AArch64: Improve immediate generation
Checks
Commit Message
Further improve immediate generation by adding support for 2-instruction
MOV/EOR bitmask immediates. This reduces the number of 3/4-instruction
immediates in SPECCPU2017 by ~2%.
Passes regress, OK for commit?
gcc/ChangeLog:
* config/aarch64/aarch64.cc (aarch64_internal_mov_immediate)
Add support for immediates using MOV/EOR bitmask.
gcc/testsuite:
* gcc.target/aarch64/imm_choice_comparison.c: Fix test.
* gcc.target/aarch64/moveor_imm.c: Add new test.
* gcc.target/aarch64/pr106583.c: Fix test.
---
Comments
On 19/10/2023 13:43, Wilco Dijkstra wrote:
> Further improve immediate generation by adding support for 2-instruction
> MOV/EOR bitmask immediates. This reduces the number of 3/4-instruction
> immediates in SPECCPU2017 by ~2%.
>
> Passes regress, OK for commit?
>
> gcc/ChangeLog:
> * config/aarch64/aarch64.cc (aarch64_internal_mov_immediate)
> Add support for immediates using MOV/EOR bitmask.
>
> gcc/testsuite:
> * gcc.target/aarch64/imm_choice_comparison.c: Fix test.
> * gcc.target/aarch64/moveor_imm.c: Add new test.
> * gcc.target/aarch64/pr106583.c: Fix test.
>
> ---
>
> diff --git a/gcc/config/aarch64/aarch64.cc b/gcc/config/aarch64/aarch64.cc
> index 578a253d6e0e133e19592553fc873b3e73f9f218..ed5be2b64c9a767d74e9d78415da964c669001aa 100644
> --- a/gcc/config/aarch64/aarch64.cc
> +++ b/gcc/config/aarch64/aarch64.cc
> @@ -5748,6 +5748,26 @@ aarch64_internal_mov_immediate (rtx dest, rtx imm, bool generate,
> }
> return 2;
> }
> +
> + /* Try 2 bitmask immediates which are xor'd together. */
> + for (i = 0; i < 64; i += 16)
> + {
> + val2 = (val >> i) & mask;
> + val2 |= val2 << 16;
> + val2 |= val2 << 32;
> + if (aarch64_bitmask_imm (val2) && aarch64_bitmask_imm (val ^ val2))
> + break;
> + }
> +
> + if (i != 64)
> + {
> + if (generate)
> + {
> + emit_insn (gen_rtx_SET (dest, GEN_INT (val2)));
> + emit_insn (gen_xordi3 (dest, dest, GEN_INT (val ^ val2)));
> + }
> + return 2;
> + }
> }
>
> /* Try a bitmask plus 2 movk to generate the immediate in 3 instructions. */
> diff --git a/gcc/testsuite/gcc.target/aarch64/imm_choice_comparison.c b/gcc/testsuite/gcc.target/aarch64/imm_choice_comparison.c
> index ebc44d6dbc7287d907603d77d7b54496de177c4b..2434ca380ca2cad3e1e4181deeaad680f518b866 100644
> --- a/gcc/testsuite/gcc.target/aarch64/imm_choice_comparison.c
> +++ b/gcc/testsuite/gcc.target/aarch64/imm_choice_comparison.c
> @@ -6,7 +6,7 @@
> int
> foo (long long x)
> {
> - return x <= 0x1999999999999998;
> + return x <= 0x0000999999999998;
> }
>
> int
> diff --git a/gcc/testsuite/gcc.target/aarch64/moveor_imm.c b/gcc/testsuite/gcc.target/aarch64/moveor_imm.c
> new file mode 100644
> index 0000000000000000000000000000000000000000..5f4997b50398fdda5924610959e0c54967ad0735
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/moveor_imm.c
> @@ -0,0 +1,31 @@
> +/* { dg-do assemble } */
> +/* { dg-options "-O2 --save-temps" } */
> +
> +long f1 (void)
> +{
> + return 0x2aaaaaaaaaaaaaab;
> +}
> +
> +long f2 (void)
> +{
> + return 0x10f0f0f0f0f0f0f1;
> +}
> +
> +long f3 (void)
> +{
> + return 0xccccccccccccccd;
> +}
> +
> +long f4 (void)
> +{
> + return 0x1999999999999998;
> +}
> +
> +long f5 (void)
> +{
> + return 0x3f3333333f333333;
> +}
> +
> +/* { dg-final { scan-assembler-not {\tmovk\t} } } */
> +/* { dg-final { scan-assembler-times {\tmov\t} 5 } } */
> +/* { dg-final { scan-assembler-times {\teor\t} 5 } } */
> diff --git a/gcc/testsuite/gcc.target/aarch64/pr106583.c b/gcc/testsuite/gcc.target/aarch64/pr106583.c
> index 0f931580817d78dc1cc58f03b251bd21bec71f59..79ada5160ce059d66eeaee407ca02488b2a1f114 100644
> --- a/gcc/testsuite/gcc.target/aarch64/pr106583.c
> +++ b/gcc/testsuite/gcc.target/aarch64/pr106583.c
> @@ -3,7 +3,7 @@
>
> long f1 (void)
> {
> - return 0x7efefefefefefeff;
> + return 0x75fefefefefefeff;
> }
>
> long f2 (void)
>
I think the tests should be converted to use check-function-bodies,
rather than scanning for counts on the entire file. It makes it far
more obvious what's changed if a test starts to fail. The functions are
all trivial, so the test can be quite precise.
Otherwise, this LGTM.
R.
@@ -5748,6 +5748,26 @@ aarch64_internal_mov_immediate (rtx dest, rtx imm, bool generate,
}
return 2;
}
+
+ /* Try 2 bitmask immediates which are xor'd together. */
+ for (i = 0; i < 64; i += 16)
+ {
+ val2 = (val >> i) & mask;
+ val2 |= val2 << 16;
+ val2 |= val2 << 32;
+ if (aarch64_bitmask_imm (val2) && aarch64_bitmask_imm (val ^ val2))
+ break;
+ }
+
+ if (i != 64)
+ {
+ if (generate)
+ {
+ emit_insn (gen_rtx_SET (dest, GEN_INT (val2)));
+ emit_insn (gen_xordi3 (dest, dest, GEN_INT (val ^ val2)));
+ }
+ return 2;
+ }
}
/* Try a bitmask plus 2 movk to generate the immediate in 3 instructions. */
@@ -6,7 +6,7 @@
int
foo (long long x)
{
- return x <= 0x1999999999999998;
+ return x <= 0x0000999999999998;
}
int
new file mode 100644
@@ -0,0 +1,31 @@
+/* { dg-do assemble } */
+/* { dg-options "-O2 --save-temps" } */
+
+long f1 (void)
+{
+ return 0x2aaaaaaaaaaaaaab;
+}
+
+long f2 (void)
+{
+ return 0x10f0f0f0f0f0f0f1;
+}
+
+long f3 (void)
+{
+ return 0xccccccccccccccd;
+}
+
+long f4 (void)
+{
+ return 0x1999999999999998;
+}
+
+long f5 (void)
+{
+ return 0x3f3333333f333333;
+}
+
+/* { dg-final { scan-assembler-not {\tmovk\t} } } */
+/* { dg-final { scan-assembler-times {\tmov\t} 5 } } */
+/* { dg-final { scan-assembler-times {\teor\t} 5 } } */
@@ -3,7 +3,7 @@
long f1 (void)
{
- return 0x7efefefefefefeff;
+ return 0x75fefefefefefeff;
}
long f2 (void)