Unnecessary moves around dpadd and dpsub are caused by different pseudos
being assigned to the input-output operands which correspond to the same
register.
This forces the same pseudo to the input-output operands, which removes
unnecesary moves.
Tested on mips-mti-linux-gnu.
gcc/ChangeLog:
* gcc/config/mips/mips.c (mips_expand_builtin_insn): Force the
operands which correspond to the same input-output register to
have the same pseudo assigned to them.
gcc/testsuite/ChangeLog:
* gcc/testsuite/gcc.target/mips/msa-dpadd-dpsub.c: New test.
---
gcc/config/mips/mips.c | 20 ++++++++++++++++++
gcc/testsuite/gcc.target/mips/msa-dpadd-dpsub.c | 28 +++++++++++++++++++++++++
2 files changed, 48 insertions(+)
create mode 100644 gcc/testsuite/gcc.target/mips/msa-dpadd-dpsub.c
diff --git a/gcc/config/mips/mips.c b/gcc/config/mips/mips.c
index 7f6a0db..3a77097 100644
--- a/gcc/config/mips/mips.c
+++ b/gcc/config/mips/mips.c
@@ -16960,6 +16960,26 @@ mips_expand_builtin_insn (enum insn_code icode,
unsigned int nops,
std::swap (ops[1], ops[3]);
break;
+ case CODE_FOR_msa_dpadd_s_w:
+ case CODE_FOR_msa_dpadd_s_h:
+ case CODE_FOR_msa_dpadd_s_d:
+ case CODE_FOR_msa_dpadd_u_w:
+ case CODE_FOR_msa_dpadd_u_h:
+ case CODE_FOR_msa_dpadd_u_d:
+ case CODE_FOR_msa_dpsub_s_w:
+ case CODE_FOR_msa_dpsub_s_h:
+ case CODE_FOR_msa_dpsub_s_d:
+ case CODE_FOR_msa_dpsub_u_w:
+ case CODE_FOR_msa_dpsub_u_h:
+ case CODE_FOR_msa_dpsub_u_d:
+ /* Force the operands which correspond to the same in-out register
+ to have the same pseudo assigned to them. If the input operand
+ is not REG, create one for it. */
+ if (!REG_P (ops[1].value))
+ ops[1].value = copy_to_mode_reg (ops[1].mode, ops[1].value);
+ create_output_operand (&ops[0], ops[1].value, ops[1].mode);
+ break;
+
default:
break;
}
diff --git a/gcc/testsuite/gcc.target/mips/msa-dpadd-dpsub.c
b/gcc/testsuite/gcc.target/mips/msa-dpadd-dpsub.c
new file mode 100644
index 0000000..c665bdf
--- /dev/null
+++ b/gcc/testsuite/gcc.target/mips/msa-dpadd-dpsub.c
@@ -0,0 +1,28 @@
+/* { dg-do compile } */
+/* { dg-options "-mfp64 -mhard-float -mmsa" } */
+/* { dg-skip-if "code quality test" { *-*-* } { "-O0" } { "" } } */
+
+typedef short v8i16 __attribute__ ((vector_size (16)));
+typedef int v4i32 __attribute__ ((vector_size (16)));
+
+void foo (int *x, v8i16 *y, v8i16 *z)
+{
+ v4i32 acc[4];
+
+ acc[0] = __builtin_msa_ld_w(x, 0);
+ acc[1] = __builtin_msa_ld_w(x, 16);
+ acc[2] = __builtin_msa_ld_w(x, 32);
+ acc[3] = __builtin_msa_ld_w(x, 48);
+
+ acc[0] = __builtin_msa_dpadd_s_w(acc[0], y[0], z[0]);
+ acc[1] = __builtin_msa_dpadd_s_w(acc[1], y[1], z[0]);
+ acc[2] = __builtin_msa_dpsub_s_w(acc[2], y[0], z[1]);
+ acc[3] = __builtin_msa_dpsub_s_w(acc[3], y[1], z[1]);
+
+ __builtin_msa_st_w(acc[0], x, 0);
+ __builtin_msa_st_w(acc[1], x, 16);
+ __builtin_msa_st_w(acc[2], x, 32);
+ __builtin_msa_st_w(acc[3], x, 48);
+}
+
+/* { dg-final { scan-assembler-not "move.v" } } */
--
2.7.4
