tree-optimization: Query ranger for non-SSA niter bound expressions

Message ID 20260602172349.20432-1-abhishek.kaushik@arm.com
State New
Headers
Series tree-optimization: Query ranger for non-SSA niter bound expressions |

Checks

Context Check Description
linaro-tcwg-bot/tcwg_gcc_build--master-aarch64 success Build passed
linaro-tcwg-bot/tcwg_simplebootstrap_build--master-aarch64-bootstrap success Build passed
linaro-tcwg-bot/tcwg_gcc_build--master-arm success Build passed
linaro-tcwg-bot/tcwg_gcc_check--master-aarch64 success Test passed
linaro-tcwg-bot/tcwg_gcc_check--master-arm success Test passed
linaro-tcwg-bot/tcwg_simplebootstrap_build--master-arm-bootstrap success Build passed

Commit Message

Abhishek Kaushik June 2, 2026, 5:23 p.m. UTC
  determine_value_range only queried ranger when VAR was an SSA_NAME.
However, number_of_iterations_exit_assumptions calls
expand_simple_operations on the IV bases before asking for the range.
This can expose simple GENERIC expressions, including conversions, even
when the original value had an SSA_NAME with useful range information.

Ranger can analyze such expressions at a given program point, so query it
for integral bound expressions using the loop header as context rather
than restricting the query to SSA_NAMEs.  This lets niter analysis
recover the narrower range for value-preserving conversions such as
uint8_t to int, while still handling non-value-preserving conversions and
wrapping expressions conservatively.

The new test covers a direct converted uint8_t bound, a masked uint16_t
bound whose useful range comes from ranger, a guarded uint16_t bound
whose range is context-sensitive, a non-value-preserving uint8_t to
int8_t conversion, and a wrapping unsigned expression.

Bootstrapped and regression tested on aarch64-unknown-linux-gnu.

gcc/ChangeLog:

	* tree-ssa-loop-niter.cc (determine_value_range): Query ranger
	for integral expressions, not only SSA_NAMEs.  Pass the loop
	header context to range queries.

gcc/testsuite/ChangeLog:

	* gcc.target/aarch64/sve2/niter-convert-range.c: New test.
---
 .../aarch64/sve2/niter-convert-range.c        | 68 +++++++++++++++++++
 gcc/tree-ssa-loop-niter.cc                    | 10 +--
 2 files changed, 74 insertions(+), 4 deletions(-)
 create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c
  

Comments

Richard Biener June 3, 2026, 7:56 a.m. UTC | #1
On Tue, 2 Jun 2026, Abhishek Kaushik wrote:

> determine_value_range only queried ranger when VAR was an SSA_NAME.
> However, number_of_iterations_exit_assumptions calls
> expand_simple_operations on the IV bases before asking for the range.
> This can expose simple GENERIC expressions, including conversions, even
> when the original value had an SSA_NAME with useful range information.
> 
> Ranger can analyze such expressions at a given program point, so query it
> for integral bound expressions using the loop header as context rather
> than restricting the query to SSA_NAMEs.  This lets niter analysis
> recover the narrower range for value-preserving conversions such as
> uint8_t to int, while still handling non-value-preserving conversions and
> wrapping expressions conservatively.
> 
> The new test covers a direct converted uint8_t bound, a masked uint16_t
> bound whose useful range comes from ranger, a guarded uint16_t bound
> whose range is context-sensitive, a non-value-preserving uint8_t to
> int8_t conversion, and a wrapping unsigned expression.
> 
> Bootstrapped and regression tested on aarch64-unknown-linux-gnu.
> 
> gcc/ChangeLog:
> 
> 	* tree-ssa-loop-niter.cc (determine_value_range): Query ranger
> 	for integral expressions, not only SSA_NAMEs.  Pass the loop
> 	header context to range queries.
> 
> gcc/testsuite/ChangeLog:
> 
> 	* gcc.target/aarch64/sve2/niter-convert-range.c: New test.
> ---
>  .../aarch64/sve2/niter-convert-range.c        | 68 +++++++++++++++++++
>  gcc/tree-ssa-loop-niter.cc                    | 10 +--
>  2 files changed, 74 insertions(+), 4 deletions(-)
>  create mode 100644 gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c
> 
> diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c b/gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c
> new file mode 100644
> index 00000000000..c96b2f3999c
> --- /dev/null
> +++ b/gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c
> @@ -0,0 +1,68 @@
> +/* { dg-do compile } */
> +/* { dg-options "-O2 -march=armv8.3-a+sve2 -fdump-tree-vect-details" } */
> +
> +#include <stdint.h>
> +
> +int
> +foo (char *buf, uint8_t len)
> +{
> +  int x = 0;
> +
> +  for (int16_t i = 0, y = 0; i < len; i++, y = i * 10)
> +    x += (int) y * (buf[i] - '0');
> +
> +  return x;
> +}
> +
> +int
> +masked_len_foo (unsigned char *buf, uint16_t len)
> +{
> +  uint16_t n = len & 255;
> +  int x = 0;
> +
> +  for (int16_t i = 0, y = 0; i < n; i++, y = i * 10)
> +    x += (int) y * (buf[i] - '0');
> +
> +  return x;
> +}
> +
> +int
> +guarded_len_foo (char *buf, uint16_t len)
> +{
> +  if (len > 255)
> +    return 0;
> +
> +  int x = 0;
> +
> +  for (int16_t i = 0, y = 0; i < len; i++, y = i * 10)
> +    x += (int) y * (buf[i] - '0');
> +
> +  return x;
> +}
> +
> +int
> +non_value_preserving (unsigned char *buf, uint8_t len)
> +{
> +  int x = 0;
> +
> +  for (int16_t i = 0, y = 0; i < (int) (int8_t) len; i++, y = i * 10)
> +    x += (int) y * (buf[i] - '0');
> +
> +  return x;
> +}
> +
> +unsigned
> +niter_convert_range_wrap (unsigned char *buf, uint8_t len)
> +{
> +  unsigned x = 0;
> +
> +  for (unsigned i = 0, y = 0; i < (unsigned) len - 1; i++, y = i * 10)
> +    x += y * (buf[i] - '0');
> +
> +  return x;
> +}
> +
> +/* { dg-final { scan-tree-dump-times {bounds on difference of bases: 0 [.][.][.] 126} 3 "vect" } } */
> +/* { dg-final { scan-tree-dump-times {bounds on difference of bases: 0 [.][.][.] 254} 9 "vect" } } */
> +/* { dg-final { scan-tree-dump-times {bounds on difference of bases: -1 [.][.][.] 4294967294} 3 "vect" } } */
> +/* { dg-final { scan-tree-dump-times "loop vectorized using variable length vectors" 5 "vect" } } */
> diff --git a/gcc/tree-ssa-loop-niter.cc b/gcc/tree-ssa-loop-niter.cc
> index f179be1ffc7..b3a0aebfba2 100644
> --- a/gcc/tree-ssa-loop-niter.cc
> +++ b/gcc/tree-ssa-loop-niter.cc
> @@ -356,15 +356,16 @@ determine_value_range (class loop *loop, tree type, tree var, mpz_t off,
>    get_type_static_bounds (type, min, max);
>  
>    /* See if we have some range info from VRP.  */
> -  if (TREE_CODE (var) == SSA_NAME && INTEGRAL_TYPE_P (type))
> +  if (INTEGRAL_TYPE_P (type))
>      {
>        edge e = loop_preheader_edge (loop);
>        signop sgn = TYPE_SIGN (type);
>        gphi_iterator gsi;
> +      gimple *ctx = last_nondebug_stmt (loop->header);
>  
>        /* Either for VAR itself...  */
>        int_range_max var_range (TREE_TYPE (var));
> -      get_range_query (cfun)->range_of_expr (var_range, var);
> +      get_range_query (cfun)->range_of_expr (var_range, var, ctx);
>        if (var_range.varying_p () || var_range.undefined_p ())
>  	rtype = VR_VARYING;
>        else
> @@ -383,7 +384,8 @@ determine_value_range (class loop *loop, tree type, tree var, mpz_t off,
>  	  gphi *phi = gsi.phi ();
>  	  if (PHI_ARG_DEF_FROM_EDGE (phi, e) == var
>  	      && get_range_query (cfun)->range_of_expr (phi_range,
> -						    gimple_phi_result (phi))
> +						    gimple_phi_result (phi),
> +						    ctx)

I believe you can use ->range_on_edge here, so no need to compute 'ctx'
but instead pass in the preheader edge as context (last_nondebug_stmt
of the loop header looks not exactly correct).

>  	      && !phi_range.varying_p ()
>  	      && !phi_range.undefined_p ())
>  	    {
> @@ -404,7 +406,7 @@ determine_value_range (class loop *loop, tree type, tree var, mpz_t off,
>  		  if (wi::gt_p (minv, maxv, sgn))
>  		    {
>  		      int_range_max vr (TREE_TYPE (var));
> -		      get_range_query (cfun)->range_of_expr (vr, var);
> +		      get_range_query (cfun)->range_of_expr (vr, var, ctx);

Similar here.  Would be nice to amend the functions documentation
that the range computation is valid on the entry edge of the loop.

Thanks,
Richard.

>  		      if (vr.varying_p () || vr.undefined_p ())
>  			rtype = VR_VARYING;
>  		      else
>
  

Patch

diff --git a/gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c b/gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c
new file mode 100644
index 00000000000..c96b2f3999c
--- /dev/null
+++ b/gcc/testsuite/gcc.target/aarch64/sve2/niter-convert-range.c
@@ -0,0 +1,68 @@ 
+/* { dg-do compile } */
+/* { dg-options "-O2 -march=armv8.3-a+sve2 -fdump-tree-vect-details" } */
+
+#include <stdint.h>
+
+int
+foo (char *buf, uint8_t len)
+{
+  int x = 0;
+
+  for (int16_t i = 0, y = 0; i < len; i++, y = i * 10)
+    x += (int) y * (buf[i] - '0');
+
+  return x;
+}
+
+int
+masked_len_foo (unsigned char *buf, uint16_t len)
+{
+  uint16_t n = len & 255;
+  int x = 0;
+
+  for (int16_t i = 0, y = 0; i < n; i++, y = i * 10)
+    x += (int) y * (buf[i] - '0');
+
+  return x;
+}
+
+int
+guarded_len_foo (char *buf, uint16_t len)
+{
+  if (len > 255)
+    return 0;
+
+  int x = 0;
+
+  for (int16_t i = 0, y = 0; i < len; i++, y = i * 10)
+    x += (int) y * (buf[i] - '0');
+
+  return x;
+}
+
+int
+non_value_preserving (unsigned char *buf, uint8_t len)
+{
+  int x = 0;
+
+  for (int16_t i = 0, y = 0; i < (int) (int8_t) len; i++, y = i * 10)
+    x += (int) y * (buf[i] - '0');
+
+  return x;
+}
+
+unsigned
+niter_convert_range_wrap (unsigned char *buf, uint8_t len)
+{
+  unsigned x = 0;
+
+  for (unsigned i = 0, y = 0; i < (unsigned) len - 1; i++, y = i * 10)
+    x += y * (buf[i] - '0');
+
+  return x;
+}
+
+/* { dg-final { scan-tree-dump-times {bounds on difference of bases: 0 [.][.][.] 126} 3 "vect" } } */
+/* { dg-final { scan-tree-dump-times {bounds on difference of bases: 0 [.][.][.] 254} 9 "vect" } } */
+/* { dg-final { scan-tree-dump-times {bounds on difference of bases: -1 [.][.][.] 4294967294} 3 "vect" } } */
+/* { dg-final { scan-tree-dump-times "loop vectorized using variable length vectors" 5 "vect" } } */
diff --git a/gcc/tree-ssa-loop-niter.cc b/gcc/tree-ssa-loop-niter.cc
index f179be1ffc7..b3a0aebfba2 100644
--- a/gcc/tree-ssa-loop-niter.cc
+++ b/gcc/tree-ssa-loop-niter.cc
@@ -356,15 +356,16 @@  determine_value_range (class loop *loop, tree type, tree var, mpz_t off,
   get_type_static_bounds (type, min, max);
 
   /* See if we have some range info from VRP.  */
-  if (TREE_CODE (var) == SSA_NAME && INTEGRAL_TYPE_P (type))
+  if (INTEGRAL_TYPE_P (type))
     {
       edge e = loop_preheader_edge (loop);
       signop sgn = TYPE_SIGN (type);
       gphi_iterator gsi;
+      gimple *ctx = last_nondebug_stmt (loop->header);
 
       /* Either for VAR itself...  */
       int_range_max var_range (TREE_TYPE (var));
-      get_range_query (cfun)->range_of_expr (var_range, var);
+      get_range_query (cfun)->range_of_expr (var_range, var, ctx);
       if (var_range.varying_p () || var_range.undefined_p ())
 	rtype = VR_VARYING;
       else
@@ -383,7 +384,8 @@  determine_value_range (class loop *loop, tree type, tree var, mpz_t off,
 	  gphi *phi = gsi.phi ();
 	  if (PHI_ARG_DEF_FROM_EDGE (phi, e) == var
 	      && get_range_query (cfun)->range_of_expr (phi_range,
-						    gimple_phi_result (phi))
+						    gimple_phi_result (phi),
+						    ctx)
 	      && !phi_range.varying_p ()
 	      && !phi_range.undefined_p ())
 	    {
@@ -404,7 +406,7 @@  determine_value_range (class loop *loop, tree type, tree var, mpz_t off,
 		  if (wi::gt_p (minv, maxv, sgn))
 		    {
 		      int_range_max vr (TREE_TYPE (var));
-		      get_range_query (cfun)->range_of_expr (vr, var);
+		      get_range_query (cfun)->range_of_expr (vr, var, ctx);
 		      if (vr.varying_p () || vr.undefined_p ())
 			rtype = VR_VARYING;
 		      else