))
+   (use (match_operand:HI 3 "memory_operand" "m,m"))
+   (clobber (match_operand:X87MODEI12 4 "memory_operand" "=m,m"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "#"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "floor")
+   (set_attr "mode" "<MODE>")])
+
+(define_split 
+  [(set (match_operand:X87MODEI12 0 "register_operand" "")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:X87MODEI12 4 "memory_operand" ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 4) (unspec:X87MODEI12 [(match_dup 1)]
+				  UNSPEC_FIST_FLOOR))
+	      (use (match_dup 2))
+	      (use (match_dup 3))])
+   (set (match_dup 0) (match_dup 4))]
+  "")
+
+(define_split 
+  [(set (match_operand:X87MODEI12 0 "memory_operand" "")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_FLOOR))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:X87MODEI12 4 "memory_operand" ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 0) (unspec:X87MODEI12 [(match_dup 1)]
+				  UNSPEC_FIST_FLOOR))
+	      (use (match_dup 2))
+	      (use (match_dup 3))])]
+  "")
+
+(define_expand "lfloor<mode>2"
+  [(parallel [(set (match_operand:X87MODEI 0 "nonimmediate_operand" "")
+		   (unspec:X87MODEI [(match_operand:XF 1 "register_operand" "")]
+		    UNSPEC_FIST_FLOOR))
+	      (clobber (reg:CC FLAGS_REG))])]
+  "TARGET_USE_FANCY_MATH_387
+   && (!TARGET_SSE_MATH || TARGET_MIX_SSE_I387)
+   && flag_unsafe_math_optimizations"
+  "")
+
 ;; Rounding mode control word calculation could clobber FLAGS_REG.
 (define_insn_and_split "frndintxf2_ceil"
   [(set (match_operand:XF 0 "register_operand" "=f")
@@ -16518,6 +16677,163 @@
   DONE;
 })
 
+(define_insn_and_split "*fist<mode>2_ceil_1"
+  [(set (match_operand:X87MODEI 0 "nonimmediate_operand" "=m,?r")
+	(unspec:X87MODEI [(match_operand:XF 1 "register_operand" "f,f")]
+	 UNSPEC_FIST_CEIL))
+   (clobber (reg:CC FLAGS_REG))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations
+   && !(reload_completed || reload_in_progress)"
+  "#"
+  "&& 1"
+  [(const_int 0)]
+{
+  ix86_optimize_mode_switching = 1;
+  operands[2] = assign_386_stack_local (HImode, 1);
+  operands[3] = assign_386_stack_local (HImode, 2);
+  if (memory_operand (operands[0], VOIDmode))
+    emit_insn (gen_fist<mode>2_ceil (operands[0], operands[1],
+				     operands[2], operands[3]));
+  else
+    {
+      operands[4] = assign_386_stack_local (<MODE>mode, 0);
+      emit_insn (gen_fist<mode>2_ceil_with_temp (operands[0], operands[1],
+						 operands[2], operands[3],
+						 operands[4]));
+    }
+  DONE;
+}
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "ceil")
+   (set_attr "mode" "<MODE>")])
+
+(define_insn "fistdi2_ceil"
+  [(set (match_operand:DI 0 "memory_operand" "=m")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "f")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" "m"))
+   (use (match_operand:HI 3 "memory_operand" "m"))
+   (clobber (match_scratch:XF 4 "=&1f"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "* return output_fix_trunc (insn, operands, 0);"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "ceil")
+   (set_attr "mode" "DI")])
+
+(define_insn "fistdi2_ceil_with_temp"
+  [(set (match_operand:DI 0 "nonimmediate_operand" "=m,?r")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "f,f")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" "m,m"))
+   (use (match_operand:HI 3 "memory_operand" "m,m"))
+   (clobber (match_operand:DI 4 "memory_operand" "=m,m"))
+   (clobber (match_scratch:XF 5 "=&1f,&1f"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "#"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "ceil")
+   (set_attr "mode" "DI")])
+
+(define_split 
+  [(set (match_operand:DI 0 "register_operand" "")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:DI 4 "memory_operand" ""))
+   (clobber (match_scratch 5 ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 4) (unspec:DI [(match_dup 1)] UNSPEC_FIST_CEIL))
+	      (use (match_dup 2))
+	      (use (match_dup 3))
+	      (clobber (match_dup 5))])
+   (set (match_dup 0) (match_dup 4))]
+  "")
+
+(define_split 
+  [(set (match_operand:DI 0 "memory_operand" "")
+	(unspec:DI [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:DI 4 "memory_operand" ""))
+   (clobber (match_scratch 5 ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 0) (unspec:DI [(match_dup 1)] UNSPEC_FIST_CEIL))
+	      (use (match_dup 2))
+	      (use (match_dup 3))
+	      (clobber (match_dup 5))])]
+  "")
+
+(define_insn "fist<mode>2_ceil"
+  [(set (match_operand:X87MODEI12 0 "memory_operand" "=m")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "f")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" "m"))
+   (use (match_operand:HI 3 "memory_operand" "m"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "* return output_fix_trunc (insn, operands, 0);"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "ceil")
+   (set_attr "mode" "<MODE>")])
+
+(define_insn "fist<mode>2_ceil_with_temp"
+  [(set (match_operand:X87MODEI12 0 "nonimmediate_operand" "=m,?r")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "f,f")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" "m,m"))
+   (use (match_operand:HI 3 "memory_operand" "m,m"))
+   (clobber (match_operand:X87MODEI12 4 "memory_operand" "=m,m"))]
+  "TARGET_USE_FANCY_MATH_387
+   && flag_unsafe_math_optimizations"
+  "#"
+  [(set_attr "type" "fistp")
+   (set_attr "i387_cw" "ceil")
+   (set_attr "mode" "<MODE>")])
+
+(define_split 
+  [(set (match_operand:X87MODEI12 0 "register_operand" "")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:X87MODEI12 4 "memory_operand" ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 4) (unspec:X87MODEI12 [(match_dup 1)]
+				  UNSPEC_FIST_CEIL))
+	      (use (match_dup 2))
+	      (use (match_dup 3))])
+   (set (match_dup 0) (match_dup 4))]
+  "")
+
+(define_split 
+  [(set (match_operand:X87MODEI12 0 "memory_operand" "")
+	(unspec:X87MODEI12 [(match_operand:XF 1 "register_operand" "")]
+	 UNSPEC_FIST_CEIL))
+   (use (match_operand:HI 2 "memory_operand" ""))
+   (use (match_operand:HI 3 "memory_operand" ""))
+   (clobber (match_operand:X87MODEI12 4 "memory_operand" ""))]
+  "reload_completed"
+  [(parallel [(set (match_dup 0) (unspec:X87MODEI12 [(match_dup 1)]
+				  UNSPEC_FIST_CEIL))
+	      (use (match_dup 2))
+	      (use (match_dup 3))])]
+  "")
+
+(define_expand "lceil<mode>2"
+  [(parallel [(set (match_operand:X87MODEI 0 "nonimmediate_operand" "")
+		   (unspec:X87MODEI [(match_operand:XF 1 "register_operand" "")]
+		    UNSPEC_FIST_CEIL))
+	      (clobber (reg:CC FLAGS_REG))])]
+  "TARGET_USE_FANCY_MATH_387
+   && (!TARGET_SSE_MATH || TARGET_MIX_SSE_I387)
+   && flag_unsafe_math_optimizations"
+  "")
+
 ;; Rounding mode control word calculation could clobber FLAGS_REG.
 (define_insn_and_split "frndintxf2_trunc"
   [(set (match_operand:XF 0 "register_operand" "=f")
diff --git a/gcc/reg-stack.c b/gcc/reg-stack.c
index 8ae4d37..a49d9e9 100644
--- a/gcc/reg-stack.c
+++ b/gcc/reg-stack.c
@@ -1671,6 +1671,10 @@ subst_stack_regs_pat (rtx insn, stack regstack, rtx pat)
 	    switch (XINT (pat_src, 1))
 	      {
 	      case UNSPEC_FIST:
+
+	      case UNSPEC_FIST_FLOOR:
+	      case UNSPEC_FIST_CEIL:
+
 		/* These insns only operate on the top of the stack.  */
 
 		src1 = get_true_reg (&XVECEXP (pat_src, 0, 0));