Patchwork [SH] PR 52479 - Remove fsca for DFmode

login
register
mail settings
Submitter Oleg Endo
Date March 20, 2012, 8:28 p.m.
Message ID <1332275308.1876.194.camel@yam-132-YW-E178-FTW>
Download mbox | patch
Permalink /patch/147834/
State New
Headers show

Comments

Oleg Endo - March 20, 2012, 8:28 p.m.
Hi,

The attached patch removes the fsca instruction support for DFmode on
SH4A when -ffast-math is enabled.

Tested against rev 185555 with 
make -k check RUNTESTFLAGS="--target_board=sh-sim
\{-m2/-ml,-m2/-mb,-m2a-single/-mb,
-m4-single/-ml,-m4-single/-mb,
-m4a-single/-ml,-m4a-single/-mb}"

and no new failures.

Cheers,
Oleg

ChangeLog:

	PR target/52479
	* config/sh/sh-protos.h (sh_fsca_df2int): Remove.
	* config/sh/sh.c (sh_fsca_df2int_rtx, sh_fsca_df2int): Remove.
	* config/sh/sh.md (sindf2, cosdf2): Remove.

testsuite/ChangeLog:

	PR target/52479
	* gcc.target/sh/sh4a-cos.c: Remove.
	* gcc.target/sh/sh4a-sin.c: Remove.
Kaz Kojima - March 21, 2012, 3:17 a.m.
Oleg Endo <oleg.endo@t-online.de> wrote:
> The attached patch removes the fsca instruction support for DFmode on
> SH4A when -ffast-math is enabled.
> 
> Tested against rev 185555 with 
> make -k check RUNTESTFLAGS="--target_board=sh-sim
> \{-m2/-ml,-m2/-mb,-m2a-single/-mb,
> -m4-single/-ml,-m4-single/-mb,
> -m4a-single/-ml,-m4a-single/-mb}"
> 
> and no new failures.

OK for trunk.

Regards,
	kaz

Patch

Index: gcc/testsuite/gcc.target/sh/sh4a-cos.c
===================================================================
--- gcc/testsuite/gcc.target/sh/sh4a-cos.c	(revision 185554)
+++ gcc/testsuite/gcc.target/sh/sh4a-cos.c	(working copy)
@@ -1,11 +0,0 @@ 
-/* Verify that we generate single-precision sine and cosine approximate
-   (fsca) in fast math mode on SH4A with FPU.  */
-/* { dg-do compile { target "sh*-*-*" } } */
-/* { dg-options "-O -ffast-math" } */
-/* { dg-skip-if "" { "sh*-*-*" } { "*" } { "-m4a" "-m4a-single" "-m4a-single-only" } }  */
-/* { dg-final { scan-assembler "fsca" } } */
-
-#include <math.h>
-
-double test(double f) { return cos(f); }
-
Index: gcc/testsuite/gcc.target/sh/sh4a-sin.c
===================================================================
--- gcc/testsuite/gcc.target/sh/sh4a-sin.c	(revision 185554)
+++ gcc/testsuite/gcc.target/sh/sh4a-sin.c	(working copy)
@@ -1,11 +0,0 @@ 
-/* Verify that we generate single-precision sine and cosine approximate
-   (fsca) in fast math mode on SH4A with FPU.  */
-/* { dg-do compile { target "sh*-*-*" } } */
-/* { dg-options "-O -ffast-math" } */
-/* { dg-skip-if "" { "sh*-*-*" } { "*" } { "-m4a" "-m4a-single" "-m4a-single-only" } }  */
-/* { dg-final { scan-assembler "fsca" } } */
-
-#include <math.h>
-
-double test(double f) { return sin(f); }
-
Index: gcc/testsuite/gcc.target/sh/sh4a-sincos.c
===================================================================
--- gcc/testsuite/gcc.target/sh/sh4a-sincos.c	(revision 185554)
+++ gcc/testsuite/gcc.target/sh/sh4a-sincos.c	(working copy)
@@ -1,12 +0,0 @@ 
-/* Verify that we generate a single single-precision sine and cosine
-   approximate (fsca) in fast math mode when a function computes both
-   sine and cosine.  */
-/* { dg-do compile { target "sh*-*-*" } } */
-/* { dg-options "-O -ffast-math" } */
-/* { dg-skip-if "" { "sh*-*-*" } { "*" } { "-m4a" "-m4a-single" "-m4a-single-only" } }  */
-/* { dg-final { scan-assembler-times "fsca" 1 } } */
-
-#include <math.h>
-
-double test(double f) { return sin(f) + cos(f); }
-
Index: gcc/config/sh/sh-protos.h
===================================================================
--- gcc/config/sh/sh-protos.h	(revision 185554)
+++ gcc/config/sh/sh-protos.h	(working copy)
@@ -39,7 +39,6 @@ 
 
 #ifdef RTX_CODE
 extern rtx sh_fsca_sf2int (void);
-extern rtx sh_fsca_df2int (void);
 extern rtx sh_fsca_int2sf (void);
 
 /* Declare functions defined in sh.c and used in templates.  */
Index: gcc/config/sh/sh.c
===================================================================
--- gcc/config/sh/sh.c	(revision 185554)
+++ gcc/config/sh/sh.c	(working copy)
@@ -11997,27 +11997,6 @@ 
   return sh_fsca_sf2int_rtx;
 }
 
-/* This function returns a constant rtx that represents pi / 2**15 in
-   DFmode.  it's used to scale DFmode angles, in radians, to a
-   fixed-point signed 16.16-bit fraction of a full circle, i.e., 2*pi
-   maps to 0x10000).  */
-
-static GTY(()) rtx sh_fsca_df2int_rtx;
-
-rtx
-sh_fsca_df2int (void)
-{
-  if (! sh_fsca_df2int_rtx)
-    {
-      REAL_VALUE_TYPE rv;
-
-      real_from_string (&rv, "10430.378350470453");
-      sh_fsca_df2int_rtx = const_double_from_real_value (rv, DFmode);
-    }
-
-  return sh_fsca_df2int_rtx;
-}
-
 /* This function returns a constant rtx that represents 2**15 / pi in
    SFmode.  it's used to scale a fixed-point signed 16.16-bit fraction
    of a full circle back to a SFmode value, i.e., 0x10000 maps to
Index: gcc/config/sh/sh.md
===================================================================
--- gcc/config/sh/sh.md	(revision 185555)
+++ gcc/config/sh/sh.md	(working copy)
@@ -10658,48 +10658,6 @@ 
   DONE;
 })
 
-(define_expand "sindf2"
-  [(set (match_operand:DF 0 "fp_arith_reg_operand" "")
-	(unspec:DF [(match_operand:DF 1 "fp_arith_reg_operand" "")]
-		   UNSPEC_FSINA))]
-  "TARGET_SH4A_FP && ! TARGET_FPU_SINGLE && flag_unsafe_math_optimizations"
-{
-  rtx scaled = gen_reg_rtx (DFmode);
-  rtx truncated = gen_reg_rtx (SImode);
-  rtx fsca = gen_reg_rtx (V2SFmode);
-  rtx scale_reg = force_reg (DFmode, sh_fsca_df2int ());
-  rtx sfresult = gen_reg_rtx (SFmode);
-
-  emit_df_insn (gen_muldf3 (scaled, operands[1], scale_reg));
-  emit_df_insn (gen_fix_truncdfsi2 (truncated, scaled));
-  emit_sf_insn (gen_fsca (fsca, truncated, sh_fsca_int2sf (),
-			  get_fpscr_rtx ()));
-  emit_move_insn (sfresult, gen_rtx_SUBREG (SFmode, fsca, 0));
-  emit_df_insn (gen_extendsfdf2 (operands[0], sfresult));
-  DONE;
-})
-
-(define_expand "cosdf2"
-  [(set (match_operand:DF 0 "fp_arith_reg_operand" "")
-	(unspec:DF [(match_operand:DF 1 "fp_arith_reg_operand" "")]
-		   UNSPEC_FCOSA))]
-  "TARGET_SH4A_FP && ! TARGET_FPU_SINGLE && flag_unsafe_math_optimizations"
-{
-  rtx scaled = gen_reg_rtx (DFmode);
-  rtx truncated = gen_reg_rtx (SImode);
-  rtx fsca = gen_reg_rtx (V2SFmode);
-  rtx scale_reg = force_reg (DFmode, sh_fsca_df2int ());
-  rtx sfresult = gen_reg_rtx (SFmode);
-
-  emit_df_insn (gen_muldf3 (scaled, operands[1], scale_reg));
-  emit_df_insn (gen_fix_truncdfsi2 (truncated, scaled));
-  emit_sf_insn (gen_fsca (fsca, truncated, sh_fsca_int2sf (),
-			  get_fpscr_rtx ()));
-  emit_move_insn (sfresult, gen_rtx_SUBREG (SFmode, fsca, 4));
-  emit_df_insn (gen_extendsfdf2 (operands[0], sfresult));
-  DONE;
-})
-
 (define_expand "abssf2"
   [(set (match_operand:SF 0 "fp_arith_reg_operand" "")
 	(abs:SF (match_operand:SF 1 "fp_arith_reg_operand" "")))]