Re: [PATCH] rs6000: Teach rs6000_opaque_type_invalid_use_p about inline asm [PR108272]
on 2023/1/6 17:26, Kewen.Lin via Gcc-patches wrote:
> Hi,
>
> As PR108272 shows, there are some invalid uses of MMA opaque
> types in inline asm statements. This patch is to teach the
> function rs6000_opaque_type_invalid_use_p for inline asm,
> check and error any invalid use of MMA opaque types in input
> and output operands.
>
> Bootstrapped and regtested on powerpc64-linux-gnu P8 and
> powerpc64le-linux-gnu P9 and P10.
>
> I'm going to push this next week if no objections.
Committed the attached one in r13-5184-g074b0c03eabeb8.
BR,
Kewen
Subject: [PATCH] rs6000: Teach rs6000_opaque_type_invalid_use_p about inline
asm [PR108272]
As PR108272 shows, there are some invalid uses of MMA opaque
types in inline asm statements. This patch is to teach the
function rs6000_opaque_type_invalid_use_p for inline asm,
check and error any invalid use of MMA opaque types in input
and output operands.
PR target/108272
gcc/ChangeLog:
* config/rs6000/rs6000.cc (rs6000_opaque_type_invalid_use_p): Add the
support for invalid uses in inline asm, factor out the checking and
erroring to lambda function check_and_error_invalid_use.
gcc/testsuite/ChangeLog:
* gcc.target/powerpc/pr108272-1.c: New test.
* gcc.target/powerpc/pr108272-2.c: New test.
* gcc.target/powerpc/pr108272-3.c: New test.
* gcc.target/powerpc/pr108272-4.c: New test.
---
gcc/config/rs6000/rs6000.cc | 51 +++
gcc/testsuite/gcc.target/powerpc/pr108272-1.c | 17 +++
gcc/testsuite/gcc.target/powerpc/pr108272-2.c | 17 +++
gcc/testsuite/gcc.target/powerpc/pr108272-3.c | 17 +++
gcc/testsuite/gcc.target/powerpc/pr108272-4.c | 18 +++
5 files changed, 110 insertions(+), 10 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr108272-1.c
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr108272-2.c
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr108272-3.c
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr108272-4.c
diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc
index 3baa2c3b7b0..4287a6b1a61 100644
--- a/gcc/config/rs6000/rs6000.cc
+++ b/gcc/config/rs6000/rs6000.cc
@@ -28928,9 +28928,9 @@ constant_generates_xxspltidp (vec_const_128bit_type
*vsx_const)
__vector_pair built-in types. They are target specific and
only available when MMA is supported. With MMA supported, it
simply returns true, otherwise it checks if the given gimple
- STMT is an assignment stmt and uses either of these two opaque
- types unexpectedly, if yes, it would raise an error message
- and returns true, otherwise it returns false. */
+ STMT is an assignment or asm stmt and uses either of these two
+ opaque types unexpectedly, if yes, it would raise an error
+ message and returns true, otherwise it returns false. */
bool
rs6000_opaque_type_invalid_use_p (gimple *stmt)
@@ -28938,23 +28938,54 @@ rs6000_opaque_type_invalid_use_p (gimple *stmt)
if (TARGET_MMA)
return false;
+ /* If the given TYPE is one MMA opaque type, emit the corresponding
+ error messages and return true, otherwise return false. */
+ auto check_and_error_invalid_use = [](tree type)
+ {
+tree mv = TYPE_MAIN_VARIANT (type);
+if (mv == vector_quad_type_node)
+ {
+ error ("type %<__vector_quad%> requires the %qs option", "-mmma");
+ return true;
+ }
+else if (mv == vector_pair_type_node)
+ {
+ error ("type %<__vector_pair%> requires the %qs option", "-mmma");
+ return true;
+ }
+return false;
+ };
+
if (stmt)
{
/* The usage of MMA opaque types is very limited for now,
-to check with gassign is enough so far. */
+to check with gassign and gasm is enough so far. */
if (gassign *ga = dyn_cast (stmt))
{
tree lhs = gimple_assign_lhs (ga);
tree type = TREE_TYPE (lhs);
- if (type == vector_quad_type_node)
+ if (check_and_error_invalid_use (type))
+ return true;
+ }
+ else if (gasm *gs = dyn_cast (stmt))
+ {
+ unsigned ninputs = gimple_asm_ninputs (gs);
+ for (unsigned i = 0; i < ninputs; i++)
{
- error ("type %<__vector_quad%> requires the %qs option", "-mmma");
- return true;
+ tree op = gimple_asm_input_op (gs, i);
+ tree val = TREE_VALUE (op);
+ tree type = TREE_TYPE (val);
+ if (check_and_error_invalid_use (type))
+ return true;
}
- else if (type == vector_pair_type_node)
+ unsigned noutputs = gimple_asm_noutputs (gs);
+ for (unsigned i = 0; i < noutputs; i++)
{
- error ("type %<__vector_pair%> requires the %qs option", "-mmma&quo
[PATCH] rs6000: Teach rs6000_opaque_type_invalid_use_p about inline asm [PR108272]
Hi,
As PR108272 shows, there are some invalid uses of MMA opaque
types in inline asm statements. This patch is to teach the
function rs6000_opaque_type_invalid_use_p for inline asm,
check and error any invalid use of MMA opaque types in input
and output operands.
Bootstrapped and regtested on powerpc64-linux-gnu P8 and
powerpc64le-linux-gnu P9 and P10.
I'm going to push this next week if no objections.
Kewen
BR,
-
PR target/108272
gcc/ChangeLog:
* config/rs6000/rs6000.cc (rs6000_opaque_type_invalid_use_p): Add the
support for invalid uses in inline asm, factor out the checking and
erroring to lambda function check_and_error_invalid_use.
gcc/testsuite/ChangeLog:
* gcc.target/powerpc/pr108272-1.c: New test.
* gcc.target/powerpc/pr108272-2.c: New test.
---
gcc/config/rs6000/rs6000.cc | 50 +++
gcc/testsuite/gcc.target/powerpc/pr108272-1.c | 17 +++
gcc/testsuite/gcc.target/powerpc/pr108272-2.c | 17 +++
3 files changed, 74 insertions(+), 10 deletions(-)
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr108272-1.c
create mode 100644 gcc/testsuite/gcc.target/powerpc/pr108272-2.c
diff --git a/gcc/config/rs6000/rs6000.cc b/gcc/config/rs6000/rs6000.cc
index d362668ba13..46f74922755 100644
--- a/gcc/config/rs6000/rs6000.cc
+++ b/gcc/config/rs6000/rs6000.cc
@@ -28903,9 +28903,9 @@ constant_generates_xxspltidp (vec_const_128bit_type
*vsx_const)
__vector_pair built-in types. They are target specific and
only available when MMA is supported. With MMA supported, it
simply returns true, otherwise it checks if the given gimple
- STMT is an assignment stmt and uses either of these two opaque
- types unexpectedly, if yes, it would raise an error message
- and returns true, otherwise it returns false. */
+ STMT is an assignment or asm stmt and uses either of these two
+ opaque types unexpectedly, if yes, it would raise an error
+ message and returns true, otherwise it returns false. */
bool
rs6000_opaque_type_invalid_use_p (gimple *stmt)
@@ -28913,23 +28913,53 @@ rs6000_opaque_type_invalid_use_p (gimple *stmt)
if (TARGET_MMA)
return false;
+ /* If the given TYPE is one MMA opaque type, emit the corresponding
+ error messages and return true, otherwise return false. */
+ auto check_and_error_invalid_use = [](tree type)
+ {
+if (type == vector_quad_type_node)
+ {
+ error ("type %<__vector_quad%> requires the %qs option", "-mmma");
+ return true;
+ }
+else if (type == vector_pair_type_node)
+ {
+ error ("type %<__vector_pair%> requires the %qs option", "-mmma");
+ return true;
+ }
+return false;
+ };
+
if (stmt)
{
/* The usage of MMA opaque types is very limited for now,
-to check with gassign is enough so far. */
+to check with gassign and gasm is enough so far. */
if (gassign *ga = dyn_cast (stmt))
{
tree lhs = gimple_assign_lhs (ga);
tree type = TREE_TYPE (lhs);
- if (type == vector_quad_type_node)
+ if (check_and_error_invalid_use (type))
+ return true;
+ }
+ else if (gasm *gs = dyn_cast (stmt))
+ {
+ unsigned ninputs = gimple_asm_ninputs (gs);
+ for (unsigned i = 0; i < ninputs; i++)
{
- error ("type %<__vector_quad%> requires the %qs option", "-mmma");
- return true;
+ tree op = gimple_asm_input_op (gs, i);
+ tree val = TREE_VALUE (op);
+ tree type = TREE_TYPE (val);
+ if (check_and_error_invalid_use (type))
+ return true;
}
- else if (type == vector_pair_type_node)
+ unsigned noutputs = gimple_asm_noutputs (gs);
+ for (unsigned i = 0; i < noutputs; i++)
{
- error ("type %<__vector_pair%> requires the %qs option", "-mmma");
- return true;
+ tree op = gimple_asm_output_op (gs, i);
+ tree val = TREE_VALUE (op);
+ tree type = TREE_TYPE (val);
+ if (check_and_error_invalid_use (type))
+ return true;
}
}
}
diff --git a/gcc/testsuite/gcc.target/powerpc/pr108272-1.c
b/gcc/testsuite/gcc.target/powerpc/pr108272-1.c
new file mode 100644
index 000..b99e6a4d86d
--- /dev/null
+++ b/gcc/testsuite/gcc.target/powerpc/pr108272-1.c
@@ -0,0 +1,17 @@
+/* { dg-require-effective-target powerpc_p9modulo_ok } */
+/* If the default cpu type is power10 or later, type __vector_quad is
+ supported. To keep the test point available all the time, this case
+ specifies -mdejagnu-cpu=power9 here. */
+/* { dg-options "-mdejagnu-cpu=power9" } */
+
+/* Verify there is no ICE and don't check the error messages on unsupported
+ type since they could be fragile and are not test points of this case. */
+
+/* { dg-excess-errors "pr108272-1" } */
+
+void
