[v3] rs6000: Fix some issues in rs6000_can_inline_p [PR102059]

Message ID	85f65aa7-84d3-0e57-ccfb-b95aa24bbe8b@linux.ibm.com
State	New
Headers	show Return-Path: <gcc-patches-bounces+incoming=patchwork.ozlabs.org@gcc.gnu.org> DMARC-Filter: OpenDMARC Filter v1.4.1 sourceware.org 978EE3858C2C To: GCC Patches <gcc-patches@gcc.gnu.org> Subject: [PATCH v3] rs6000: Fix some issues in rs6000_can_inline_p [PR102059] Message-ID: <85f65aa7-84d3-0e57-ccfb-b95aa24bbe8b@linux.ibm.com> Date: Wed, 5 Jan 2022 15:34:41 +0800 User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:78.0) Gecko/20100101 Thunderbird/78.10.0 Content-Type: multipart/mixed; boundary="------------8692737A988F293C4BFF955E" Content-Language: en-US MIME-Version: 1.0 Precedence: list From: "Kewen.Lin via Gcc-patches" <gcc-patches@gcc.gnu.org> Reply-To: "Kewen.Lin" <linkw@linux.ibm.com> Cc: Michael Meissner <meissner@linux.ibm.com>, Peter Bergner <bergner@linux.ibm.com>, Bill Schmidt <wschmidt@linux.ibm.com>, David Edelsohn <dje.gcc@gmail.com>, Segher Boessenkool <segher@kernel.crashing.org> Errors-To: gcc-patches-bounces+incoming=patchwork.ozlabs.org@gcc.gnu.org Sender: "Gcc-patches" <gcc-patches-bounces+incoming=patchwork.ozlabs.org@gcc.gnu.org>
Series	[v3] rs6000: Fix some issues in rs6000_can_inline_p [PR102059] \| expand [v3] rs6000: Fix some issues in rs6000_can_inline_p [PR102059]

diff --git a/gcc/config/rs6000/rs6000.c b/gcc/config/rs6000/rs6000.c index 7d07b47d9e3..60e131f2191 100644 --- a/gcc/config/rs6000/rs6000.c +++ b/gcc/config/rs6000/rs6000.c @@ -25379,55 +25379,87 @@ rs6000_update_ipa_fn_target_info (unsigned int &info, const gimple *stmt) static bool rs6000_can_inline_p (tree caller, tree callee) { - bool ret = false; tree caller_tree = DECL_FUNCTION_SPECIFIC_TARGET (caller); tree callee_tree = DECL_FUNCTION_SPECIFIC_TARGET (callee); - /* If the callee has no option attributes, then it is ok to inline. */ + /* If the caller/callee has option attributes, then use them. + Otherwise, use the command line options. */ if (!callee_tree) - ret = true; - - else - { - HOST_WIDE_INT caller_isa; - struct cl_target_option *callee_opts = TREE_TARGET_OPTION (callee_tree); - HOST_WIDE_INT callee_isa = callee_opts->x_rs6000_isa_flags; - HOST_WIDE_INT explicit_isa = callee_opts->x_rs6000_isa_flags_explicit; + callee_tree = target_option_default_node; + if (!caller_tree) + caller_tree = target_option_default_node; + + struct cl_target_option *caller_opts = TREE_TARGET_OPTION (caller_tree); + struct cl_target_option *callee_opts = TREE_TARGET_OPTION (callee_tree); + HOST_WIDE_INT caller_isa = caller_opts->x_rs6000_isa_flags; + HOST_WIDE_INT callee_isa = callee_opts->x_rs6000_isa_flags; + + bool always_inline + = DECL_DISREGARD_INLINE_LIMITS (callee) + && lookup_attribute ("always_inline", DECL_ATTRIBUTES (callee)); + + /* Some features can be tolerated for always inlines. */ + unsigned HOST_WIDE_INT always_inline_safe_mask + /* Fusion option masks. */ + = OPTION_MASK_P8_FUSION | OPTION_MASK_P10_FUSION + | OPTION_MASK_P8_FUSION_SIGN | OPTION_MASK_P10_FUSION + | OPTION_MASK_P10_FUSION_LD_CMPI | OPTION_MASK_P10_FUSION_2LOGICAL + | OPTION_MASK_P10_FUSION_LOGADD | OPTION_MASK_P10_FUSION_ADDLOG + | OPTION_MASK_P10_FUSION_2ADD + /* Like fusion, some option masks which are just for optimization. */ + | OPTION_MASK_SAVE_TOC_INDIRECT | OPTION_MASK_PCREL_OPT; + + /* Some features are totally safe for inlining (or always inlines), + let's exclude them from the following checkings. */ + HOST_WIDE_INT safe_mask = always_inline ? always_inline_safe_mask : 0; + cgraph_node *callee_node = cgraph_node::get (callee); + if (ipa_fn_summaries && ipa_fn_summaries->get (callee_node) != NULL) + { + unsigned int info = ipa_fn_summaries->get (callee_node)->target_info; + if ((info & RS6000_FN_TARGET_INFO_HTM) == 0) + safe_mask |= OPTION_MASK_HTM; + } + + callee_isa &= ~safe_mask; + + /* The callee's options must be a subset of the caller's options, i.e. + a vsx function may inline an altivec function, but a no-vsx function + must not inline a vsx function. */ + if ((caller_isa & callee_isa) != callee_isa) + { + if (TARGET_DEBUG_TARGET) + fprintf (stderr, + "rs6000_can_inline_p:, caller %s, callee %s, cannot " + "inline since callee's options set isn't a subset of " + "caller's options set.\n", + get_decl_name (caller), get_decl_name (callee)); + return false; + } - /* If the caller has option attributes, then use them. - Otherwise, use the command line options. */ - if (caller_tree) - caller_isa = TREE_TARGET_OPTION (caller_tree)->x_rs6000_isa_flags; - else - caller_isa = rs6000_isa_flags; + /* For those options that the callee has explicitly enabled or disabled, + then we must enforce that the callee's and caller's options match + exactly; see PR70010. */ + HOST_WIDE_INT explicit_isa = callee_opts->x_rs6000_isa_flags_explicit; + explicit_isa &= ~safe_mask; - cgraph_node *callee_node = cgraph_node::get (callee); - if (ipa_fn_summaries && ipa_fn_summaries->get (callee_node) != NULL) - { - unsigned int info = ipa_fn_summaries->get (callee_node)->target_info; - if ((info & RS6000_FN_TARGET_INFO_HTM) == 0) - { - callee_isa &= ~OPTION_MASK_HTM; - explicit_isa &= ~OPTION_MASK_HTM; - } - } - - /* The callee's options must be a subset of the caller's options, i.e. - a vsx function may inline an altivec function, but a no-vsx function - must not inline a vsx function. However, for those options that the - callee has explicitly enabled or disabled, then we must enforce that - the callee's and caller's options match exactly; see PR70010. */ - if (((caller_isa & callee_isa) == callee_isa) - && (caller_isa & explicit_isa) == (callee_isa & explicit_isa)) - ret = true; + if ((caller_isa & explicit_isa) != (callee_isa & explicit_isa)) + { + if (TARGET_DEBUG_TARGET) + fprintf (stderr, + "rs6000_can_inline_p:, caller %s, callee %s, cannot " + "inline since callee's options set isn't a subset of " + "caller's options set by considering callee's " + "explicitly set options.\n", + get_decl_name (caller), get_decl_name (callee)); + return false; } if (TARGET_DEBUG_TARGET) - fprintf (stderr, "rs6000_can_inline_p:, caller %s, callee %s, %s inline\n", - get_decl_name (caller), get_decl_name (callee), - (ret ? "can" : "cannot")); + fprintf (stderr, + "rs6000_can_inline_p:, caller %s, callee %s, can inline.\n", + get_decl_name (caller), get_decl_name (callee)); - return ret; + return true; } /* Allocate a stack temp and fixup the address so it meets the particular diff --git a/gcc/testsuite/gcc.dg/lto/pr102059-1_0.c b/gcc/testsuite/gcc.dg/lto/pr102059-1_0.c index e1004f5cfbf..b215b701097 100644 --- a/gcc/testsuite/gcc.dg/lto/pr102059-1_0.c +++ b/gcc/testsuite/gcc.dg/lto/pr102059-1_0.c @@ -1,7 +1,7 @@ /* { dg-lto-do link } */ /* { dg-skip-if "power10 and above only" { ! { power10_ok } } } */ /* -Wno-attributes suppresses always_inline warnings. */ -/* { dg-lto-options { "-O2 -mdejagnu-cpu=power8 -flto -Wno-attributes -mno-power8-fusion" } } */ +/* { dg-lto-options { "-O2 -mdejagnu-cpu=power8 -flto -Wno-attributes" } } */ int __attribute__ ((always_inline)) foo1 (int *b) diff --git a/gcc/testsuite/gcc.target/powerpc/pr102059-4.c b/gcc/testsuite/gcc.target/powerpc/pr102059-4.c new file mode 100644 index 00000000000..d2a002cf141 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/pr102059-4.c @@ -0,0 +1,24 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target powerpc_htm_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power8" } */ + +/* Verify it emits inlining error msg at non-LTO mode. */ + +#include <htmintrin.h> + +static inline int __attribute__ ((always_inline)) +foo (int *b) /* { dg-error "inlining failed in call to .* target specific option mismatch" } */ +{ + int res = _HTM_STATE(__builtin_ttest()); + *b += res; + return *b; +} + +#pragma GCC target "cpu=power10" +int +bar (int *a) +{ + *a = foo (a); /* { dg-message "called from here" } */ + return 0; +} + diff --git a/gcc/testsuite/gcc.target/powerpc/pr102059-5.c b/gcc/testsuite/gcc.target/powerpc/pr102059-5.c new file mode 100644 index 00000000000..1d5d6c38bf3 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/pr102059-5.c @@ -0,0 +1,20 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -mno-vsx" } */ + +/* Verify it emits inlining error msg when the callee has explicit + disabling option from command line. */ + +vector int c, a, b; + +static inline void __attribute__ ((__always_inline__)) +foo () /* { dg-error "inlining failed in call to .* target specific option mismatch" } */ +{ + c = a + b; +} + +__attribute__ ((target ("vsx"))) +int main () +{ + foo (); /* { dg-message "called from here" } */ + c = a + b; +} diff --git a/gcc/testsuite/gcc.target/powerpc/pr102059-6.c b/gcc/testsuite/gcc.target/powerpc/pr102059-6.c new file mode 100644 index 00000000000..9684cab986a --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/pr102059-6.c @@ -0,0 +1,95 @@ +/* { dg-do compile } */ +/* -Wno-attributes suppresses always_inline warnings. */ +/* { dg-options "-O2 -mdejagnu-cpu=power8 -Wno-attributes" } */ + +/* Verify it doesn't emit inlining error msg since some mismatched + features are considered as safe for always_inline. */ + +/* 1. Callee enables Power8 fusion implicitly, while caller + with Power9 doesn't support power8 fusion at all. */ + +__attribute__ ((always_inline)) +int callee1 (int *b) +{ + *b += 1; + return *b; +} + +#pragma GCC target "cpu=power9" +int caller1 (int *a) +{ + *a = callee1 (a); + return 0; +} + +/* 2. Caller enables indirect toc save feature while callee + disables it explicitly. */ + +#pragma GCC target "save-toc-indirect" +__attribute__ ((always_inline)) +int callee2 (int *b) +{ + *b += 2; + return *b; +} + +#pragma GCC target "no-save-toc-indirect" +int caller2 (int *a) +{ + *a = callee2 (a); + return 0; +} + +/* 3. Caller disables Power10 fusion explicitly, while callee + still supports it as Power10 turns it on by default. */ + +#pragma GCC target "cpu=power10" +__attribute__ ((always_inline)) +int callee3 (int *b) +{ + *b += 3; + return *b; +} + +#pragma GCC target "cpu=power10,no-power10-fusion" +int caller3 (int *a) +{ + *a = callee3 (a); + return 0; +} + +/* 4. Caller enables Power10 fusion implicitly, while callee + disables it explicitly. */ + +#pragma GCC target "no-power10-fusion" +__attribute__ ((always_inline)) +int callee4 (int *b) +{ + *b += 4; + return *b; +} + +#pragma GCC target "cpu=power10" +int caller4 (int *a) +{ + *a = callee4 (a); + return 0; +} + +/* 5. Caller disables pcrel-opt while callee enables it explicitly. */ + +#pragma GCC target "cpu=power10,no-pcrel-opt" +__attribute__ ((always_inline)) +int callee5 (int *b) +{ + *b += 5; + return *b; +} + +#pragma GCC target "cpu=power10,pcrel-opt" +int caller5 (int *a) +{ + *a = callee5 (a); + return 0; +} + diff --git a/gcc/testsuite/gcc.target/powerpc/pr102059-7.c b/gcc/testsuite/gcc.target/powerpc/pr102059-7.c new file mode 100644 index 00000000000..0f27f2ce7d7 --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/pr102059-7.c @@ -0,0 +1,22 @@ +/* { dg-do compile } */ +/* -Wno-attributes suppresses always_inline warnings. */ +/* { dg-options "-O2 -mdejagnu-cpu=power8 -mno-power8-fusion -Wno-attributes" } */ + +/* Verify it doesn't emit inlining error msg since the flag power8 + fusion is considered as safe for always_inline, it's still safe + even the flag is set explicitly. */ + +__attribute__ ((always_inline)) +int foo (int *b) +{ + *b += 10; + return *b; +} + +#pragma GCC target "power8-fusion" +int bar (int *a) +{ + *a = foo (a); + return 0; +} + diff --git a/gcc/testsuite/gcc.target/powerpc/pr102059-8.c b/gcc/testsuite/gcc.target/powerpc/pr102059-8.c new file mode 100644 index 00000000000..826ce88025f --- /dev/null +++ b/gcc/testsuite/gcc.target/powerpc/pr102059-8.c @@ -0,0 +1,22 @@ +/* { dg-do compile } */ +/* { dg-require-effective-target powerpc_htm_ok } */ +/* { dg-options "-O2 -mdejagnu-cpu=power8 -mhtm" } */ + +/* Verify the inlining won't perform when the callee requires + some target feature which isn't supported by caller, even + if the callee doesn't have any target attributes or pragmas. + If the inlining performs here, the compilation will fail. */ + +int +foo (int *b) +{ + *b += __builtin_ttest (); + return *b; +} + +__attribute__ ((target ("no-htm"))) int +bar (int *a) +{ + *a = foo (a); + return 0; +}

[v3] rs6000: Fix some issues in rs6000_can_inline_p [PR102059]

Commit Message

Comments

Patch