procyberian/gcc
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

nvptx: Support '-mfake-ptx-alloca'

Browse source 
With '-mfake-ptx-alloca' enabled, the user-visible behavior changes only
for configurations where PTX 'alloca' is not available.  Rather than a
compile-time 'sorry, unimplemented: dynamic stack allocation not supported'
in presence of dynamic stack allocation, compilation and assembly then
succeeds.  However, attempting to link in such '*.o' files then fails due
to unresolved symbol '__GCC_nvptx__PTX_alloca_not_supported'.

This is meant to be used in scenarios where large volumes of code are
compiled, a small fraction of which runs into dynamic stack allocation, but
these parts are not important for specific use cases, and we'd thus like the
build to succeed, and error out just upon actual, very rare use of the
offending '*.o' files.

	gcc/
	* config/nvptx/nvptx.opt (-mfake-ptx-alloca): New.
	* config/nvptx/nvptx-protos.h (nvptx_output_fake_ptx_alloca):
	Declare.
	* config/nvptx/nvptx.cc (nvptx_output_fake_ptx_alloca): New.
	* config/nvptx/nvptx.md (define_insn "@nvptx_alloca_<mode>")
	[!(TARGET_PTX_7_3 && TARGET_SM52)]: Use it for
	'-mfake-ptx-alloca'.
	gcc/testsuite/
	* gcc.target/nvptx/alloca-1-O0_-mfake-ptx-alloca.c: New.
	* gcc.target/nvptx/alloca-2-O0_-mfake-ptx-alloca.c: Likewise.
	* gcc.target/nvptx/alloca-4-O3_-mfake-ptx-alloca.c: Likewise.
	* gcc.target/nvptx/vla-1-O0_-mfake-ptx-alloca.c: Likewise.
	* gcc.target/nvptx/alloca-4-O3.c:
	'dg-additional-options -mfake-ptx-alloca'.
This commit is contained in:
Thomas Schwinge 2025-02-24 16:13:11 +01:00
parent 22e76700ae
commit 1146410c0f
9 changed files with 187 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

1
gcc/config/nvptx/nvptx-protos.h
View file

@ -55,6 +55,7 @@ extern rtx nvptx_expand_compare (rtx);
extern const char *nvptx_ptx_type_from_mode (machine_mode, bool);
extern const char *nvptx_output_mov_insn (rtx, rtx);
extern const char *nvptx_output_call_insn (rtx_insn *, rtx, rtx);
extern const char *nvptx_output_fake_ptx_alloca (void);
extern const char *nvptx_output_return (void);
extern const char *nvptx_output_set_softstack (unsigned);
extern const char *nvptx_output_simt_enter (rtx, rtx, rtx);

21
gcc/config/nvptx/nvptx.cc
View file

@ -1758,6 +1758,27 @@ nvptx_output_set_softstack (unsigned src_regno)
}
return "";
}
/* Output a fake PTX 'alloca'. */
const char *
nvptx_output_fake_ptx_alloca (void)
{
#define FAKE_PTX_ALLOCA_NAME "__GCC_nvptx__PTX_alloca_not_supported"
static tree decl;
if (!decl)
{
tree alloca_type = TREE_TYPE (builtin_decl_explicit (BUILT_IN_ALLOCA));
decl = build_decl (UNKNOWN_LOCATION, FUNCTION_DECL,
get_identifier (FAKE_PTX_ALLOCA_NAME), alloca_type);
DECL_EXTERNAL (decl) = 1;
TREE_PUBLIC (decl) = 1;
nvptx_record_needed_fndecl (decl);
}
return "\tcall\t(%0), " FAKE_PTX_ALLOCA_NAME ", (%1);";
#undef FAKE_PTX_ALLOCA_NAME
}
/* Output a return instruction. Also copy the return value to its outgoing
location. */

4
gcc/config/nvptx/nvptx.md
View file

@ -1705,6 +1705,8 @@
output_asm_insn ("}", NULL);
return "";
}
else if (nvptx_fake_ptx_alloca)
return nvptx_output_fake_ptx_alloca ();
else
{
sorry_at (INSN_LOCATION (insn),
@ -1733,6 +1735,7 @@
gcc_checking_assert (REG_P (operands[0]));
emit_insn (gen_nvptx_stacksave (Pmode, operands[0], operands[1]));
}
/* We don't bother to special-case '-mfake-ptx-alloca' here. */
else
{
/* The concept of a '%stack' pointer doesn't apply like this.
@ -1765,6 +1768,7 @@
operands[1] = force_reg (Pmode, operands[1]);
emit_insn (gen_nvptx_stackrestore (Pmode, operands[0], operands[1]));
}
/* We don't bother to special-case '-mfake-ptx-alloca' here. */
else if (!TARGET_SOFT_STACK)
; /* See 'save_stack_block'. */
else if (TARGET_SOFT_STACK)

15
gcc/config/nvptx/nvptx.opt
View file

@ -167,3 +167,18 @@ Target Var(nvptx_alias) Init(0) Undocumented
mexperimental
Target Var(nvptx_experimental) Init(0) Undocumented
mfake-ptx-alloca
Target Var(nvptx_fake_ptx_alloca) Init(0) Undocumented
; With '-mfake-ptx-alloca' enabled, the user-visible behavior changes only
; for configurations where PTX 'alloca' is not available. Rather than a
; compile-time 'sorry, unimplemented: dynamic stack allocation not supported'
; in presence of dynamic stack allocation, compilation and assembly then
; succeeds. However, attempting to link in such '*.o' files then fails due
; to unresolved symbol '__GCC_nvptx__PTX_alloca_not_supported'.
;
; This is meant to be used in scenarios where large volumes of code are
; compiled, a small fraction of which runs into dynamic stack allocation, but
; these parts are not important for specific use cases, and we'd thus like the
; build to succeed, and error out just upon actual, very rare use of the
; offending '*.o' files.

49
gcc/testsuite/gcc.target/nvptx/alloca-1-O0_-mfake-ptx-alloca.c Normal file
View file

@ -0,0 +1,49 @@
/* { dg-do assemble } */
/* { dg-options {-O0 -mno-soft-stack} } */
/* { dg-additional-options -march=sm_30 } */
/* { dg-additional-options -mfake-ptx-alloca } */
/* { dg-additional-options -save-temps } */
/* { dg-final { check-function-bodies {** } {} } } */
void sink(void *);
void f(void)
{
sink(__builtin_alloca(123));
/* { dg-bogus {sorry, unimplemented: dynamic stack allocation not supported} {} { target *-*-* } .-1 } */
}
/*
** f:
** \.visible \.func f
** {
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** mov\.u64 \11, 16;
** add\.u64 \2, \11, -1;
** add\.u64 \3, \2, 123;
** div\.u64 \4, \3, 16;
** mul\.lo\.u64 \5, \4, 16;
** call \(\6\), __GCC_nvptx__PTX_alloca_not_supported, \(\5\);
** add\.u64 \7, \6, 15;
** shr\.u64 \8, \7, 4;
** shl\.b64 \9, \8, 4;
** mov\.u64 \1, \9;
** mov\.u64 \10, \1;
** {
** \.param\.u64 %out_arg1;
** st\.param\.u64 \[%out_arg1\], \10;
** call sink, \(%out_arg1\);
** }
** ret;
*/
/* { dg-final { scan-assembler-times {(?n)^\.extern \.func \(\.param\.u64 %value_out\) __GCC_nvptx__PTX_alloca_not_supported \(\.param\.u64 %in_ar0\);$} 1 } } */

18
gcc/testsuite/gcc.target/nvptx/alloca-2-O0_-mfake-ptx-alloca.c Normal file
View file

@ -0,0 +1,18 @@
/* { dg-do link } */
/* { dg-options {-O0 -mno-soft-stack} } */
/* { dg-additional-options -march=sm_30 } */
/* { dg-additional-options -mfake-ptx-alloca } */
/* { dg-additional-options -save-temps } */
int
main(void)
{
return !(__builtin_alloca(100) != __builtin_alloca(10));
}
/* { dg-final { scan-assembler-times {(?n)\tcall\t\(%r[0-9]+\), __GCC_nvptx__PTX_alloca_not_supported, \(%r[0-9]+\);$} 2 } } */
/* { dg-final { scan-assembler-times {(?n)^\.extern \.func \(\.param\.u64 %value_out\) __GCC_nvptx__PTX_alloca_not_supported \(\.param\.u64 %in_ar0\);$} 1 } } */
/* { dg-message __GCC_nvptx__PTX_alloca_not_supported {unresolved symbol} { target *-*-* } 0 } */
/* { dg-final output-exists-not } */

2
gcc/testsuite/gcc.target/nvptx/alloca-4-O3.c
View file

@ -1,6 +1,8 @@
/* { dg-do assemble } */
/* { dg-options {-O3 -mno-soft-stack} } */
/* { dg-add-options nvptx_alloca_ptx } */
/* Verify the fake one isn't used if the real PTX 'alloca' is available.
{ dg-additional-options -mfake-ptx-alloca } */
/* { dg-additional-options -save-temps } */
/* { dg-final { check-function-bodies {** } {} } } */

48
gcc/testsuite/gcc.target/nvptx/alloca-4-O3_-mfake-ptx-alloca.c Normal file
View file

@ -0,0 +1,48 @@
/* { dg-do assemble } */
/* { dg-options {-O3 -mno-soft-stack} } */
/* { dg-additional-options {-march=sm_30 -mfake-ptx-alloca} } */
/* { dg-additional-options -save-temps } */
/* { dg-final { check-function-bodies {** } {} } } */
void sink(void *);
void f(void)
{
void *p;
p = __builtin_stack_save();
sink(__builtin_alloca(25));
/* { dg-bogus {sorry, unimplemented: dynamic stack allocation not supported} {} { target *-*-* } .-1 } */
__builtin_stack_restore(p);
sink(__builtin_alloca(13));
/* { dg-bogus {sorry, unimplemented: dynamic stack allocation not supported} {} { target *-*-* } .-1 } */
}
/*
** f:
** .visible .func f
** {
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** \.reg\.u64 (%r[0-9]+);
** call \(\1\), __GCC_nvptx__PTX_alloca_not_supported, \(32\);
** add\.u64 \2, \1, 15;
** and\.b64 \3, \2, -16;
** {
** \.param\.u64 %out_arg1;
** st\.param\.u64 \[%out_arg1\], \3;
** call sink, \(%out_arg1\);
** }
** call \(\4\), __GCC_nvptx__PTX_alloca_not_supported, \(16\);
** add\.u64 \5, \4, 15;
** and\.b64 \6, \5, -16;
** {
** \.param\.u64 %out_arg1;
** st\.param\.u64 \[%out_arg1\], \6;
** call sink, \(%out_arg1\);
** }
** ret;
*/
/* { dg-final { scan-assembler-times {(?n)^\.extern \.func \(\.param\.u64 %value_out\) __GCC_nvptx__PTX_alloca_not_supported \(\.param\.u64 %in_ar0\);$} 1 } } */

29
gcc/testsuite/gcc.target/nvptx/vla-1-O0_-mfake-ptx-alloca.c Normal file
View file

@ -0,0 +1,29 @@
/* { dg-do assemble } */
/* { dg-options {-O0 -mno-soft-stack} } */
/* { dg-additional-options -march=sm_30 } */
/* { dg-additional-options -mfake-ptx-alloca } */
/* { dg-additional-options -save-temps } */
/* { dg-final { check-function-bodies {** } {} } } */
void sink(void *);
void f(int s)
{
char a[s];
/* { dg-bogus {sorry, unimplemented: dynamic stack allocation not supported} {} { target *-*-* } .-1 } */
sink(a);
}
/*
** f:
** ...
** cvt\.s64\.s32 (%r[0-9]+), (%r[0-9]+);
** mov\.u64 (%r[0-9]+), 16;
** add\.u64 (%r[0-9]+), \3, -1;
** add\.u64 (%r[0-9]+), \1, \4;
** div\.u64 (%r[0-9]+), \5, 16;
** mul\.lo\.u64 (%r[0-9]+), \6, 16;
** call \((%r[0-9]+)\), __GCC_nvptx__PTX_alloca_not_supported, \(\7\);
** ...
*/
/* { dg-final { scan-assembler-times {(?n)^\.extern \.func \(\.param\.u64 %value_out\) __GCC_nvptx__PTX_alloca_not_supported \(\.param\.u64 %in_ar0\);$} 1 } } */