On 08/19/2012 11:22 AM, Richard Sandiford wrote:
Not sure whether a peephole is the right choice here. In practice,
I'd imagine these opportunities would only come from a DImode move of
$0 into a doubleword register, so we could simply emit the pattern in
mips_split_doubleword_move.
That would also allow us to use it for plain HI and LO. It wasn't
obvious from the patch why it was restricted to the DSP extension
registers.
Please also add a scan-assembler test.
How is this version of the fix?
-Sandra
2012-08-22 Sandra Loosemore <san...@codesourcery.com>
gcc/
* mips.c (mips_split_doubleword_move): Use mult instruction to
zero-initialize accumulator.
gcc/testsuite/
* gcc.target/mips/mips32-dsp-accinit.c: New.
Index: gcc/config/mips/mips.c
===================================================================
--- gcc/config/mips/mips.c (revision 190463)
+++ gcc/config/mips/mips.c (working copy)
@@ -4158,6 +4158,14 @@ mips_split_doubleword_move (rtx dest, rt
else
emit_insn (gen_mfhisi_di (mips_subword (dest, true), src));
}
+ else if (!TARGET_64BIT && !TARGET_MIPS16 && ISA_HAS_DSP_MULT
+ && src == const0_rtx
+ && REG_P (dest) && ACC_REG_P (REGNO (dest)))
+ /* Zero-initialize accumulator using "mult $dest,$0,$0" instead
+ of a mthi/mtlo pair. */
+ emit_insn (gen_mulsidi3_32bit (dest,
+ gen_rtx_REG (SImode, GP_REG_FIRST),
+ gen_rtx_REG (SImode, GP_REG_FIRST)));
else
{
/* The operation can be split into two normal moves. Decide in
Index: gcc/testsuite/gcc.target/mips/mips32-dsp-accinit.c
===================================================================
--- gcc/testsuite/gcc.target/mips/mips32-dsp-accinit.c (revision 0)
+++ gcc/testsuite/gcc.target/mips/mips32-dsp-accinit.c (revision 0)
@@ -0,0 +1,16 @@
+/* { dg-options "-O2 -march=74kc -mgp32" } */
+
+/* Check that the zero-initialization of the accumulator feeding into
+ the madd is done by means of a mult instruction instead of mthi/mtlo. */
+
+NOMIPS16 long long f (int n, int *v, int m)
+{
+ long long result = 0;
+ int i;
+
+ for (i = 0; i < n; i++)
+ result = __builtin_mips_madd (result, v[i], m);
+ return result;
+}
+
+/* { dg-final { scan-assembler "mult\t\\\$ac.,\\\$0,\\\$0" } } */
