diff mbox series

SVE intrinsics: Add constant folding for svindex.

Message ID 2C4AC631-3F0B-4131-95C3-E6AA81AAB71C@nvidia.com
State New
Headers show
Series SVE intrinsics: Add constant folding for svindex. | expand

Commit Message

Jennifer Schmitz Oct. 16, 2024, 8:32 a.m. UTC 
This patch folds svindex with constant arguments into a vector series.
We implemented this in svindex_impl::fold using the function build_vec_series.
For example,
svuint64_t f1 ()
{
  return svindex_u642 (10, 3);
}
compiled with -O2 -march=armv8.2-a+sve, is folded to {10, 13, 16, ...}
in the gimple pass lower.
This optimization benefits cases where svindex is used in combination with
other gimple-level optimizations.
For example,
svuint64_t f2 ()
{
    return svmul_x (svptrue_b64 (), svindex_u64 (10, 3), 5);
}
has previously been compiled to
f2:
        index   z0.d, #10, #3
        mul     z0.d, z0.d, #5
        ret
Now, it is compiled to
f2:
        mov     x0, 50
        index   z0.d, x0, #15
        ret

For non-constant arguments, build_vec_series produces a VEC_SERIES_EXPR,
which is translated back at RTL level to an index instruction without codegen
changes.

We added test cases checking
- the application of the transform during gimple for constant arguments,
- the interaction with another gimple-level optimization.

The patch was bootstrapped and regtested on aarch64-linux-gnu, no regression.
OK for mainline?

Signed-off-by: Jennifer Schmitz <jschmitz@nvidia.com>

gcc/
	* config/aarch64/aarch64-sve-builtins-base.cc
	(svindex_impl::fold): Add constant folding.

gcc/testsuite/
	* gcc.target/aarch64/sve/index_const_fold.c: New test.
---
 .../aarch64/aarch64-sve-builtins-base.cc      | 12 +++++++
 .../gcc.target/aarch64/sve/index_const_fold.c | 35 +++++++++++++++++++
 2 files changed, 47 insertions(+)
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c

Comments

Jennifer Schmitz Oct. 16, 2024, 3:41 p.m. UTC | #1 
I resubmitted an corrected version of this patch in 
https://gcc.gnu.org/pipermail/gcc-patches/2024-October/665648.html

> On 16 Oct 2024, at 10:32, Jennifer Schmitz <jschmitz@nvidia.com> wrote:
> 
> This patch folds svindex with constant arguments into a vector series.
> We implemented this in svindex_impl::fold using the function build_vec_series.
> For example,
> svuint64_t f1 ()
> {
>  return svindex_u642 (10, 3);
> }
> compiled with -O2 -march=armv8.2-a+sve, is folded to {10, 13, 16, ...}
> in the gimple pass lower.
> This optimization benefits cases where svindex is used in combination with
> other gimple-level optimizations.
> For example,
> svuint64_t f2 ()
> {
>    return svmul_x (svptrue_b64 (), svindex_u64 (10, 3), 5);
> }
> has previously been compiled to
> f2:
>        index   z0.d, #10, #3
>        mul     z0.d, z0.d, #5
>        ret
> Now, it is compiled to
> f2:
>        mov     x0, 50
>        index   z0.d, x0, #15
>        ret
> 
> For non-constant arguments, build_vec_series produces a VEC_SERIES_EXPR,
> which is translated back at RTL level to an index instruction without codegen
> changes.
> 
> We added test cases checking
> - the application of the transform during gimple for constant arguments,
> - the interaction with another gimple-level optimization.
> 
> The patch was bootstrapped and regtested on aarch64-linux-gnu, no regression.
> OK for mainline?
> 
> Signed-off-by: Jennifer Schmitz <jschmitz@nvidia.com>
> 
> gcc/
> * config/aarch64/aarch64-sve-builtins-base.cc
> (svindex_impl::fold): Add constant folding.
> 
> gcc/testsuite/
> * gcc.target/aarch64/sve/index_const_fold.c: New test.
> ---
> .../aarch64/aarch64-sve-builtins-base.cc      | 12 +++++++
> .../gcc.target/aarch64/sve/index_const_fold.c | 35 +++++++++++++++++++
> 2 files changed, 47 insertions(+)
> create mode 100644 gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c
> 
> diff --git a/gcc/config/aarch64/aarch64-sve-builtins-base.cc b/gcc/config/aarch64/aarch64-sve-builtins-base.cc
> index 1c17149e1f0..f6b1657ecbb 100644
> --- a/gcc/config/aarch64/aarch64-sve-builtins-base.cc
> +++ b/gcc/config/aarch64/aarch64-sve-builtins-base.cc
> @@ -1304,6 +1304,18 @@ public:
> 
> class svindex_impl : public function_base
> {
> +public:
> +  gimple *
> +  fold (gimple_folder &f) const override
> +  {
> +    tree vec_type = TREE_TYPE (f.lhs);
> +    tree base = gimple_call_arg (f.call, 0);
> +    tree step = gimple_call_arg (f.call, 1);
> +
> +    return gimple_build_assign (f.lhs,
> + build_vec_series (vec_type, base, step));
> +  }
> +
> public:
>   rtx
>   expand (function_expander &e) const override
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c b/gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c
> new file mode 100644
> index 00000000000..f5e6c0f7a85
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c
> @@ -0,0 +1,35 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -fdump-tree-optimized" } */
> +
> +#include <arm_sve.h>
> +#include <stdint.h>
> +
> +#define INDEX_CONST(TYPE, TY) \
> +  sv##TYPE f_##TY##_index_const () \
> +  { \
> +    return svindex_##TY (10, 3); \
> +  }
> +
> +#define MULT_INDEX(TYPE, TY) \
> +  sv##TYPE f_##TY##_mult_index () \
> +  { \
> +    return svmul_x (svptrue_b8 (), \
> +    svindex_##TY (10, 3), \
> +    5); \
> +  }
> +
> +#define ALL_TESTS(TYPE, TY) \
> +  INDEX_CONST (TYPE, TY) \
> +  MULT_INDEX (TYPE, TY)
> +
> +ALL_TESTS (uint8_t, u8)
> +ALL_TESTS (uint16_t, u16)
> +ALL_TESTS (uint32_t, u32)
> +ALL_TESTS (uint64_t, u64)
> +ALL_TESTS (int8_t, s8)
> +ALL_TESTS (int16_t, s16)
> +ALL_TESTS (int32_t, s32)
> +ALL_TESTS (int64_t, s64)
> +
> +/* { dg-final { scan-tree-dump "return \\{ 10, 13, 16, ... \\}" 8 "optimized" } } */
> +/* { dg-final { scan-tree-dump "return \\{ 50, 65, 80, ... \\}" 8 "optimized" } } */
> -- 
> 2.44.0
diff mbox series

Patch

diff --git a/gcc/config/aarch64/aarch64-sve-builtins-base.cc b/gcc/config/aarch64/aarch64-sve-builtins-base.cc
index 1c17149e1f0..f6b1657ecbb 100644
--- a/gcc/config/aarch64/aarch64-sve-builtins-base.cc
+++ b/gcc/config/aarch64/aarch64-sve-builtins-base.cc
@@ -1304,6 +1304,18 @@  public:
 
 class svindex_impl : public function_base
 {
+public:
+  gimple *
+  fold (gimple_folder &f) const override
+  {
+    tree vec_type = TREE_TYPE (f.lhs);
+    tree base = gimple_call_arg (f.call, 0);
+    tree step = gimple_call_arg (f.call, 1);
+
+    return gimple_build_assign (f.lhs,
+				build_vec_series (vec_type, base, step));
+  }
+
 public:
   rtx
   expand (function_expander &e) const override
diff --git a/gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c b/gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c
new file mode 100644
index 00000000000..f5e6c0f7a85
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve/index_const_fold.c
@@ -0,0 +1,35 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O2 -fdump-tree-optimized" } */
+
+#include <arm_sve.h>
+#include <stdint.h>
+
+#define INDEX_CONST(TYPE, TY)				\
+  sv##TYPE f_##TY##_index_const ()			\
+  {							\
+    return svindex_##TY (10, 3);			\
+  }
+
+#define MULT_INDEX(TYPE, TY)				\
+  sv##TYPE f_##TY##_mult_index ()			\
+  {							\
+    return svmul_x (svptrue_b8 (),			\
+		    svindex_##TY (10, 3),		\
+		    5);					\
+  }
+
+#define ALL_TESTS(TYPE, TY)				\
+  INDEX_CONST (TYPE, TY)				\
+  MULT_INDEX (TYPE, TY)
+
+ALL_TESTS (uint8_t, u8)
+ALL_TESTS (uint16_t, u16)
+ALL_TESTS (uint32_t, u32)
+ALL_TESTS (uint64_t, u64)
+ALL_TESTS (int8_t, s8)
+ALL_TESTS (int16_t, s16)
+ALL_TESTS (int32_t, s32)
+ALL_TESTS (int64_t, s64)
+
+/* { dg-final { scan-tree-dump "return \\{ 10, 13, 16, ... \\}" 8 "optimized" } } */
+/* { dg-final { scan-tree-dump "return \\{ 50, 65, 80, ... \\}" 8 "optimized" } } */