* Short-cut generation of simple built-in functions @ 2015-11-07 13:31 Richard Sandiford 2015-11-10 11:29 ` Richard Biener 2015-11-22 22:19 ` H.J. Lu 0 siblings, 2 replies; 8+ messages in thread From: Richard Sandiford @ 2015-11-07 13:31 UTC (permalink / raw) To: gcc-patches This patch short-circuits the builtins.c expansion code for a particular gimple call if: - the function has an associated internal function - the target implements that internal function - the call has no side effects This allows a later patch to remove the builtins.c code, once calls with side effects have been handled. Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. OK to install? Thanks, Richard gcc/ * builtins.h (called_as_built_in): Declare. * builtins.c (called_as_built_in): Make external. * internal-fn.h (expand_internal_call): Define a variant that specifies the internal function explicitly. * internal-fn.c (expand_load_lanes_optab_fn) (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) (expand_mask_load_optab_fn, expand_mask_store_optab_fn) (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn argument. (internal_fn_expanders): Update prototype. (expand_internal_call): Define a variant that specifies the internal function explicitly. Use it to implement the previous interface. * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in functions as calls to internal functions. diff --git a/gcc/builtins.c b/gcc/builtins.c index f65011e..bbcc7dc3 100644 --- a/gcc/builtins.c +++ b/gcc/builtins.c @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) of the optimization level. This means whenever a function is invoked with its "internal" name, which normally contains the prefix "__builtin". */ -static bool +bool called_as_built_in (tree node) { /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since diff --git a/gcc/builtins.h b/gcc/builtins.h index 917eb90..1d00068 100644 --- a/gcc/builtins.h +++ b/gcc/builtins.h @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; extern bool force_folding_builtin_constant_p; extern bool is_builtin_fn (tree); +extern bool called_as_built_in (tree); extern bool get_object_alignment_1 (tree, unsigned int *, unsigned HOST_WIDE_INT *); extern unsigned int get_object_alignment (tree); diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c index bfbc958..dc7d4f5 100644 --- a/gcc/cfgexpand.c +++ b/gcc/cfgexpand.c @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) return; } + /* If this is a call to a built-in function and it has no effect other + than setting the lhs, try to implement it using an internal function + instead. */ + decl = gimple_call_fndecl (stmt); + if (gimple_call_lhs (stmt) + && !gimple_vdef (stmt) + && (optimize || (decl && called_as_built_in (decl)))) + { + internal_fn ifn = replacement_internal_fn (stmt); + if (ifn != IFN_LAST) + { + expand_internal_call (ifn, stmt); + return; + } + } + exp = build_vl_exp (CALL_EXPR, gimple_call_num_args (stmt) + 3); CALL_EXPR_FN (exp) = gimple_call_fn (stmt); - decl = gimple_call_fndecl (stmt); builtin_p = decl && DECL_BUILT_IN (decl); /* If this is not a builtin function, the function type through which the diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c index 9f9f9cf..c03c0fc 100644 --- a/gcc/internal-fn.c +++ b/gcc/internal-fn.c @@ -103,7 +103,7 @@ get_multi_vector_move (tree array_type, convert_optab optab) /* Expand LOAD_LANES call STMT using optab OPTAB. */ static void -expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) +expand_load_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) { struct expand_operand ops[2]; tree type, lhs, rhs; @@ -127,7 +127,7 @@ expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) /* Expand STORE_LANES call STMT using optab OPTAB. */ static void -expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) +expand_store_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) { struct expand_operand ops[2]; tree type, lhs, rhs; @@ -149,7 +149,7 @@ expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) } static void -expand_ANNOTATE (gcall *) +expand_ANNOTATE (internal_fn, gcall *) { gcc_unreachable (); } @@ -157,7 +157,7 @@ expand_ANNOTATE (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_LANE (gcall *) +expand_GOMP_SIMD_LANE (internal_fn, gcall *) { gcc_unreachable (); } @@ -165,7 +165,7 @@ expand_GOMP_SIMD_LANE (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_VF (gcall *) +expand_GOMP_SIMD_VF (internal_fn, gcall *) { gcc_unreachable (); } @@ -173,7 +173,7 @@ expand_GOMP_SIMD_VF (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_LAST_LANE (gcall *) +expand_GOMP_SIMD_LAST_LANE (internal_fn, gcall *) { gcc_unreachable (); } @@ -181,7 +181,7 @@ expand_GOMP_SIMD_LAST_LANE (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_ORDERED_START (gcall *) +expand_GOMP_SIMD_ORDERED_START (internal_fn, gcall *) { gcc_unreachable (); } @@ -189,7 +189,7 @@ expand_GOMP_SIMD_ORDERED_START (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_ORDERED_END (gcall *) +expand_GOMP_SIMD_ORDERED_END (internal_fn, gcall *) { gcc_unreachable (); } @@ -197,7 +197,7 @@ expand_GOMP_SIMD_ORDERED_END (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_NULL (gcall *) +expand_UBSAN_NULL (internal_fn, gcall *) { gcc_unreachable (); } @@ -205,7 +205,7 @@ expand_UBSAN_NULL (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_BOUNDS (gcall *) +expand_UBSAN_BOUNDS (internal_fn, gcall *) { gcc_unreachable (); } @@ -213,7 +213,7 @@ expand_UBSAN_BOUNDS (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_VPTR (gcall *) +expand_UBSAN_VPTR (internal_fn, gcall *) { gcc_unreachable (); } @@ -221,7 +221,7 @@ expand_UBSAN_VPTR (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_OBJECT_SIZE (gcall *) +expand_UBSAN_OBJECT_SIZE (internal_fn, gcall *) { gcc_unreachable (); } @@ -229,7 +229,7 @@ expand_UBSAN_OBJECT_SIZE (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_ASAN_CHECK (gcall *) +expand_ASAN_CHECK (internal_fn, gcall *) { gcc_unreachable (); } @@ -237,7 +237,7 @@ expand_ASAN_CHECK (gcall *) /* This should get expanded in the tsan pass. */ static void -expand_TSAN_FUNC_EXIT (gcall *) +expand_TSAN_FUNC_EXIT (internal_fn, gcall *) { gcc_unreachable (); } @@ -1639,7 +1639,7 @@ expand_mul_overflow (location_t loc, tree lhs, tree arg0, tree arg1, /* Expand UBSAN_CHECK_ADD call STMT. */ static void -expand_UBSAN_CHECK_ADD (gcall *stmt) +expand_UBSAN_CHECK_ADD (internal_fn, gcall *stmt) { location_t loc = gimple_location (stmt); tree lhs = gimple_call_lhs (stmt); @@ -1652,7 +1652,7 @@ expand_UBSAN_CHECK_ADD (gcall *stmt) /* Expand UBSAN_CHECK_SUB call STMT. */ static void -expand_UBSAN_CHECK_SUB (gcall *stmt) +expand_UBSAN_CHECK_SUB (internal_fn, gcall *stmt) { location_t loc = gimple_location (stmt); tree lhs = gimple_call_lhs (stmt); @@ -1668,7 +1668,7 @@ expand_UBSAN_CHECK_SUB (gcall *stmt) /* Expand UBSAN_CHECK_MUL call STMT. */ static void -expand_UBSAN_CHECK_MUL (gcall *stmt) +expand_UBSAN_CHECK_MUL (internal_fn, gcall *stmt) { location_t loc = gimple_location (stmt); tree lhs = gimple_call_lhs (stmt); @@ -1853,7 +1853,7 @@ expand_arith_overflow (enum tree_code code, gimple *stmt) /* Expand ADD_OVERFLOW STMT. */ static void -expand_ADD_OVERFLOW (gcall *stmt) +expand_ADD_OVERFLOW (internal_fn, gcall *stmt) { expand_arith_overflow (PLUS_EXPR, stmt); } @@ -1861,7 +1861,7 @@ expand_ADD_OVERFLOW (gcall *stmt) /* Expand SUB_OVERFLOW STMT. */ static void -expand_SUB_OVERFLOW (gcall *stmt) +expand_SUB_OVERFLOW (internal_fn, gcall *stmt) { expand_arith_overflow (MINUS_EXPR, stmt); } @@ -1869,7 +1869,7 @@ expand_SUB_OVERFLOW (gcall *stmt) /* Expand MUL_OVERFLOW STMT. */ static void -expand_MUL_OVERFLOW (gcall *stmt) +expand_MUL_OVERFLOW (internal_fn, gcall *stmt) { expand_arith_overflow (MULT_EXPR, stmt); } @@ -1877,7 +1877,7 @@ expand_MUL_OVERFLOW (gcall *stmt) /* This should get folded in tree-vectorizer.c. */ static void -expand_LOOP_VECTORIZED (gcall *) +expand_LOOP_VECTORIZED (internal_fn, gcall *) { gcc_unreachable (); } @@ -1885,7 +1885,7 @@ expand_LOOP_VECTORIZED (gcall *) /* Expand MASK_LOAD call STMT using optab OPTAB. */ static void -expand_mask_load_optab_fn (gcall *stmt, direct_optab optab) +expand_mask_load_optab_fn (internal_fn, gcall *stmt, direct_optab optab) { struct expand_operand ops[3]; tree type, lhs, rhs, maskt; @@ -1912,7 +1912,7 @@ expand_mask_load_optab_fn (gcall *stmt, direct_optab optab) /* Expand MASK_STORE call STMT using optab OPTAB. */ static void -expand_mask_store_optab_fn (gcall *stmt, direct_optab optab) +expand_mask_store_optab_fn (internal_fn, gcall *stmt, direct_optab optab) { struct expand_operand ops[3]; tree type, lhs, rhs, maskt; @@ -1935,12 +1935,12 @@ expand_mask_store_optab_fn (gcall *stmt, direct_optab optab) } static void -expand_ABNORMAL_DISPATCHER (gcall *) +expand_ABNORMAL_DISPATCHER (internal_fn, gcall *) { } static void -expand_BUILTIN_EXPECT (gcall *stmt) +expand_BUILTIN_EXPECT (internal_fn, gcall *stmt) { /* When guessing was done, the hints should be already stripped away. */ gcc_assert (!flag_guess_branch_prob || optimize == 0 || seen_error ()); @@ -1960,7 +1960,7 @@ expand_BUILTIN_EXPECT (gcall *stmt) should never be called. */ static void -expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) +expand_VA_ARG (internal_fn, gcall *) { gcc_unreachable (); } @@ -1968,7 +1968,7 @@ expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) /* Expand the IFN_UNIQUE function according to its first argument. */ static void -expand_UNIQUE (gcall *stmt) +expand_UNIQUE (internal_fn, gcall *stmt) { rtx pattern = NULL_RTX; enum ifn_unique_kind kind @@ -2014,7 +2014,7 @@ expand_UNIQUE (gcall *stmt) /* The size of an OpenACC compute dimension. */ static void -expand_GOACC_DIM_SIZE (gcall *stmt) +expand_GOACC_DIM_SIZE (internal_fn, gcall *stmt) { tree lhs = gimple_call_lhs (stmt); @@ -2035,7 +2035,7 @@ expand_GOACC_DIM_SIZE (gcall *stmt) /* The position of an OpenACC execution engine along one compute axis. */ static void -expand_GOACC_DIM_POS (gcall *stmt) +expand_GOACC_DIM_POS (internal_fn, gcall *stmt) { tree lhs = gimple_call_lhs (stmt); @@ -2056,7 +2056,7 @@ expand_GOACC_DIM_POS (gcall *stmt) /* This is expanded by oacc_device_lower pass. */ static void -expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) +expand_GOACC_LOOP (internal_fn, gcall *) { gcc_unreachable (); } @@ -2064,20 +2064,20 @@ expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) /* This is expanded by oacc_device_lower pass. */ static void -expand_GOACC_REDUCTION (gcall *stmt ATTRIBUTE_UNUSED) +expand_GOACC_REDUCTION (internal_fn, gcall *) { gcc_unreachable (); } -/* Expand call STMT using OPTAB, which has a single output operand and - NARGS input operands. */ +/* Expand a call to FN using the operands in STMT. FN has a single + output operand and NARGS input operands. */ static void -expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) +expand_direct_optab_fn (internal_fn fn, gcall *stmt, direct_optab optab, + unsigned int nargs) { expand_operand *ops = XALLOCAVEC (expand_operand, nargs + 1); - internal_fn fn = gimple_call_internal_fn (stmt); tree_pair types = direct_internal_fn_types (fn, stmt); insn_code icode = direct_optab_handler (optab, TYPE_MODE (types.first)); @@ -2115,11 +2115,11 @@ expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) /* Expanders for optabs that can use expand_direct_optab_fn. */ -#define expand_unary_optab_fn(STMT, OPTAB) \ - expand_direct_optab_fn (STMT, OPTAB, 1) +#define expand_unary_optab_fn(FN, STMT, OPTAB) \ + expand_direct_optab_fn (FN, STMT, OPTAB, 1) -#define expand_binary_optab_fn(STMT, OPTAB) \ - expand_direct_optab_fn (STMT, OPTAB, 2) +#define expand_binary_optab_fn(FN, STMT, OPTAB) \ + expand_direct_optab_fn (FN, STMT, OPTAB, 2) /* RETURN_TYPE and ARGS are a return type and argument list that are in principle compatible with FN (which satisfies direct_internal_fn_p). @@ -2215,9 +2215,9 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) #define DEF_INTERNAL_OPTAB_FN(CODE, FLAGS, OPTAB, TYPE) \ static void \ - expand_##CODE (gcall *stmt) \ + expand_##CODE (internal_fn fn, gcall *stmt) \ { \ - expand_##TYPE##_optab_fn (stmt, OPTAB##_optab); \ + expand_##TYPE##_optab_fn (fn, stmt, OPTAB##_optab); \ } #include "internal-fn.def" @@ -2227,16 +2227,24 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) expand_<NAME> (gcall *stmt) where STMT is the statement that performs the call. */ -static void (*const internal_fn_expanders[]) (gcall *) = { +static void (*const internal_fn_expanders[]) (internal_fn, gcall *) = { #define DEF_INTERNAL_FN(CODE, FLAGS, FNSPEC) expand_##CODE, #include "internal-fn.def" 0 }; +/* Expand STMT as though it were a call to internal function FN. */ + +void +expand_internal_call (internal_fn fn, gcall *stmt) +{ + internal_fn_expanders[fn] (fn, stmt); +} + /* Expand STMT, which is a call to internal function FN. */ void expand_internal_call (gcall *stmt) { - internal_fn_expanders[(int) gimple_call_internal_fn (stmt)] (stmt); + expand_internal_call (gimple_call_internal_fn (stmt), stmt); } diff --git a/gcc/internal-fn.h b/gcc/internal-fn.h index 31e895e..5ee43b8 100644 --- a/gcc/internal-fn.h +++ b/gcc/internal-fn.h @@ -162,5 +162,6 @@ extern bool direct_internal_fn_supported_p (internal_fn, tree_pair); extern bool direct_internal_fn_supported_p (internal_fn, tree); extern void expand_internal_call (gcall *); +extern void expand_internal_call (internal_fn, gcall *); #endif ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: Short-cut generation of simple built-in functions 2015-11-07 13:31 Short-cut generation of simple built-in functions Richard Sandiford @ 2015-11-10 11:29 ` Richard Biener 2015-11-10 21:24 ` Richard Sandiford 2015-11-22 22:19 ` H.J. Lu 1 sibling, 1 reply; 8+ messages in thread From: Richard Biener @ 2015-11-10 11:29 UTC (permalink / raw) To: GCC Patches, richard.sandiford On Sat, Nov 7, 2015 at 2:31 PM, Richard Sandiford <richard.sandiford@arm.com> wrote: > This patch short-circuits the builtins.c expansion code for a particular > gimple call if: > > - the function has an associated internal function > - the target implements that internal function > - the call has no side effects > > This allows a later patch to remove the builtins.c code, once calls with > side effects have been handled. > > Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. > OK to install? > > Thanks, > Richard > > > gcc/ > * builtins.h (called_as_built_in): Declare. > * builtins.c (called_as_built_in): Make external. > * internal-fn.h (expand_internal_call): Define a variant that > specifies the internal function explicitly. > * internal-fn.c (expand_load_lanes_optab_fn) > (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) > (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) > (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) > (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) > (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) > (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) > (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) > (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) > (expand_mask_load_optab_fn, expand_mask_store_optab_fn) > (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) > (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) > (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) > (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn > argument. > (internal_fn_expanders): Update prototype. > (expand_internal_call): Define a variant that specifies the > internal function explicitly. Use it to implement the previous > interface. > * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in > functions as calls to internal functions. > > diff --git a/gcc/builtins.c b/gcc/builtins.c > index f65011e..bbcc7dc3 100644 > --- a/gcc/builtins.c > +++ b/gcc/builtins.c > @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) > of the optimization level. This means whenever a function is invoked with > its "internal" name, which normally contains the prefix "__builtin". */ > > -static bool > +bool > called_as_built_in (tree node) > { > /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since > diff --git a/gcc/builtins.h b/gcc/builtins.h > index 917eb90..1d00068 100644 > --- a/gcc/builtins.h > +++ b/gcc/builtins.h > @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; > extern bool force_folding_builtin_constant_p; > > extern bool is_builtin_fn (tree); > +extern bool called_as_built_in (tree); > extern bool get_object_alignment_1 (tree, unsigned int *, > unsigned HOST_WIDE_INT *); > extern unsigned int get_object_alignment (tree); > diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c > index bfbc958..dc7d4f5 100644 > --- a/gcc/cfgexpand.c > +++ b/gcc/cfgexpand.c > @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) > return; > } > > + /* If this is a call to a built-in function and it has no effect other > + than setting the lhs, try to implement it using an internal function > + instead. */ > + decl = gimple_call_fndecl (stmt); > + if (gimple_call_lhs (stmt) > + && !gimple_vdef (stmt) I think you want && ! gimple_has_side_effects (stmt) instead of checking !gimple_vdef (stmt). > + && (optimize || (decl && called_as_built_in (decl)))) > + { > + internal_fn ifn = replacement_internal_fn (stmt); > + if (ifn != IFN_LAST) > + { > + expand_internal_call (ifn, stmt); > + return; > + } > + } > + > exp = build_vl_exp (CALL_EXPR, gimple_call_num_args (stmt) + 3); > > CALL_EXPR_FN (exp) = gimple_call_fn (stmt); > - decl = gimple_call_fndecl (stmt); > builtin_p = decl && DECL_BUILT_IN (decl); > > /* If this is not a builtin function, the function type through which the > diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c > index 9f9f9cf..c03c0fc 100644 > --- a/gcc/internal-fn.c > +++ b/gcc/internal-fn.c > @@ -103,7 +103,7 @@ get_multi_vector_move (tree array_type, convert_optab optab) > /* Expand LOAD_LANES call STMT using optab OPTAB. */ > > static void > -expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) > +expand_load_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) I'm somewhat confused by all these unused internal_fn args. I'm sure we can avoid them? > { > struct expand_operand ops[2]; > tree type, lhs, rhs; > @@ -127,7 +127,7 @@ expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) > /* Expand STORE_LANES call STMT using optab OPTAB. */ > > static void > -expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) > +expand_store_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) > { > struct expand_operand ops[2]; > tree type, lhs, rhs; > @@ -149,7 +149,7 @@ expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) > } > > static void > -expand_ANNOTATE (gcall *) > +expand_ANNOTATE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -157,7 +157,7 @@ expand_ANNOTATE (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_LANE (gcall *) > +expand_GOMP_SIMD_LANE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -165,7 +165,7 @@ expand_GOMP_SIMD_LANE (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_VF (gcall *) > +expand_GOMP_SIMD_VF (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -173,7 +173,7 @@ expand_GOMP_SIMD_VF (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_LAST_LANE (gcall *) > +expand_GOMP_SIMD_LAST_LANE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -181,7 +181,7 @@ expand_GOMP_SIMD_LAST_LANE (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_ORDERED_START (gcall *) > +expand_GOMP_SIMD_ORDERED_START (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -189,7 +189,7 @@ expand_GOMP_SIMD_ORDERED_START (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_ORDERED_END (gcall *) > +expand_GOMP_SIMD_ORDERED_END (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -197,7 +197,7 @@ expand_GOMP_SIMD_ORDERED_END (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_NULL (gcall *) > +expand_UBSAN_NULL (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -205,7 +205,7 @@ expand_UBSAN_NULL (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_BOUNDS (gcall *) > +expand_UBSAN_BOUNDS (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -213,7 +213,7 @@ expand_UBSAN_BOUNDS (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_VPTR (gcall *) > +expand_UBSAN_VPTR (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -221,7 +221,7 @@ expand_UBSAN_VPTR (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_OBJECT_SIZE (gcall *) > +expand_UBSAN_OBJECT_SIZE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -229,7 +229,7 @@ expand_UBSAN_OBJECT_SIZE (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_ASAN_CHECK (gcall *) > +expand_ASAN_CHECK (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -237,7 +237,7 @@ expand_ASAN_CHECK (gcall *) > /* This should get expanded in the tsan pass. */ > > static void > -expand_TSAN_FUNC_EXIT (gcall *) > +expand_TSAN_FUNC_EXIT (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -1639,7 +1639,7 @@ expand_mul_overflow (location_t loc, tree lhs, tree arg0, tree arg1, > /* Expand UBSAN_CHECK_ADD call STMT. */ > > static void > -expand_UBSAN_CHECK_ADD (gcall *stmt) > +expand_UBSAN_CHECK_ADD (internal_fn, gcall *stmt) > { > location_t loc = gimple_location (stmt); > tree lhs = gimple_call_lhs (stmt); > @@ -1652,7 +1652,7 @@ expand_UBSAN_CHECK_ADD (gcall *stmt) > /* Expand UBSAN_CHECK_SUB call STMT. */ > > static void > -expand_UBSAN_CHECK_SUB (gcall *stmt) > +expand_UBSAN_CHECK_SUB (internal_fn, gcall *stmt) > { > location_t loc = gimple_location (stmt); > tree lhs = gimple_call_lhs (stmt); > @@ -1668,7 +1668,7 @@ expand_UBSAN_CHECK_SUB (gcall *stmt) > /* Expand UBSAN_CHECK_MUL call STMT. */ > > static void > -expand_UBSAN_CHECK_MUL (gcall *stmt) > +expand_UBSAN_CHECK_MUL (internal_fn, gcall *stmt) > { > location_t loc = gimple_location (stmt); > tree lhs = gimple_call_lhs (stmt); > @@ -1853,7 +1853,7 @@ expand_arith_overflow (enum tree_code code, gimple *stmt) > /* Expand ADD_OVERFLOW STMT. */ > > static void > -expand_ADD_OVERFLOW (gcall *stmt) > +expand_ADD_OVERFLOW (internal_fn, gcall *stmt) > { > expand_arith_overflow (PLUS_EXPR, stmt); > } > @@ -1861,7 +1861,7 @@ expand_ADD_OVERFLOW (gcall *stmt) > /* Expand SUB_OVERFLOW STMT. */ > > static void > -expand_SUB_OVERFLOW (gcall *stmt) > +expand_SUB_OVERFLOW (internal_fn, gcall *stmt) > { > expand_arith_overflow (MINUS_EXPR, stmt); > } > @@ -1869,7 +1869,7 @@ expand_SUB_OVERFLOW (gcall *stmt) > /* Expand MUL_OVERFLOW STMT. */ > > static void > -expand_MUL_OVERFLOW (gcall *stmt) > +expand_MUL_OVERFLOW (internal_fn, gcall *stmt) > { > expand_arith_overflow (MULT_EXPR, stmt); > } > @@ -1877,7 +1877,7 @@ expand_MUL_OVERFLOW (gcall *stmt) > /* This should get folded in tree-vectorizer.c. */ > > static void > -expand_LOOP_VECTORIZED (gcall *) > +expand_LOOP_VECTORIZED (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -1885,7 +1885,7 @@ expand_LOOP_VECTORIZED (gcall *) > /* Expand MASK_LOAD call STMT using optab OPTAB. */ > > static void > -expand_mask_load_optab_fn (gcall *stmt, direct_optab optab) > +expand_mask_load_optab_fn (internal_fn, gcall *stmt, direct_optab optab) > { > struct expand_operand ops[3]; > tree type, lhs, rhs, maskt; > @@ -1912,7 +1912,7 @@ expand_mask_load_optab_fn (gcall *stmt, direct_optab optab) > /* Expand MASK_STORE call STMT using optab OPTAB. */ > > static void > -expand_mask_store_optab_fn (gcall *stmt, direct_optab optab) > +expand_mask_store_optab_fn (internal_fn, gcall *stmt, direct_optab optab) > { > struct expand_operand ops[3]; > tree type, lhs, rhs, maskt; > @@ -1935,12 +1935,12 @@ expand_mask_store_optab_fn (gcall *stmt, direct_optab optab) > } > > static void > -expand_ABNORMAL_DISPATCHER (gcall *) > +expand_ABNORMAL_DISPATCHER (internal_fn, gcall *) > { > } > > static void > -expand_BUILTIN_EXPECT (gcall *stmt) > +expand_BUILTIN_EXPECT (internal_fn, gcall *stmt) > { > /* When guessing was done, the hints should be already stripped away. */ > gcc_assert (!flag_guess_branch_prob || optimize == 0 || seen_error ()); > @@ -1960,7 +1960,7 @@ expand_BUILTIN_EXPECT (gcall *stmt) > should never be called. */ > > static void > -expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) > +expand_VA_ARG (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -1968,7 +1968,7 @@ expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) > /* Expand the IFN_UNIQUE function according to its first argument. */ > > static void > -expand_UNIQUE (gcall *stmt) > +expand_UNIQUE (internal_fn, gcall *stmt) > { > rtx pattern = NULL_RTX; > enum ifn_unique_kind kind > @@ -2014,7 +2014,7 @@ expand_UNIQUE (gcall *stmt) > /* The size of an OpenACC compute dimension. */ > > static void > -expand_GOACC_DIM_SIZE (gcall *stmt) > +expand_GOACC_DIM_SIZE (internal_fn, gcall *stmt) > { > tree lhs = gimple_call_lhs (stmt); > > @@ -2035,7 +2035,7 @@ expand_GOACC_DIM_SIZE (gcall *stmt) > /* The position of an OpenACC execution engine along one compute axis. */ > > static void > -expand_GOACC_DIM_POS (gcall *stmt) > +expand_GOACC_DIM_POS (internal_fn, gcall *stmt) > { > tree lhs = gimple_call_lhs (stmt); > > @@ -2056,7 +2056,7 @@ expand_GOACC_DIM_POS (gcall *stmt) > /* This is expanded by oacc_device_lower pass. */ > > static void > -expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) > +expand_GOACC_LOOP (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -2064,20 +2064,20 @@ expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) > /* This is expanded by oacc_device_lower pass. */ > > static void > -expand_GOACC_REDUCTION (gcall *stmt ATTRIBUTE_UNUSED) > +expand_GOACC_REDUCTION (internal_fn, gcall *) > { > gcc_unreachable (); > } > > -/* Expand call STMT using OPTAB, which has a single output operand and > - NARGS input operands. */ > +/* Expand a call to FN using the operands in STMT. FN has a single > + output operand and NARGS input operands. */ > > static void > -expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) > +expand_direct_optab_fn (internal_fn fn, gcall *stmt, direct_optab optab, > + unsigned int nargs) > { > expand_operand *ops = XALLOCAVEC (expand_operand, nargs + 1); > > - internal_fn fn = gimple_call_internal_fn (stmt); > tree_pair types = direct_internal_fn_types (fn, stmt); > insn_code icode = direct_optab_handler (optab, TYPE_MODE (types.first)); > > @@ -2115,11 +2115,11 @@ expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) > > /* Expanders for optabs that can use expand_direct_optab_fn. */ > > -#define expand_unary_optab_fn(STMT, OPTAB) \ > - expand_direct_optab_fn (STMT, OPTAB, 1) > +#define expand_unary_optab_fn(FN, STMT, OPTAB) \ > + expand_direct_optab_fn (FN, STMT, OPTAB, 1) > > -#define expand_binary_optab_fn(STMT, OPTAB) \ > - expand_direct_optab_fn (STMT, OPTAB, 2) > +#define expand_binary_optab_fn(FN, STMT, OPTAB) \ > + expand_direct_optab_fn (FN, STMT, OPTAB, 2) > > /* RETURN_TYPE and ARGS are a return type and argument list that are > in principle compatible with FN (which satisfies direct_internal_fn_p). > @@ -2215,9 +2215,9 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) > > #define DEF_INTERNAL_OPTAB_FN(CODE, FLAGS, OPTAB, TYPE) \ > static void \ > - expand_##CODE (gcall *stmt) \ > + expand_##CODE (internal_fn fn, gcall *stmt) \ > { \ > - expand_##TYPE##_optab_fn (stmt, OPTAB##_optab); \ > + expand_##TYPE##_optab_fn (fn, stmt, OPTAB##_optab); \ > } > #include "internal-fn.def" > > @@ -2227,16 +2227,24 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) > expand_<NAME> (gcall *stmt) > > where STMT is the statement that performs the call. */ > -static void (*const internal_fn_expanders[]) (gcall *) = { > +static void (*const internal_fn_expanders[]) (internal_fn, gcall *) = { > #define DEF_INTERNAL_FN(CODE, FLAGS, FNSPEC) expand_##CODE, > #include "internal-fn.def" > 0 > }; > > +/* Expand STMT as though it were a call to internal function FN. */ > + > +void > +expand_internal_call (internal_fn fn, gcall *stmt) > +{ > + internal_fn_expanders[fn] (fn, stmt); That is, just don't pass 'fn' here? Or do we ever expect to have the same internal_fn_expander for different fn? Otherwise looks ok to me. Richard. > +} > + > /* Expand STMT, which is a call to internal function FN. */ > > void > expand_internal_call (gcall *stmt) > { > - internal_fn_expanders[(int) gimple_call_internal_fn (stmt)] (stmt); > + expand_internal_call (gimple_call_internal_fn (stmt), stmt); > } > diff --git a/gcc/internal-fn.h b/gcc/internal-fn.h > index 31e895e..5ee43b8 100644 > --- a/gcc/internal-fn.h > +++ b/gcc/internal-fn.h > @@ -162,5 +162,6 @@ extern bool direct_internal_fn_supported_p (internal_fn, tree_pair); > extern bool direct_internal_fn_supported_p (internal_fn, tree); > > extern void expand_internal_call (gcall *); > +extern void expand_internal_call (internal_fn, gcall *); > > #endif > ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: Short-cut generation of simple built-in functions 2015-11-10 11:29 ` Richard Biener @ 2015-11-10 21:24 ` Richard Sandiford 2015-11-11 10:18 ` Richard Biener 0 siblings, 1 reply; 8+ messages in thread From: Richard Sandiford @ 2015-11-10 21:24 UTC (permalink / raw) To: Richard Biener; +Cc: GCC Patches Richard Biener <richard.guenther@gmail.com> writes: > On Sat, Nov 7, 2015 at 2:31 PM, Richard Sandiford > <richard.sandiford@arm.com> wrote: >> This patch short-circuits the builtins.c expansion code for a particular >> gimple call if: >> >> - the function has an associated internal function >> - the target implements that internal function >> - the call has no side effects >> >> This allows a later patch to remove the builtins.c code, once calls with >> side effects have been handled. >> >> Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. >> OK to install? >> >> Thanks, >> Richard >> >> >> gcc/ >> * builtins.h (called_as_built_in): Declare. >> * builtins.c (called_as_built_in): Make external. >> * internal-fn.h (expand_internal_call): Define a variant that >> specifies the internal function explicitly. >> * internal-fn.c (expand_load_lanes_optab_fn) >> (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) >> (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) >> (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) >> (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) >> (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) >> (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) >> (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) >> (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) >> (expand_mask_load_optab_fn, expand_mask_store_optab_fn) >> (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) >> (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) >> (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) >> (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn >> argument. >> (internal_fn_expanders): Update prototype. >> (expand_internal_call): Define a variant that specifies the >> internal function explicitly. Use it to implement the previous >> interface. >> * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in >> functions as calls to internal functions. >> >> diff --git a/gcc/builtins.c b/gcc/builtins.c >> index f65011e..bbcc7dc3 100644 >> --- a/gcc/builtins.c >> +++ b/gcc/builtins.c >> @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) >> of the optimization level. This means whenever a function is invoked with >> its "internal" name, which normally contains the prefix "__builtin". */ >> >> -static bool >> +bool >> called_as_built_in (tree node) >> { >> /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since >> diff --git a/gcc/builtins.h b/gcc/builtins.h >> index 917eb90..1d00068 100644 >> --- a/gcc/builtins.h >> +++ b/gcc/builtins.h >> @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; >> extern bool force_folding_builtin_constant_p; >> >> extern bool is_builtin_fn (tree); >> +extern bool called_as_built_in (tree); >> extern bool get_object_alignment_1 (tree, unsigned int *, >> unsigned HOST_WIDE_INT *); >> extern unsigned int get_object_alignment (tree); >> diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c >> index bfbc958..dc7d4f5 100644 >> --- a/gcc/cfgexpand.c >> +++ b/gcc/cfgexpand.c >> @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) >> return; >> } >> >> + /* If this is a call to a built-in function and it has no effect other >> + than setting the lhs, try to implement it using an internal function >> + instead. */ >> + decl = gimple_call_fndecl (stmt); >> + if (gimple_call_lhs (stmt) >> + && !gimple_vdef (stmt) > > I think you want && ! gimple_has_side_effects (stmt) > instead of checking !gimple_vdef (stmt). OK, I can do that, but what would the difference be in practice for these types of call? I.e. are there cases for built-ins where: (A) gimple_vdef (stmt) && !gimple_side_effects (stmt) or: (B) !gimple_vdef (stmt) && gimple_side_effects (stmt) ? It just seems like this check should be the opposite of the one used in the call-cdce patch (when deciding whether to optimise a call with an lhs). In order to keep them in sync I'd need to use gimple_side_effects rather than gimple_vdef there too, but is (B) a possibility there? >> + && (optimize || (decl && called_as_built_in (decl)))) >> + { >> + internal_fn ifn = replacement_internal_fn (stmt); >> + if (ifn != IFN_LAST) >> + { >> + expand_internal_call (ifn, stmt); >> + return; >> + } >> + } >> + >> exp = build_vl_exp (CALL_EXPR, gimple_call_num_args (stmt) + 3); >> >> CALL_EXPR_FN (exp) = gimple_call_fn (stmt); >> - decl = gimple_call_fndecl (stmt); >> builtin_p = decl && DECL_BUILT_IN (decl); >> >> /* If this is not a builtin function, the function type through which the >> diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c >> index 9f9f9cf..c03c0fc 100644 >> --- a/gcc/internal-fn.c >> +++ b/gcc/internal-fn.c >> @@ -103,7 +103,7 @@ get_multi_vector_move (tree array_type, > convert_optab optab) >> /* Expand LOAD_LANES call STMT using optab OPTAB. */ >> >> static void >> -expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) >> +expand_load_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) > > I'm somewhat confused by all these unused internal_fn args. I'm sure > we can avoid them? They're needed for: >> -/* Expand call STMT using OPTAB, which has a single output operand and >> - NARGS input operands. */ >> +/* Expand a call to FN using the operands in STMT. FN has a single >> + output operand and NARGS input operands. */ >> >> static void >> -expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) >> +expand_direct_optab_fn (internal_fn fn, gcall *stmt, direct_optab optab, >> + unsigned int nargs) >> { >> expand_operand *ops = XALLOCAVEC (expand_operand, nargs + 1); >> >> - internal_fn fn = gimple_call_internal_fn (stmt); >> tree_pair types = direct_internal_fn_types (fn, stmt); >> insn_code icode = direct_optab_handler (optab, TYPE_MODE (types.first)); where we previously assumed that the function was the same as the call. Thanks, Richard ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: Short-cut generation of simple built-in functions 2015-11-10 21:24 ` Richard Sandiford @ 2015-11-11 10:18 ` Richard Biener 2015-11-13 14:32 ` Richard Sandiford 0 siblings, 1 reply; 8+ messages in thread From: Richard Biener @ 2015-11-11 10:18 UTC (permalink / raw) To: Richard Biener, GCC Patches, richard.sandiford On Tue, Nov 10, 2015 at 10:24 PM, Richard Sandiford <richard.sandiford@arm.com> wrote: > Richard Biener <richard.guenther@gmail.com> writes: >> On Sat, Nov 7, 2015 at 2:31 PM, Richard Sandiford >> <richard.sandiford@arm.com> wrote: >>> This patch short-circuits the builtins.c expansion code for a particular >>> gimple call if: >>> >>> - the function has an associated internal function >>> - the target implements that internal function >>> - the call has no side effects >>> >>> This allows a later patch to remove the builtins.c code, once calls with >>> side effects have been handled. >>> >>> Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. >>> OK to install? >>> >>> Thanks, >>> Richard >>> >>> >>> gcc/ >>> * builtins.h (called_as_built_in): Declare. >>> * builtins.c (called_as_built_in): Make external. >>> * internal-fn.h (expand_internal_call): Define a variant that >>> specifies the internal function explicitly. >>> * internal-fn.c (expand_load_lanes_optab_fn) >>> (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) >>> (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) >>> (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) >>> (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) >>> (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) >>> (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) >>> (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) >>> (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) >>> (expand_mask_load_optab_fn, expand_mask_store_optab_fn) >>> (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) >>> (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) >>> (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) >>> (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn >>> argument. >>> (internal_fn_expanders): Update prototype. >>> (expand_internal_call): Define a variant that specifies the >>> internal function explicitly. Use it to implement the previous >>> interface. >>> * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in >>> functions as calls to internal functions. >>> >>> diff --git a/gcc/builtins.c b/gcc/builtins.c >>> index f65011e..bbcc7dc3 100644 >>> --- a/gcc/builtins.c >>> +++ b/gcc/builtins.c >>> @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) >>> of the optimization level. This means whenever a function is invoked with >>> its "internal" name, which normally contains the prefix "__builtin". */ >>> >>> -static bool >>> +bool >>> called_as_built_in (tree node) >>> { >>> /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since >>> diff --git a/gcc/builtins.h b/gcc/builtins.h >>> index 917eb90..1d00068 100644 >>> --- a/gcc/builtins.h >>> +++ b/gcc/builtins.h >>> @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; >>> extern bool force_folding_builtin_constant_p; >>> >>> extern bool is_builtin_fn (tree); >>> +extern bool called_as_built_in (tree); >>> extern bool get_object_alignment_1 (tree, unsigned int *, >>> unsigned HOST_WIDE_INT *); >>> extern unsigned int get_object_alignment (tree); >>> diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c >>> index bfbc958..dc7d4f5 100644 >>> --- a/gcc/cfgexpand.c >>> +++ b/gcc/cfgexpand.c >>> @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) >>> return; >>> } >>> >>> + /* If this is a call to a built-in function and it has no effect other >>> + than setting the lhs, try to implement it using an internal function >>> + instead. */ >>> + decl = gimple_call_fndecl (stmt); >>> + if (gimple_call_lhs (stmt) >>> + && !gimple_vdef (stmt) >> >> I think you want && ! gimple_has_side_effects (stmt) >> instead of checking !gimple_vdef (stmt). > > OK, I can do that, but what would the difference be in practice for > these types of call? I.e. are there cases for built-ins where: > > (A) gimple_vdef (stmt) && !gimple_side_effects (stmt) > > or: > > (B) !gimple_vdef (stmt) && gimple_side_effects (stmt) > > ? There was talk to make calls use volatile to prevent CSE and friends. Using gimple_has_side_effects is just the better check. > It just seems like this check should be the opposite of the one used > in the call-cdce patch (when deciding whether to optimise a call > with an lhs). In order to keep them in sync I'd need to use > gimple_side_effects rather than gimple_vdef there too, but is > (B) a possibility there? Not sure if the tests should be in-sync. I'm also not sure what you really want to check with >>> + /* If this is a call to a built-in function and it has no effect other >>> + than setting the lhs, try to implement it using an internal function >>> + instead. */ >>> + decl = gimple_call_fndecl (stmt); >>> + if (gimple_call_lhs (stmt) >>> + && !gimple_vdef (stmt) I know you want to catch errno setting here but shouldn't that use a different kind of check and only done for a specific subset of functions? For example do you want to replace sqrt() with an IFN in case it has a VUSE (it will have that with -frounding-math)? In this case you'll lose the implicit dependence on fesetround and friends. This implicit dependence is not checked by gimple_has_side_effects either btw. >>> + && (optimize || (decl && called_as_built_in (decl)))) >>> + { >>> + internal_fn ifn = replacement_internal_fn (stmt); >>> + if (ifn != IFN_LAST) >>> + { >>> + expand_internal_call (ifn, stmt); >>> + return; >>> + } >>> + } >>> + >>> exp = build_vl_exp (CALL_EXPR, gimple_call_num_args (stmt) + 3); >>> >>> CALL_EXPR_FN (exp) = gimple_call_fn (stmt); >>> - decl = gimple_call_fndecl (stmt); >>> builtin_p = decl && DECL_BUILT_IN (decl); >>> >>> /* If this is not a builtin function, the function type through which the >>> diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c >>> index 9f9f9cf..c03c0fc 100644 >>> --- a/gcc/internal-fn.c >>> +++ b/gcc/internal-fn.c >>> @@ -103,7 +103,7 @@ get_multi_vector_move (tree array_type, >> convert_optab optab) >>> /* Expand LOAD_LANES call STMT using optab OPTAB. */ >>> >>> static void >>> -expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) >>> +expand_load_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) >> >> I'm somewhat confused by all these unused internal_fn args. I'm sure >> we can avoid them? > > They're needed for: > >>> -/* Expand call STMT using OPTAB, which has a single output operand and >>> - NARGS input operands. */ >>> +/* Expand a call to FN using the operands in STMT. FN has a single >>> + output operand and NARGS input operands. */ >>> >>> static void >>> -expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) >>> +expand_direct_optab_fn (internal_fn fn, gcall *stmt, direct_optab optab, >>> + unsigned int nargs) >>> { >>> expand_operand *ops = XALLOCAVEC (expand_operand, nargs + 1); >>> >>> - internal_fn fn = gimple_call_internal_fn (stmt); >>> tree_pair types = direct_internal_fn_types (fn, stmt); >>> insn_code icode = direct_optab_handler (optab, TYPE_MODE (types.first)); > > where we previously assumed that the function was the same as the call. Ah, ok. Thanks, Richard. > Thanks, > Richard > ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: Short-cut generation of simple built-in functions 2015-11-11 10:18 ` Richard Biener @ 2015-11-13 14:32 ` Richard Sandiford 2015-11-17 9:55 ` Richard Sandiford 0 siblings, 1 reply; 8+ messages in thread From: Richard Sandiford @ 2015-11-13 14:32 UTC (permalink / raw) To: Richard Biener; +Cc: GCC Patches Richard Biener <richard.guenther@gmail.com> writes: > On Tue, Nov 10, 2015 at 10:24 PM, Richard Sandiford > <richard.sandiford@arm.com> wrote: >> Richard Biener <richard.guenther@gmail.com> writes: >>> On Sat, Nov 7, 2015 at 2:31 PM, Richard Sandiford >>> <richard.sandiford@arm.com> wrote: >>>> This patch short-circuits the builtins.c expansion code for a particular >>>> gimple call if: >>>> >>>> - the function has an associated internal function >>>> - the target implements that internal function >>>> - the call has no side effects >>>> >>>> This allows a later patch to remove the builtins.c code, once calls with >>>> side effects have been handled. >>>> >>>> Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. >>>> OK to install? >>>> >>>> Thanks, >>>> Richard >>>> >>>> >>>> gcc/ >>>> * builtins.h (called_as_built_in): Declare. >>>> * builtins.c (called_as_built_in): Make external. >>>> * internal-fn.h (expand_internal_call): Define a variant that >>>> specifies the internal function explicitly. >>>> * internal-fn.c (expand_load_lanes_optab_fn) >>>> (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) >>>> (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) >>>> (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) >>>> (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) >>>> (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) >>>> (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) >>>> (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) >>>> (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) >>>> (expand_mask_load_optab_fn, expand_mask_store_optab_fn) >>>> (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) >>>> (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) >>>> (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) >>>> (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn >>>> argument. >>>> (internal_fn_expanders): Update prototype. >>>> (expand_internal_call): Define a variant that specifies the >>>> internal function explicitly. Use it to implement the previous >>>> interface. >>>> * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in >>>> functions as calls to internal functions. >>>> >>>> diff --git a/gcc/builtins.c b/gcc/builtins.c >>>> index f65011e..bbcc7dc3 100644 >>>> --- a/gcc/builtins.c >>>> +++ b/gcc/builtins.c >>>> @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) >>>> of the optimization level. This means whenever a function is invoked with >>>> its "internal" name, which normally contains the prefix "__builtin". */ >>>> >>>> -static bool >>>> +bool >>>> called_as_built_in (tree node) >>>> { >>>> /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since >>>> diff --git a/gcc/builtins.h b/gcc/builtins.h >>>> index 917eb90..1d00068 100644 >>>> --- a/gcc/builtins.h >>>> +++ b/gcc/builtins.h >>>> @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; >>>> extern bool force_folding_builtin_constant_p; >>>> >>>> extern bool is_builtin_fn (tree); >>>> +extern bool called_as_built_in (tree); >>>> extern bool get_object_alignment_1 (tree, unsigned int *, >>>> unsigned HOST_WIDE_INT *); >>>> extern unsigned int get_object_alignment (tree); >>>> diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c >>>> index bfbc958..dc7d4f5 100644 >>>> --- a/gcc/cfgexpand.c >>>> +++ b/gcc/cfgexpand.c >>>> @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) >>>> return; >>>> } >>>> >>>> + /* If this is a call to a built-in function and it has no effect other >>>> + than setting the lhs, try to implement it using an internal function >>>> + instead. */ >>>> + decl = gimple_call_fndecl (stmt); >>>> + if (gimple_call_lhs (stmt) >>>> + && !gimple_vdef (stmt) >>> >>> I think you want && ! gimple_has_side_effects (stmt) >>> instead of checking !gimple_vdef (stmt). >> >> OK, I can do that, but what would the difference be in practice for >> these types of call? I.e. are there cases for built-ins where: >> >> (A) gimple_vdef (stmt) && !gimple_side_effects (stmt) >> >> or: >> >> (B) !gimple_vdef (stmt) && gimple_side_effects (stmt) >> >> ? > > There was talk to make calls use volatile to prevent CSE and friends. > > Using gimple_has_side_effects is just the better check. > >> It just seems like this check should be the opposite of the one used >> in the call-cdce patch (when deciding whether to optimise a call >> with an lhs). In order to keep them in sync I'd need to use >> gimple_side_effects rather than gimple_vdef there too, but is >> (B) a possibility there? > > Not sure if the tests should be in-sync. Well, with the series, we have two opportunities to use optabs: - in tree-call-cdce.c, if we need to keep the original call (or something that sets EDOM) - here in cfgexpand.c, if we don't need to keep the original call. So both are really asking "do I need to keep the call?" If we don't do either then we don't use the optab. > I'm also not sure what you really want to check with > >>>> + /* If this is a call to a built-in function and it has no effect other >>>> + than setting the lhs, try to implement it using an internal function >>>> + instead. */ >>>> + decl = gimple_call_fndecl (stmt); >>>> + if (gimple_call_lhs (stmt) >>>> + && !gimple_vdef (stmt) > > I know you want to catch errno setting here but shouldn't that use a > different kind of check and only done for a specific subset of > functions? For example do you want to replace sqrt() with > an IFN in case it has a VUSE (it will have that with -frounding-math)? > In this case you'll lose the implicit dependence on fesetround and > friends. This implicit dependence is not checked by gimple_has_side_effects > either btw. Yeah, we want to replace even in that case. Remember that this is expand code, so we're going to generate rtl whatever happens. It's up to the rtl patterns to represent any ordering requirements. (Not that we do have explicit rtl dependencies for rounding mode, but we don't for +, -. * or / either.) If for some reason a particular .md pattern doesn't honour the current rounding mode when normally the optab would be expected to, the .md file should limit the pattern to !flag_rounding_math. Thanks, Richard ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: Short-cut generation of simple built-in functions 2015-11-13 14:32 ` Richard Sandiford @ 2015-11-17 9:55 ` Richard Sandiford 2015-11-17 14:34 ` Richard Biener 0 siblings, 1 reply; 8+ messages in thread From: Richard Sandiford @ 2015-11-17 9:55 UTC (permalink / raw) To: Richard Biener; +Cc: GCC Patches Richard Sandiford <richard.sandiford@arm.com> writes: > Richard Biener <richard.guenther@gmail.com> writes: >> On Tue, Nov 10, 2015 at 10:24 PM, Richard Sandiford >> <richard.sandiford@arm.com> wrote: >>> Richard Biener <richard.guenther@gmail.com> writes: >>>> On Sat, Nov 7, 2015 at 2:31 PM, Richard Sandiford >>>> <richard.sandiford@arm.com> wrote: >>>>> This patch short-circuits the builtins.c expansion code for a particular >>>>> gimple call if: >>>>> >>>>> - the function has an associated internal function >>>>> - the target implements that internal function >>>>> - the call has no side effects >>>>> >>>>> This allows a later patch to remove the builtins.c code, once calls with >>>>> side effects have been handled. >>>>> >>>>> Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. >>>>> OK to install? >>>>> >>>>> Thanks, >>>>> Richard >>>>> >>>>> >>>>> gcc/ >>>>> * builtins.h (called_as_built_in): Declare. >>>>> * builtins.c (called_as_built_in): Make external. >>>>> * internal-fn.h (expand_internal_call): Define a variant that >>>>> specifies the internal function explicitly. >>>>> * internal-fn.c (expand_load_lanes_optab_fn) >>>>> (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) >>>>> (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) >>>>> (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) >>>>> (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) >>>>> (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) >>>>> (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) >>>>> (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) >>>>> (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) >>>>> (expand_mask_load_optab_fn, expand_mask_store_optab_fn) >>>>> (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) >>>>> (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) >>>>> (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) >>>>> (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn >>>>> argument. >>>>> (internal_fn_expanders): Update prototype. >>>>> (expand_internal_call): Define a variant that specifies the >>>>> internal function explicitly. Use it to implement the previous >>>>> interface. >>>>> * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in >>>>> functions as calls to internal functions. >>>>> >>>>> diff --git a/gcc/builtins.c b/gcc/builtins.c >>>>> index f65011e..bbcc7dc3 100644 >>>>> --- a/gcc/builtins.c >>>>> +++ b/gcc/builtins.c >>>>> @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) >>>>> of the optimization level. This means whenever a function is invoked with >>>>> its "internal" name, which normally contains the prefix "__builtin". */ >>>>> >>>>> -static bool >>>>> +bool >>>>> called_as_built_in (tree node) >>>>> { >>>>> /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since >>>>> diff --git a/gcc/builtins.h b/gcc/builtins.h >>>>> index 917eb90..1d00068 100644 >>>>> --- a/gcc/builtins.h >>>>> +++ b/gcc/builtins.h >>>>> @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; >>>>> extern bool force_folding_builtin_constant_p; >>>>> >>>>> extern bool is_builtin_fn (tree); >>>>> +extern bool called_as_built_in (tree); >>>>> extern bool get_object_alignment_1 (tree, unsigned int *, >>>>> unsigned HOST_WIDE_INT *); >>>>> extern unsigned int get_object_alignment (tree); >>>>> diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c >>>>> index bfbc958..dc7d4f5 100644 >>>>> --- a/gcc/cfgexpand.c >>>>> +++ b/gcc/cfgexpand.c >>>>> @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) >>>>> return; >>>>> } >>>>> >>>>> + /* If this is a call to a built-in function and it has no effect other >>>>> + than setting the lhs, try to implement it using an internal function >>>>> + instead. */ >>>>> + decl = gimple_call_fndecl (stmt); >>>>> + if (gimple_call_lhs (stmt) >>>>> + && !gimple_vdef (stmt) >>>> >>>> I think you want && ! gimple_has_side_effects (stmt) >>>> instead of checking !gimple_vdef (stmt). >>> >>> OK, I can do that, but what would the difference be in practice for >>> these types of call? I.e. are there cases for built-ins where: >>> >>> (A) gimple_vdef (stmt) && !gimple_side_effects (stmt) >>> >>> or: >>> >>> (B) !gimple_vdef (stmt) && gimple_side_effects (stmt) >>> >>> ? >> >> There was talk to make calls use volatile to prevent CSE and friends. >> >> Using gimple_has_side_effects is just the better check. >> >>> It just seems like this check should be the opposite of the one used >>> in the call-cdce patch (when deciding whether to optimise a call >>> with an lhs). In order to keep them in sync I'd need to use >>> gimple_side_effects rather than gimple_vdef there too, but is >>> (B) a possibility there? >> >> Not sure if the tests should be in-sync. > > Well, with the series, we have two opportunities to use optabs: > > - in tree-call-cdce.c, if we need to keep the original call > (or something that sets EDOM) > > - here in cfgexpand.c, if we don't need to keep the original call. > > So both are really asking "do I need to keep the call?" > > If we don't do either then we don't use the optab. Anyway, the updated patch below uses gimple_side_effects_p. >> I'm also not sure what you really want to check with >> >>>>> + /* If this is a call to a built-in function and it has no effect other >>>>> + than setting the lhs, try to implement it using an internal function >>>>> + instead. */ >>>>> + decl = gimple_call_fndecl (stmt); >>>>> + if (gimple_call_lhs (stmt) >>>>> + && !gimple_vdef (stmt) >> >> I know you want to catch errno setting here but shouldn't that use a >> different kind of check and only done for a specific subset of >> functions? For example do you want to replace sqrt() with >> an IFN in case it has a VUSE (it will have that with -frounding-math)? >> In this case you'll lose the implicit dependence on fesetround and >> friends. This implicit dependence is not checked by gimple_has_side_effects >> either btw. > > Yeah, we want to replace even in that case. Remember that this is > expand code, so we're going to generate rtl whatever happens. It's up > to the rtl patterns to represent any ordering requirements. (Not that > we do have explicit rtl dependencies for rounding mode, but we don't for > +, -. * or / either.) > > If for some reason a particular .md pattern doesn't honour the current > rounding mode when normally the optab would be expected to, the .md file > should limit the pattern to !flag_rounding_math. Tested as before. OK to install? Thanks, Richard gcc/ * builtins.h (called_as_built_in): Declare. * builtins.c (called_as_built_in): Make external. * internal-fn.h (expand_internal_call): Define a variant that specifies the internal function explicitly. * internal-fn.c (expand_load_lanes_optab_fn) (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) (expand_mask_load_optab_fn, expand_mask_store_optab_fn) (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn argument. (internal_fn_expanders): Update prototype. (expand_internal_call): Define a variant that specifies the internal function explicitly. Use it to implement the previous interface. * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in functions as calls to internal functions. diff --git a/gcc/builtins.c b/gcc/builtins.c index 6df1b9b..c422d0d 100644 --- a/gcc/builtins.c +++ b/gcc/builtins.c @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) of the optimization level. This means whenever a function is invoked with its "internal" name, which normally contains the prefix "__builtin". */ -static bool +bool called_as_built_in (tree node) { /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since diff --git a/gcc/builtins.h b/gcc/builtins.h index 917eb90..1d00068 100644 --- a/gcc/builtins.h +++ b/gcc/builtins.h @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; extern bool force_folding_builtin_constant_p; extern bool is_builtin_fn (tree); +extern bool called_as_built_in (tree); extern bool get_object_alignment_1 (tree, unsigned int *, unsigned HOST_WIDE_INT *); extern unsigned int get_object_alignment (tree); diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c index e55467a..1990e10 100644 --- a/gcc/cfgexpand.c +++ b/gcc/cfgexpand.c @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) return; } + /* If this is a call to a built-in function and it has no effect other + than setting the lhs, try to implement it using an internal function + instead. */ + decl = gimple_call_fndecl (stmt); + if (gimple_call_lhs (stmt) + && !gimple_has_side_effects (stmt) + && (optimize || (decl && called_as_built_in (decl)))) + { + internal_fn ifn = replacement_internal_fn (stmt); + if (ifn != IFN_LAST) + { + expand_internal_call (ifn, stmt); + return; + } + } + exp = build_vl_exp (CALL_EXPR, gimple_call_num_args (stmt) + 3); CALL_EXPR_FN (exp) = gimple_call_fn (stmt); - decl = gimple_call_fndecl (stmt); builtin_p = decl && DECL_BUILT_IN (decl); /* If this is not a builtin function, the function type through which the diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c index b853162..f23d799 100644 --- a/gcc/internal-fn.c +++ b/gcc/internal-fn.c @@ -103,7 +103,7 @@ get_multi_vector_move (tree array_type, convert_optab optab) /* Expand LOAD_LANES call STMT using optab OPTAB. */ static void -expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) +expand_load_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) { struct expand_operand ops[2]; tree type, lhs, rhs; @@ -127,7 +127,7 @@ expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) /* Expand STORE_LANES call STMT using optab OPTAB. */ static void -expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) +expand_store_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) { struct expand_operand ops[2]; tree type, lhs, rhs; @@ -149,7 +149,7 @@ expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) } static void -expand_ANNOTATE (gcall *) +expand_ANNOTATE (internal_fn, gcall *) { gcc_unreachable (); } @@ -157,7 +157,7 @@ expand_ANNOTATE (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_LANE (gcall *) +expand_GOMP_SIMD_LANE (internal_fn, gcall *) { gcc_unreachable (); } @@ -165,7 +165,7 @@ expand_GOMP_SIMD_LANE (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_VF (gcall *) +expand_GOMP_SIMD_VF (internal_fn, gcall *) { gcc_unreachable (); } @@ -173,7 +173,7 @@ expand_GOMP_SIMD_VF (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_LAST_LANE (gcall *) +expand_GOMP_SIMD_LAST_LANE (internal_fn, gcall *) { gcc_unreachable (); } @@ -181,7 +181,7 @@ expand_GOMP_SIMD_LAST_LANE (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_ORDERED_START (gcall *) +expand_GOMP_SIMD_ORDERED_START (internal_fn, gcall *) { gcc_unreachable (); } @@ -189,7 +189,7 @@ expand_GOMP_SIMD_ORDERED_START (gcall *) /* This should get expanded in adjust_simduid_builtins. */ static void -expand_GOMP_SIMD_ORDERED_END (gcall *) +expand_GOMP_SIMD_ORDERED_END (internal_fn, gcall *) { gcc_unreachable (); } @@ -197,7 +197,7 @@ expand_GOMP_SIMD_ORDERED_END (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_NULL (gcall *) +expand_UBSAN_NULL (internal_fn, gcall *) { gcc_unreachable (); } @@ -205,7 +205,7 @@ expand_UBSAN_NULL (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_BOUNDS (gcall *) +expand_UBSAN_BOUNDS (internal_fn, gcall *) { gcc_unreachable (); } @@ -213,7 +213,7 @@ expand_UBSAN_BOUNDS (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_VPTR (gcall *) +expand_UBSAN_VPTR (internal_fn, gcall *) { gcc_unreachable (); } @@ -221,7 +221,7 @@ expand_UBSAN_VPTR (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_UBSAN_OBJECT_SIZE (gcall *) +expand_UBSAN_OBJECT_SIZE (internal_fn, gcall *) { gcc_unreachable (); } @@ -229,7 +229,7 @@ expand_UBSAN_OBJECT_SIZE (gcall *) /* This should get expanded in the sanopt pass. */ static void -expand_ASAN_CHECK (gcall *) +expand_ASAN_CHECK (internal_fn, gcall *) { gcc_unreachable (); } @@ -237,7 +237,7 @@ expand_ASAN_CHECK (gcall *) /* This should get expanded in the tsan pass. */ static void -expand_TSAN_FUNC_EXIT (gcall *) +expand_TSAN_FUNC_EXIT (internal_fn, gcall *) { gcc_unreachable (); } @@ -1639,7 +1639,7 @@ expand_mul_overflow (location_t loc, tree lhs, tree arg0, tree arg1, /* Expand UBSAN_CHECK_ADD call STMT. */ static void -expand_UBSAN_CHECK_ADD (gcall *stmt) +expand_UBSAN_CHECK_ADD (internal_fn, gcall *stmt) { location_t loc = gimple_location (stmt); tree lhs = gimple_call_lhs (stmt); @@ -1652,7 +1652,7 @@ expand_UBSAN_CHECK_ADD (gcall *stmt) /* Expand UBSAN_CHECK_SUB call STMT. */ static void -expand_UBSAN_CHECK_SUB (gcall *stmt) +expand_UBSAN_CHECK_SUB (internal_fn, gcall *stmt) { location_t loc = gimple_location (stmt); tree lhs = gimple_call_lhs (stmt); @@ -1668,7 +1668,7 @@ expand_UBSAN_CHECK_SUB (gcall *stmt) /* Expand UBSAN_CHECK_MUL call STMT. */ static void -expand_UBSAN_CHECK_MUL (gcall *stmt) +expand_UBSAN_CHECK_MUL (internal_fn, gcall *stmt) { location_t loc = gimple_location (stmt); tree lhs = gimple_call_lhs (stmt); @@ -1853,7 +1853,7 @@ expand_arith_overflow (enum tree_code code, gimple *stmt) /* Expand ADD_OVERFLOW STMT. */ static void -expand_ADD_OVERFLOW (gcall *stmt) +expand_ADD_OVERFLOW (internal_fn, gcall *stmt) { expand_arith_overflow (PLUS_EXPR, stmt); } @@ -1861,7 +1861,7 @@ expand_ADD_OVERFLOW (gcall *stmt) /* Expand SUB_OVERFLOW STMT. */ static void -expand_SUB_OVERFLOW (gcall *stmt) +expand_SUB_OVERFLOW (internal_fn, gcall *stmt) { expand_arith_overflow (MINUS_EXPR, stmt); } @@ -1869,7 +1869,7 @@ expand_SUB_OVERFLOW (gcall *stmt) /* Expand MUL_OVERFLOW STMT. */ static void -expand_MUL_OVERFLOW (gcall *stmt) +expand_MUL_OVERFLOW (internal_fn, gcall *stmt) { expand_arith_overflow (MULT_EXPR, stmt); } @@ -1877,7 +1877,7 @@ expand_MUL_OVERFLOW (gcall *stmt) /* This should get folded in tree-vectorizer.c. */ static void -expand_LOOP_VECTORIZED (gcall *) +expand_LOOP_VECTORIZED (internal_fn, gcall *) { gcc_unreachable (); } @@ -1885,7 +1885,7 @@ expand_LOOP_VECTORIZED (gcall *) /* Expand MASK_LOAD call STMT using optab OPTAB. */ static void -expand_mask_load_optab_fn (gcall *stmt, convert_optab optab) +expand_mask_load_optab_fn (internal_fn, gcall *stmt, convert_optab optab) { struct expand_operand ops[3]; tree type, lhs, rhs, maskt; @@ -1914,7 +1914,7 @@ expand_mask_load_optab_fn (gcall *stmt, convert_optab optab) /* Expand MASK_STORE call STMT using optab OPTAB. */ static void -expand_mask_store_optab_fn (gcall *stmt, convert_optab optab) +expand_mask_store_optab_fn (internal_fn, gcall *stmt, convert_optab optab) { struct expand_operand ops[3]; tree type, lhs, rhs, maskt; @@ -1939,12 +1939,12 @@ expand_mask_store_optab_fn (gcall *stmt, convert_optab optab) } static void -expand_ABNORMAL_DISPATCHER (gcall *) +expand_ABNORMAL_DISPATCHER (internal_fn, gcall *) { } static void -expand_BUILTIN_EXPECT (gcall *stmt) +expand_BUILTIN_EXPECT (internal_fn, gcall *stmt) { /* When guessing was done, the hints should be already stripped away. */ gcc_assert (!flag_guess_branch_prob || optimize == 0 || seen_error ()); @@ -1964,7 +1964,7 @@ expand_BUILTIN_EXPECT (gcall *stmt) should never be called. */ static void -expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) +expand_VA_ARG (internal_fn, gcall *) { gcc_unreachable (); } @@ -1972,7 +1972,7 @@ expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) /* Expand the IFN_UNIQUE function according to its first argument. */ static void -expand_UNIQUE (gcall *stmt) +expand_UNIQUE (internal_fn, gcall *stmt) { rtx pattern = NULL_RTX; enum ifn_unique_kind kind @@ -2018,7 +2018,7 @@ expand_UNIQUE (gcall *stmt) /* The size of an OpenACC compute dimension. */ static void -expand_GOACC_DIM_SIZE (gcall *stmt) +expand_GOACC_DIM_SIZE (internal_fn, gcall *stmt) { tree lhs = gimple_call_lhs (stmt); @@ -2039,7 +2039,7 @@ expand_GOACC_DIM_SIZE (gcall *stmt) /* The position of an OpenACC execution engine along one compute axis. */ static void -expand_GOACC_DIM_POS (gcall *stmt) +expand_GOACC_DIM_POS (internal_fn, gcall *stmt) { tree lhs = gimple_call_lhs (stmt); @@ -2060,7 +2060,7 @@ expand_GOACC_DIM_POS (gcall *stmt) /* This is expanded by oacc_device_lower pass. */ static void -expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) +expand_GOACC_LOOP (internal_fn, gcall *) { gcc_unreachable (); } @@ -2068,20 +2068,20 @@ expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) /* This is expanded by oacc_device_lower pass. */ static void -expand_GOACC_REDUCTION (gcall *stmt ATTRIBUTE_UNUSED) +expand_GOACC_REDUCTION (internal_fn, gcall *) { gcc_unreachable (); } -/* Expand call STMT using OPTAB, which has a single output operand and - NARGS input operands. */ +/* Expand a call to FN using the operands in STMT. FN has a single + output operand and NARGS input operands. */ static void -expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) +expand_direct_optab_fn (internal_fn fn, gcall *stmt, direct_optab optab, + unsigned int nargs) { expand_operand *ops = XALLOCAVEC (expand_operand, nargs + 1); - internal_fn fn = gimple_call_internal_fn (stmt); tree_pair types = direct_internal_fn_types (fn, stmt); insn_code icode = direct_optab_handler (optab, TYPE_MODE (types.first)); @@ -2119,11 +2119,11 @@ expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) /* Expanders for optabs that can use expand_direct_optab_fn. */ -#define expand_unary_optab_fn(STMT, OPTAB) \ - expand_direct_optab_fn (STMT, OPTAB, 1) +#define expand_unary_optab_fn(FN, STMT, OPTAB) \ + expand_direct_optab_fn (FN, STMT, OPTAB, 1) -#define expand_binary_optab_fn(STMT, OPTAB) \ - expand_direct_optab_fn (STMT, OPTAB, 2) +#define expand_binary_optab_fn(FN, STMT, OPTAB) \ + expand_direct_optab_fn (FN, STMT, OPTAB, 2) /* RETURN_TYPE and ARGS are a return type and argument list that are in principle compatible with FN (which satisfies direct_internal_fn_p). @@ -2219,9 +2219,9 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) #define DEF_INTERNAL_OPTAB_FN(CODE, FLAGS, OPTAB, TYPE) \ static void \ - expand_##CODE (gcall *stmt) \ + expand_##CODE (internal_fn fn, gcall *stmt) \ { \ - expand_##TYPE##_optab_fn (stmt, OPTAB##_optab); \ + expand_##TYPE##_optab_fn (fn, stmt, OPTAB##_optab); \ } #include "internal-fn.def" @@ -2231,16 +2231,24 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) expand_<NAME> (gcall *stmt) where STMT is the statement that performs the call. */ -static void (*const internal_fn_expanders[]) (gcall *) = { +static void (*const internal_fn_expanders[]) (internal_fn, gcall *) = { #define DEF_INTERNAL_FN(CODE, FLAGS, FNSPEC) expand_##CODE, #include "internal-fn.def" 0 }; +/* Expand STMT as though it were a call to internal function FN. */ + +void +expand_internal_call (internal_fn fn, gcall *stmt) +{ + internal_fn_expanders[fn] (fn, stmt); +} + /* Expand STMT, which is a call to internal function FN. */ void expand_internal_call (gcall *stmt) { - internal_fn_expanders[(int) gimple_call_internal_fn (stmt)] (stmt); + expand_internal_call (gimple_call_internal_fn (stmt), stmt); } diff --git a/gcc/internal-fn.h b/gcc/internal-fn.h index 31e895e..5ee43b8 100644 --- a/gcc/internal-fn.h +++ b/gcc/internal-fn.h @@ -162,5 +162,6 @@ extern bool direct_internal_fn_supported_p (internal_fn, tree_pair); extern bool direct_internal_fn_supported_p (internal_fn, tree); extern void expand_internal_call (gcall *); +extern void expand_internal_call (internal_fn, gcall *); #endif ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: Short-cut generation of simple built-in functions 2015-11-17 9:55 ` Richard Sandiford @ 2015-11-17 14:34 ` Richard Biener 0 siblings, 0 replies; 8+ messages in thread From: Richard Biener @ 2015-11-17 14:34 UTC (permalink / raw) To: Richard Biener, GCC Patches, richard.sandiford On Tue, Nov 17, 2015 at 10:55 AM, Richard Sandiford <richard.sandiford@arm.com> wrote: > Richard Sandiford <richard.sandiford@arm.com> writes: >> Richard Biener <richard.guenther@gmail.com> writes: >>> On Tue, Nov 10, 2015 at 10:24 PM, Richard Sandiford >>> <richard.sandiford@arm.com> wrote: >>>> Richard Biener <richard.guenther@gmail.com> writes: >>>>> On Sat, Nov 7, 2015 at 2:31 PM, Richard Sandiford >>>>> <richard.sandiford@arm.com> wrote: >>>>>> This patch short-circuits the builtins.c expansion code for a particular >>>>>> gimple call if: >>>>>> >>>>>> - the function has an associated internal function >>>>>> - the target implements that internal function >>>>>> - the call has no side effects >>>>>> >>>>>> This allows a later patch to remove the builtins.c code, once calls with >>>>>> side effects have been handled. >>>>>> >>>>>> Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. >>>>>> OK to install? >>>>>> >>>>>> Thanks, >>>>>> Richard >>>>>> >>>>>> >>>>>> gcc/ >>>>>> * builtins.h (called_as_built_in): Declare. >>>>>> * builtins.c (called_as_built_in): Make external. >>>>>> * internal-fn.h (expand_internal_call): Define a variant that >>>>>> specifies the internal function explicitly. >>>>>> * internal-fn.c (expand_load_lanes_optab_fn) >>>>>> (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) >>>>>> (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) >>>>>> (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) >>>>>> (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) >>>>>> (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) >>>>>> (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) >>>>>> (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) >>>>>> (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) >>>>>> (expand_mask_load_optab_fn, expand_mask_store_optab_fn) >>>>>> (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) >>>>>> (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) >>>>>> (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) >>>>>> (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn >>>>>> argument. >>>>>> (internal_fn_expanders): Update prototype. >>>>>> (expand_internal_call): Define a variant that specifies the >>>>>> internal function explicitly. Use it to implement the previous >>>>>> interface. >>>>>> * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in >>>>>> functions as calls to internal functions. >>>>>> >>>>>> diff --git a/gcc/builtins.c b/gcc/builtins.c >>>>>> index f65011e..bbcc7dc3 100644 >>>>>> --- a/gcc/builtins.c >>>>>> +++ b/gcc/builtins.c >>>>>> @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) >>>>>> of the optimization level. This means whenever a function is invoked with >>>>>> its "internal" name, which normally contains the prefix "__builtin". */ >>>>>> >>>>>> -static bool >>>>>> +bool >>>>>> called_as_built_in (tree node) >>>>>> { >>>>>> /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since >>>>>> diff --git a/gcc/builtins.h b/gcc/builtins.h >>>>>> index 917eb90..1d00068 100644 >>>>>> --- a/gcc/builtins.h >>>>>> +++ b/gcc/builtins.h >>>>>> @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; >>>>>> extern bool force_folding_builtin_constant_p; >>>>>> >>>>>> extern bool is_builtin_fn (tree); >>>>>> +extern bool called_as_built_in (tree); >>>>>> extern bool get_object_alignment_1 (tree, unsigned int *, >>>>>> unsigned HOST_WIDE_INT *); >>>>>> extern unsigned int get_object_alignment (tree); >>>>>> diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c >>>>>> index bfbc958..dc7d4f5 100644 >>>>>> --- a/gcc/cfgexpand.c >>>>>> +++ b/gcc/cfgexpand.c >>>>>> @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) >>>>>> return; >>>>>> } >>>>>> >>>>>> + /* If this is a call to a built-in function and it has no effect other >>>>>> + than setting the lhs, try to implement it using an internal function >>>>>> + instead. */ >>>>>> + decl = gimple_call_fndecl (stmt); >>>>>> + if (gimple_call_lhs (stmt) >>>>>> + && !gimple_vdef (stmt) >>>>> >>>>> I think you want && ! gimple_has_side_effects (stmt) >>>>> instead of checking !gimple_vdef (stmt). >>>> >>>> OK, I can do that, but what would the difference be in practice for >>>> these types of call? I.e. are there cases for built-ins where: >>>> >>>> (A) gimple_vdef (stmt) && !gimple_side_effects (stmt) >>>> >>>> or: >>>> >>>> (B) !gimple_vdef (stmt) && gimple_side_effects (stmt) >>>> >>>> ? >>> >>> There was talk to make calls use volatile to prevent CSE and friends. >>> >>> Using gimple_has_side_effects is just the better check. >>> >>>> It just seems like this check should be the opposite of the one used >>>> in the call-cdce patch (when deciding whether to optimise a call >>>> with an lhs). In order to keep them in sync I'd need to use >>>> gimple_side_effects rather than gimple_vdef there too, but is >>>> (B) a possibility there? >>> >>> Not sure if the tests should be in-sync. >> >> Well, with the series, we have two opportunities to use optabs: >> >> - in tree-call-cdce.c, if we need to keep the original call >> (or something that sets EDOM) >> >> - here in cfgexpand.c, if we don't need to keep the original call. >> >> So both are really asking "do I need to keep the call?" >> >> If we don't do either then we don't use the optab. > > Anyway, the updated patch below uses gimple_side_effects_p. > >>> I'm also not sure what you really want to check with >>> >>>>>> + /* If this is a call to a built-in function and it has no effect other >>>>>> + than setting the lhs, try to implement it using an internal function >>>>>> + instead. */ >>>>>> + decl = gimple_call_fndecl (stmt); >>>>>> + if (gimple_call_lhs (stmt) >>>>>> + && !gimple_vdef (stmt) >>> >>> I know you want to catch errno setting here but shouldn't that use a >>> different kind of check and only done for a specific subset of >>> functions? For example do you want to replace sqrt() with >>> an IFN in case it has a VUSE (it will have that with -frounding-math)? >>> In this case you'll lose the implicit dependence on fesetround and >>> friends. This implicit dependence is not checked by gimple_has_side_effects >>> either btw. >> >> Yeah, we want to replace even in that case. Remember that this is >> expand code, so we're going to generate rtl whatever happens. It's up >> to the rtl patterns to represent any ordering requirements. (Not that >> we do have explicit rtl dependencies for rounding mode, but we don't for >> +, -. * or / either.) >> >> If for some reason a particular .md pattern doesn't honour the current >> rounding mode when normally the optab would be expected to, the .md file >> should limit the pattern to !flag_rounding_math. > > Tested as before. OK to install? Ok. Thanks, Richard. > Thanks, > Richard > > > gcc/ > * builtins.h (called_as_built_in): Declare. > * builtins.c (called_as_built_in): Make external. > * internal-fn.h (expand_internal_call): Define a variant that > specifies the internal function explicitly. > * internal-fn.c (expand_load_lanes_optab_fn) > (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) > (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) > (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) > (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) > (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) > (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) > (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) > (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) > (expand_mask_load_optab_fn, expand_mask_store_optab_fn) > (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) > (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) > (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) > (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn > argument. > (internal_fn_expanders): Update prototype. > (expand_internal_call): Define a variant that specifies the > internal function explicitly. Use it to implement the previous > interface. > * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in > functions as calls to internal functions. > > diff --git a/gcc/builtins.c b/gcc/builtins.c > index 6df1b9b..c422d0d 100644 > --- a/gcc/builtins.c > +++ b/gcc/builtins.c > @@ -222,7 +222,7 @@ is_builtin_fn (tree decl) > of the optimization level. This means whenever a function is invoked with > its "internal" name, which normally contains the prefix "__builtin". */ > > -static bool > +bool > called_as_built_in (tree node) > { > /* Note that we must use DECL_NAME, not DECL_ASSEMBLER_NAME_SET_P since > diff --git a/gcc/builtins.h b/gcc/builtins.h > index 917eb90..1d00068 100644 > --- a/gcc/builtins.h > +++ b/gcc/builtins.h > @@ -50,6 +50,7 @@ extern struct target_builtins *this_target_builtins; > extern bool force_folding_builtin_constant_p; > > extern bool is_builtin_fn (tree); > +extern bool called_as_built_in (tree); > extern bool get_object_alignment_1 (tree, unsigned int *, > unsigned HOST_WIDE_INT *); > extern unsigned int get_object_alignment (tree); > diff --git a/gcc/cfgexpand.c b/gcc/cfgexpand.c > index e55467a..1990e10 100644 > --- a/gcc/cfgexpand.c > +++ b/gcc/cfgexpand.c > @@ -2551,10 +2551,25 @@ expand_call_stmt (gcall *stmt) > return; > } > > + /* If this is a call to a built-in function and it has no effect other > + than setting the lhs, try to implement it using an internal function > + instead. */ > + decl = gimple_call_fndecl (stmt); > + if (gimple_call_lhs (stmt) > + && !gimple_has_side_effects (stmt) > + && (optimize || (decl && called_as_built_in (decl)))) > + { > + internal_fn ifn = replacement_internal_fn (stmt); > + if (ifn != IFN_LAST) > + { > + expand_internal_call (ifn, stmt); > + return; > + } > + } > + > exp = build_vl_exp (CALL_EXPR, gimple_call_num_args (stmt) + 3); > > CALL_EXPR_FN (exp) = gimple_call_fn (stmt); > - decl = gimple_call_fndecl (stmt); > builtin_p = decl && DECL_BUILT_IN (decl); > > /* If this is not a builtin function, the function type through which the > diff --git a/gcc/internal-fn.c b/gcc/internal-fn.c > index b853162..f23d799 100644 > --- a/gcc/internal-fn.c > +++ b/gcc/internal-fn.c > @@ -103,7 +103,7 @@ get_multi_vector_move (tree array_type, convert_optab optab) > /* Expand LOAD_LANES call STMT using optab OPTAB. */ > > static void > -expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) > +expand_load_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) > { > struct expand_operand ops[2]; > tree type, lhs, rhs; > @@ -127,7 +127,7 @@ expand_load_lanes_optab_fn (gcall *stmt, convert_optab optab) > /* Expand STORE_LANES call STMT using optab OPTAB. */ > > static void > -expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) > +expand_store_lanes_optab_fn (internal_fn, gcall *stmt, convert_optab optab) > { > struct expand_operand ops[2]; > tree type, lhs, rhs; > @@ -149,7 +149,7 @@ expand_store_lanes_optab_fn (gcall *stmt, convert_optab optab) > } > > static void > -expand_ANNOTATE (gcall *) > +expand_ANNOTATE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -157,7 +157,7 @@ expand_ANNOTATE (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_LANE (gcall *) > +expand_GOMP_SIMD_LANE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -165,7 +165,7 @@ expand_GOMP_SIMD_LANE (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_VF (gcall *) > +expand_GOMP_SIMD_VF (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -173,7 +173,7 @@ expand_GOMP_SIMD_VF (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_LAST_LANE (gcall *) > +expand_GOMP_SIMD_LAST_LANE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -181,7 +181,7 @@ expand_GOMP_SIMD_LAST_LANE (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_ORDERED_START (gcall *) > +expand_GOMP_SIMD_ORDERED_START (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -189,7 +189,7 @@ expand_GOMP_SIMD_ORDERED_START (gcall *) > /* This should get expanded in adjust_simduid_builtins. */ > > static void > -expand_GOMP_SIMD_ORDERED_END (gcall *) > +expand_GOMP_SIMD_ORDERED_END (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -197,7 +197,7 @@ expand_GOMP_SIMD_ORDERED_END (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_NULL (gcall *) > +expand_UBSAN_NULL (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -205,7 +205,7 @@ expand_UBSAN_NULL (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_BOUNDS (gcall *) > +expand_UBSAN_BOUNDS (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -213,7 +213,7 @@ expand_UBSAN_BOUNDS (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_VPTR (gcall *) > +expand_UBSAN_VPTR (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -221,7 +221,7 @@ expand_UBSAN_VPTR (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_UBSAN_OBJECT_SIZE (gcall *) > +expand_UBSAN_OBJECT_SIZE (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -229,7 +229,7 @@ expand_UBSAN_OBJECT_SIZE (gcall *) > /* This should get expanded in the sanopt pass. */ > > static void > -expand_ASAN_CHECK (gcall *) > +expand_ASAN_CHECK (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -237,7 +237,7 @@ expand_ASAN_CHECK (gcall *) > /* This should get expanded in the tsan pass. */ > > static void > -expand_TSAN_FUNC_EXIT (gcall *) > +expand_TSAN_FUNC_EXIT (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -1639,7 +1639,7 @@ expand_mul_overflow (location_t loc, tree lhs, tree arg0, tree arg1, > /* Expand UBSAN_CHECK_ADD call STMT. */ > > static void > -expand_UBSAN_CHECK_ADD (gcall *stmt) > +expand_UBSAN_CHECK_ADD (internal_fn, gcall *stmt) > { > location_t loc = gimple_location (stmt); > tree lhs = gimple_call_lhs (stmt); > @@ -1652,7 +1652,7 @@ expand_UBSAN_CHECK_ADD (gcall *stmt) > /* Expand UBSAN_CHECK_SUB call STMT. */ > > static void > -expand_UBSAN_CHECK_SUB (gcall *stmt) > +expand_UBSAN_CHECK_SUB (internal_fn, gcall *stmt) > { > location_t loc = gimple_location (stmt); > tree lhs = gimple_call_lhs (stmt); > @@ -1668,7 +1668,7 @@ expand_UBSAN_CHECK_SUB (gcall *stmt) > /* Expand UBSAN_CHECK_MUL call STMT. */ > > static void > -expand_UBSAN_CHECK_MUL (gcall *stmt) > +expand_UBSAN_CHECK_MUL (internal_fn, gcall *stmt) > { > location_t loc = gimple_location (stmt); > tree lhs = gimple_call_lhs (stmt); > @@ -1853,7 +1853,7 @@ expand_arith_overflow (enum tree_code code, gimple *stmt) > /* Expand ADD_OVERFLOW STMT. */ > > static void > -expand_ADD_OVERFLOW (gcall *stmt) > +expand_ADD_OVERFLOW (internal_fn, gcall *stmt) > { > expand_arith_overflow (PLUS_EXPR, stmt); > } > @@ -1861,7 +1861,7 @@ expand_ADD_OVERFLOW (gcall *stmt) > /* Expand SUB_OVERFLOW STMT. */ > > static void > -expand_SUB_OVERFLOW (gcall *stmt) > +expand_SUB_OVERFLOW (internal_fn, gcall *stmt) > { > expand_arith_overflow (MINUS_EXPR, stmt); > } > @@ -1869,7 +1869,7 @@ expand_SUB_OVERFLOW (gcall *stmt) > /* Expand MUL_OVERFLOW STMT. */ > > static void > -expand_MUL_OVERFLOW (gcall *stmt) > +expand_MUL_OVERFLOW (internal_fn, gcall *stmt) > { > expand_arith_overflow (MULT_EXPR, stmt); > } > @@ -1877,7 +1877,7 @@ expand_MUL_OVERFLOW (gcall *stmt) > /* This should get folded in tree-vectorizer.c. */ > > static void > -expand_LOOP_VECTORIZED (gcall *) > +expand_LOOP_VECTORIZED (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -1885,7 +1885,7 @@ expand_LOOP_VECTORIZED (gcall *) > /* Expand MASK_LOAD call STMT using optab OPTAB. */ > > static void > -expand_mask_load_optab_fn (gcall *stmt, convert_optab optab) > +expand_mask_load_optab_fn (internal_fn, gcall *stmt, convert_optab optab) > { > struct expand_operand ops[3]; > tree type, lhs, rhs, maskt; > @@ -1914,7 +1914,7 @@ expand_mask_load_optab_fn (gcall *stmt, convert_optab optab) > /* Expand MASK_STORE call STMT using optab OPTAB. */ > > static void > -expand_mask_store_optab_fn (gcall *stmt, convert_optab optab) > +expand_mask_store_optab_fn (internal_fn, gcall *stmt, convert_optab optab) > { > struct expand_operand ops[3]; > tree type, lhs, rhs, maskt; > @@ -1939,12 +1939,12 @@ expand_mask_store_optab_fn (gcall *stmt, convert_optab optab) > } > > static void > -expand_ABNORMAL_DISPATCHER (gcall *) > +expand_ABNORMAL_DISPATCHER (internal_fn, gcall *) > { > } > > static void > -expand_BUILTIN_EXPECT (gcall *stmt) > +expand_BUILTIN_EXPECT (internal_fn, gcall *stmt) > { > /* When guessing was done, the hints should be already stripped away. */ > gcc_assert (!flag_guess_branch_prob || optimize == 0 || seen_error ()); > @@ -1964,7 +1964,7 @@ expand_BUILTIN_EXPECT (gcall *stmt) > should never be called. */ > > static void > -expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) > +expand_VA_ARG (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -1972,7 +1972,7 @@ expand_VA_ARG (gcall *stmt ATTRIBUTE_UNUSED) > /* Expand the IFN_UNIQUE function according to its first argument. */ > > static void > -expand_UNIQUE (gcall *stmt) > +expand_UNIQUE (internal_fn, gcall *stmt) > { > rtx pattern = NULL_RTX; > enum ifn_unique_kind kind > @@ -2018,7 +2018,7 @@ expand_UNIQUE (gcall *stmt) > /* The size of an OpenACC compute dimension. */ > > static void > -expand_GOACC_DIM_SIZE (gcall *stmt) > +expand_GOACC_DIM_SIZE (internal_fn, gcall *stmt) > { > tree lhs = gimple_call_lhs (stmt); > > @@ -2039,7 +2039,7 @@ expand_GOACC_DIM_SIZE (gcall *stmt) > /* The position of an OpenACC execution engine along one compute axis. */ > > static void > -expand_GOACC_DIM_POS (gcall *stmt) > +expand_GOACC_DIM_POS (internal_fn, gcall *stmt) > { > tree lhs = gimple_call_lhs (stmt); > > @@ -2060,7 +2060,7 @@ expand_GOACC_DIM_POS (gcall *stmt) > /* This is expanded by oacc_device_lower pass. */ > > static void > -expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) > +expand_GOACC_LOOP (internal_fn, gcall *) > { > gcc_unreachable (); > } > @@ -2068,20 +2068,20 @@ expand_GOACC_LOOP (gcall *stmt ATTRIBUTE_UNUSED) > /* This is expanded by oacc_device_lower pass. */ > > static void > -expand_GOACC_REDUCTION (gcall *stmt ATTRIBUTE_UNUSED) > +expand_GOACC_REDUCTION (internal_fn, gcall *) > { > gcc_unreachable (); > } > > -/* Expand call STMT using OPTAB, which has a single output operand and > - NARGS input operands. */ > +/* Expand a call to FN using the operands in STMT. FN has a single > + output operand and NARGS input operands. */ > > static void > -expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) > +expand_direct_optab_fn (internal_fn fn, gcall *stmt, direct_optab optab, > + unsigned int nargs) > { > expand_operand *ops = XALLOCAVEC (expand_operand, nargs + 1); > > - internal_fn fn = gimple_call_internal_fn (stmt); > tree_pair types = direct_internal_fn_types (fn, stmt); > insn_code icode = direct_optab_handler (optab, TYPE_MODE (types.first)); > > @@ -2119,11 +2119,11 @@ expand_direct_optab_fn (gcall *stmt, direct_optab optab, unsigned int nargs) > > /* Expanders for optabs that can use expand_direct_optab_fn. */ > > -#define expand_unary_optab_fn(STMT, OPTAB) \ > - expand_direct_optab_fn (STMT, OPTAB, 1) > +#define expand_unary_optab_fn(FN, STMT, OPTAB) \ > + expand_direct_optab_fn (FN, STMT, OPTAB, 1) > > -#define expand_binary_optab_fn(STMT, OPTAB) \ > - expand_direct_optab_fn (STMT, OPTAB, 2) > +#define expand_binary_optab_fn(FN, STMT, OPTAB) \ > + expand_direct_optab_fn (FN, STMT, OPTAB, 2) > > /* RETURN_TYPE and ARGS are a return type and argument list that are > in principle compatible with FN (which satisfies direct_internal_fn_p). > @@ -2219,9 +2219,9 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) > > #define DEF_INTERNAL_OPTAB_FN(CODE, FLAGS, OPTAB, TYPE) \ > static void \ > - expand_##CODE (gcall *stmt) \ > + expand_##CODE (internal_fn fn, gcall *stmt) \ > { \ > - expand_##TYPE##_optab_fn (stmt, OPTAB##_optab); \ > + expand_##TYPE##_optab_fn (fn, stmt, OPTAB##_optab); \ > } > #include "internal-fn.def" > > @@ -2231,16 +2231,24 @@ direct_internal_fn_supported_p (internal_fn fn, tree type) > expand_<NAME> (gcall *stmt) > > where STMT is the statement that performs the call. */ > -static void (*const internal_fn_expanders[]) (gcall *) = { > +static void (*const internal_fn_expanders[]) (internal_fn, gcall *) = { > #define DEF_INTERNAL_FN(CODE, FLAGS, FNSPEC) expand_##CODE, > #include "internal-fn.def" > 0 > }; > > +/* Expand STMT as though it were a call to internal function FN. */ > + > +void > +expand_internal_call (internal_fn fn, gcall *stmt) > +{ > + internal_fn_expanders[fn] (fn, stmt); > +} > + > /* Expand STMT, which is a call to internal function FN. */ > > void > expand_internal_call (gcall *stmt) > { > - internal_fn_expanders[(int) gimple_call_internal_fn (stmt)] (stmt); > + expand_internal_call (gimple_call_internal_fn (stmt), stmt); > } > diff --git a/gcc/internal-fn.h b/gcc/internal-fn.h > index 31e895e..5ee43b8 100644 > --- a/gcc/internal-fn.h > +++ b/gcc/internal-fn.h > @@ -162,5 +162,6 @@ extern bool direct_internal_fn_supported_p (internal_fn, tree_pair); > extern bool direct_internal_fn_supported_p (internal_fn, tree); > > extern void expand_internal_call (gcall *); > +extern void expand_internal_call (internal_fn, gcall *); > > #endif > ^ permalink raw reply [flat|nested] 8+ messages in thread
* Re: Short-cut generation of simple built-in functions 2015-11-07 13:31 Short-cut generation of simple built-in functions Richard Sandiford 2015-11-10 11:29 ` Richard Biener @ 2015-11-22 22:19 ` H.J. Lu 1 sibling, 0 replies; 8+ messages in thread From: H.J. Lu @ 2015-11-22 22:19 UTC (permalink / raw) To: GCC Patches, Richard Sandiford On Sat, Nov 7, 2015 at 5:31 AM, Richard Sandiford <richard.sandiford@arm.com> wrote: > This patch short-circuits the builtins.c expansion code for a particular > gimple call if: > > - the function has an associated internal function > - the target implements that internal function > - the call has no side effects > > This allows a later patch to remove the builtins.c code, once calls with > side effects have been handled. > > Tested on x86_64-linux-gnu, aarch64-linux-gnu and arm-linux-gnueabi. > OK to install? > > Thanks, > Richard > > > gcc/ > * builtins.h (called_as_built_in): Declare. > * builtins.c (called_as_built_in): Make external. > * internal-fn.h (expand_internal_call): Define a variant that > specifies the internal function explicitly. > * internal-fn.c (expand_load_lanes_optab_fn) > (expand_store_lanes_optab_fn, expand_ANNOTATE, expand_GOMP_SIMD_LANE) > (expand_GOMP_SIMD_VF, expand_GOMP_SIMD_LAST_LANE) > (expand_GOMP_SIMD_ORDERED_START, expand_GOMP_SIMD_ORDERED_END) > (expand_UBSAN_NULL, expand_UBSAN_BOUNDS, expand_UBSAN_VPTR) > (expand_UBSAN_OBJECT_SIZE, expand_ASAN_CHECK, expand_TSAN_FUNC_EXIT) > (expand_UBSAN_CHECK_ADD, expand_UBSAN_CHECK_SUB) > (expand_UBSAN_CHECK_MUL, expand_ADD_OVERFLOW, expand_SUB_OVERFLOW) > (expand_MUL_OVERFLOW, expand_LOOP_VECTORIZED) > (expand_mask_load_optab_fn, expand_mask_store_optab_fn) > (expand_ABNORMAL_DISPATCHER, expand_BUILTIN_EXPECT, expand_VA_ARG) > (expand_UNIQUE, expand_GOACC_DIM_SIZE, expand_GOACC_DIM_POS) > (expand_GOACC_LOOP, expand_GOACC_REDUCTION, expand_direct_optab_fn) > (expand_unary_optab_fn, expand_binary_optab_fn): Add an internal_fn > argument. > (internal_fn_expanders): Update prototype. > (expand_internal_call): Define a variant that specifies the > internal function explicitly. Use it to implement the previous > interface. > * cfgexpand.c (expand_call_stmt): Try to expand calls to built-in > functions as calls to internal functions. > This caused: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=68488 H.J. ^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2015-11-22 22:01 UTC | newest] Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed) -- links below jump to the message on this page -- 2015-11-07 13:31 Short-cut generation of simple built-in functions Richard Sandiford 2015-11-10 11:29 ` Richard Biener 2015-11-10 21:24 ` Richard Sandiford 2015-11-11 10:18 ` Richard Biener 2015-11-13 14:32 ` Richard Sandiford 2015-11-17 9:55 ` Richard Sandiford 2015-11-17 14:34 ` Richard Biener 2015-11-22 22:19 ` H.J. Lu
This is a public inbox, see mirroring instructions for how to clone and mirror all data and code used for this inbox; as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).