This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

[RFC PATCH] implement asin() and acos() as built-in function

Hello!

We'd get more speedup if you could put the other opts into the .md
file also.  (Hack hack, measure, measure) it becomes a 36.5%
improvement judging by adding just inline asin and fmod to almabench.



Attached to this message is RFC patch, which implements asin() and acos() as built-in function. Patch is tested, and produces code as expected, however, I would like someone to review this patch _before_ I implement long and float versions (they are trivial, once this RFC patch is OK'd). I will prepare full patch with ChangeLog tomorrow...

BTW: I'm really curious, if it will speed up almabench and how much...

Uros. 
Index: gcc/builtins.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/builtins.c,v
retrieving revision 1.310
diff -u -p -r1.310 builtins.c
--- gcc/builtins.c	19 Apr 2004 13:03:05 -0000	1.310
+++ gcc/builtins.c	19 Apr 2004 15:47:08 -0000
@@ -1585,6 +1585,14 @@ expand_builtin_mathfn (tree exp, rtx tar
     case BUILT_IN_LOG2F:
     case BUILT_IN_LOG2L:
       errno_set = true; builtin_optab = log2_optab; break;
+    case BUILT_IN_ASIN:
+    case BUILT_IN_ASINF:
+    case BUILT_IN_ASINL:
+      builtin_optab = asin_optab; break;
+    case BUILT_IN_ACOS:
+    case BUILT_IN_ACOSF:
+    case BUILT_IN_ACOSL:
+      builtin_optab = acos_optab; break;
     case BUILT_IN_TAN:
     case BUILT_IN_TANF:
     case BUILT_IN_TANL:
@@ -5204,6 +5212,12 @@ expand_builtin (tree exp, rtx target, rt
     case BUILT_IN_TAN:
     case BUILT_IN_TANF:
     case BUILT_IN_TANL:
+    case BUILT_IN_ASIN:
+    case BUILT_IN_ASINF:
+    case BUILT_IN_ASINL:
+    case BUILT_IN_ACOS:
+    case BUILT_IN_ACOSF:
+    case BUILT_IN_ACOSL:
     case BUILT_IN_ATAN:
     case BUILT_IN_ATANF:
     case BUILT_IN_ATANL:
Index: gcc/genopinit.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/genopinit.c,v
retrieving revision 1.71
diff -u -p -r1.71 genopinit.c
--- gcc/genopinit.c	15 Apr 2004 02:43:44 -0000	1.71
+++ gcc/genopinit.c	19 Apr 2004 15:47:08 -0000
@@ -124,7 +124,9 @@ static const char * const optabs[] =
   "nearbyint_optab->handlers[$A].insn_code = CODE_FOR_$(nearbyint$a2$)",
   "sincos_optab->handlers[$A].insn_code = CODE_FOR_$(sincos$a3$)",
   "sin_optab->handlers[$A].insn_code = CODE_FOR_$(sin$a2$)",
+  "asin_optab->handlers[$A].insn_code = CODE_FOR_$(asin$a2$)",
   "cos_optab->handlers[$A].insn_code = CODE_FOR_$(cos$a2$)",
+  "acos_optab->handlers[$A].insn_code = CODE_FOR_$(acos$a2$)",
   "exp_optab->handlers[$A].insn_code = CODE_FOR_$(exp$a2$)",
   "exp10_optab->handlers[$A].insn_code = CODE_FOR_$(exp10$a2$)",
   "exp2_optab->handlers[$A].insn_code = CODE_FOR_$(exp2$a2$)",
Index: gcc/optabs.c
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/optabs.c,v
retrieving revision 1.216
diff -u -p -r1.216 optabs.c
--- gcc/optabs.c	15 Apr 2004 02:43:43 -0000	1.216
+++ gcc/optabs.c	19 Apr 2004 15:47:09 -0000
@@ -5380,7 +5380,9 @@ init_optabs (void)
   nearbyint_optab = init_optab (UNKNOWN);
   sincos_optab = init_optab (UNKNOWN);
   sin_optab = init_optab (UNKNOWN);
+  asin_optab = init_optab (UNKNOWN);
   cos_optab = init_optab (UNKNOWN);
+  acos_optab = init_optab (UNKNOWN);
   exp_optab = init_optab (UNKNOWN);
   exp10_optab = init_optab (UNKNOWN);
   exp2_optab = init_optab (UNKNOWN);
Index: gcc/optabs.h
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/optabs.h,v
retrieving revision 1.25
diff -u -p -r1.25 optabs.h
--- gcc/optabs.h	15 Apr 2004 02:43:43 -0000	1.25
+++ gcc/optabs.h	19 Apr 2004 15:47:09 -0000
@@ -152,8 +152,12 @@ enum optab_index
   OTI_sincos,
   /* Sine */
   OTI_sin,
+  /* Inverse sine */
+  OTI_asin,
   /* Cosine */
   OTI_cos,
+  /* Inverse cosine */
+  OTI_acos,
   /* Exponential */
   OTI_exp,
   /* Base-10 Exponential */
@@ -271,7 +275,9 @@ extern GTY(()) optab optab_table[OTI_MAX
 #define sqrt_optab (optab_table[OTI_sqrt])
 #define sincos_optab (optab_table[OTI_sincos])
 #define sin_optab (optab_table[OTI_sin])
+#define asin_optab (optab_table[OTI_asin])
 #define cos_optab (optab_table[OTI_cos])
+#define acos_optab (optab_table[OTI_acos])
 #define exp_optab (optab_table[OTI_exp])
 #define exp10_optab (optab_table[OTI_exp10])
 #define exp2_optab (optab_table[OTI_exp2])
Index: gcc/config/i386/i386.md
===================================================================
RCS file: /cvsroot/gcc/gcc/gcc/config/i386/i386.md,v
retrieving revision 1.528
diff -u -p -r1.528 i386.md
--- gcc/config/i386/i386.md	15 Apr 2004 02:43:44 -0000	1.528
+++ gcc/config/i386/i386.md	19 Apr 2004 15:47:13 -0000
@@ -15284,6 +15284,64 @@
   DONE;
 })
 
+;; TESTING
+
+
+(define_insn "*sqrttruncxfdf2"
+  [(set (match_operand:DF 0 "register_operand" "=f")
+	(sqrt:DF (match_operand:XF 1 "register_operand" "0")))]
+  "TARGET_80387 && !TARGET_NO_FANCY_MATH_387"
+  "fsqrt"
+  [(set_attr "type" "fpspc")
+   (set_attr "mode" "DF")
+   (set_attr "athlon_decode" "direct")])
+
+(define_expand "asindf2"
+  [(set (match_dup 2)
+	(float_extend:XF (match_operand:DF 1 "register_operand" "")))
+   (set (match_dup 3) (mult:XF (match_dup 2) (match_dup 2)))
+   (set (match_dup 5) (minus:XF (match_dup 4) (match_dup 3)))
+   (set (match_dup 6) (sqrt:DF (match_dup 5)))
+   (parallel [(set (match_operand:DF 0 "register_operand" "")
+        	   (unspec:DF [(match_dup 6) (match_dup 1)]
+			      UNSPEC_FPATAN))
+   	      (clobber (match_scratch:DF 7 ""))])]
+  "! TARGET_NO_FANCY_MATH_387 && TARGET_80387
+   && flag_unsafe_math_optimizations"
+{
+  int i;
+
+  for (i=2; i<6; i++)
+    operands[i] = gen_reg_rtx (XFmode);
+  operands[6] = gen_reg_rtx (DFmode);
+
+  emit_move_insn (operands[4], CONST1_RTX (XFmode));  /* fld1 */
+})
+
+(define_expand "acosdf2"
+  [(set (match_dup 2)
+	(float_extend:XF (match_operand:DF 1 "register_operand" "")))
+   (set (match_dup 3) (mult:XF (match_dup 2) (match_dup 2)))
+   (set (match_dup 5) (minus:XF (match_dup 4) (match_dup 3)))
+   (set (match_dup 6) (sqrt:DF (match_dup 5)))
+   (parallel [(set (match_operand:DF 0 "register_operand" "")
+        	   (unspec:DF [(match_dup 1) (match_dup 6)]
+			      UNSPEC_FPATAN))
+   	      (clobber (match_scratch:DF 7 ""))])]
+  "! TARGET_NO_FANCY_MATH_387 && TARGET_80387
+   && flag_unsafe_math_optimizations"
+{
+  int i;
+
+  for (i=2; i<6; i++)
+    operands[i] = gen_reg_rtx (XFmode);
+  operands[6] = gen_reg_rtx (DFmode);
+
+  emit_move_insn (operands[4], CONST1_RTX (XFmode));  /* fld1 */
+})
+
+;; TESTING END
+
 (define_insn "*fyl2x_sfxf3"
   [(set (match_operand:SF 0 "register_operand" "=f")
          (unspec:SF [(match_operand:SF 2 "register_operand" "0")
Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]