Message ID | 20231019061422.281377-1-hongtao.liu@intel.com |
---|---|
State | New |
Headers | show |
Series | Avoid compile time hog on vect_peel_nonlinear_iv_init for nonlinear induction vec_step_op_mul when iteration count is too big. | expand |
On Thu, Oct 19, 2023 at 8:16 AM liuhongt <hongtao.liu@intel.com> wrote: > > >So the bugs were not fixed without this hunk? IIRC in the audit > >trail we concluded the value is always positive ... (but of course > >a large unsigned value can appear negative if you test it this way?) > No, I added this incase in the future there's negative skip_niters as > you mentioned in the PR, it's just defensive programming. > > >I think you can use one of the mpz_pow* functions and > >wi::to_mpz/from_mpz for this. See tree-ssa-loop-niter.cc for the > >most heavy user of mpz (but not pow I think). > Changed. > > Bootstrapped and regtested on x86_64-pc-linux-gnu{-m32,}. > Ok for trunk. > > There's loop in vect_peel_nonlinear_iv_init to get init_expr * > pow (step_expr, skip_niters). When skipn_iters is too big, compile time > hogs. To avoid that, optimize init_expr * pow (step_expr, skip_niters) to > init_expr << (exact_log2 (step_expr) * skip_niters) when step_expr is > pow of 2, otherwise give up vectorization when skip_niters >= > TYPE_PRECISION (TREE_TYPE (init_expr)). > > Also give up vectorization when niters_skip is negative which will be > used for fully masked loop. > > gcc/ChangeLog: > > PR tree-optimization/111820 > PR tree-optimization/111833 > * tree-vect-loop-manip.cc (vect_can_peel_nonlinear_iv_p): Give > up vectorization for nonlinear iv vect_step_op_mul when > step_expr is not exact_log2 and niters is greater than > TYPE_PRECISION (TREE_TYPE (step_expr)). Also don't vectorize > for nagative niters_skip which will be used by fully masked > loop. > (vect_can_advance_ivs_p): Pass whole phi_info to > vect_can_peel_nonlinear_iv_p. > * tree-vect-loop.cc (vect_peel_nonlinear_iv_init): Optimize > init_expr * pow (step_expr, skipn) to init_expr > << (log2 (step_expr) * skipn) when step_expr is exact_log2. > > gcc/testsuite/ChangeLog: > > * gcc.target/i386/pr111820-1.c: New test. > * gcc.target/i386/pr111820-2.c: New test. > * gcc.target/i386/pr111820-3.c: New test. > * gcc.target/i386/pr103144-mul-1.c: Adjust testcase. > * gcc.target/i386/pr103144-mul-2.c: Adjust testcase. > --- > .../gcc.target/i386/pr103144-mul-1.c | 8 ++--- > .../gcc.target/i386/pr103144-mul-2.c | 8 ++--- > gcc/testsuite/gcc.target/i386/pr111820-1.c | 16 +++++++++ > gcc/testsuite/gcc.target/i386/pr111820-2.c | 16 +++++++++ > gcc/testsuite/gcc.target/i386/pr111820-3.c | 16 +++++++++ > gcc/tree-vect-loop-manip.cc | 28 +++++++++++++-- > gcc/tree-vect-loop.cc | 34 ++++++++++++++++--- > 7 files changed, 110 insertions(+), 16 deletions(-) > create mode 100644 gcc/testsuite/gcc.target/i386/pr111820-1.c > create mode 100644 gcc/testsuite/gcc.target/i386/pr111820-2.c > create mode 100644 gcc/testsuite/gcc.target/i386/pr111820-3.c > > diff --git a/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c b/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c > index 640c34fd959..913d7737dcd 100644 > --- a/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c > +++ b/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c > @@ -11,7 +11,7 @@ foo_mul (int* a, int b) > for (int i = 0; i != N; i++) > { > a[i] = b; > - b *= 3; > + b *= 4; > } > } > > @@ -23,7 +23,7 @@ foo_mul_const (int* a) > for (int i = 0; i != N; i++) > { > a[i] = b; > - b *= 3; > + b *= 4; > } > } > > @@ -34,7 +34,7 @@ foo_mul_peel (int* a, int b) > for (int i = 0; i != 39; i++) > { > a[i] = b; > - b *= 3; > + b *= 4; > } > } > > @@ -46,6 +46,6 @@ foo_mul_peel_const (int* a) > for (int i = 0; i != 39; i++) > { > a[i] = b; > - b *= 3; > + b *= 4; > } > } > diff --git a/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c b/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c > index 39fdea3a69d..b2ff186e335 100644 > --- a/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c > +++ b/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c > @@ -16,12 +16,12 @@ avx2_test (void) > > __builtin_memset (epi32_exp, 0, N * sizeof (int)); > int b = 8; > - v8si init = __extension__(v8si) { b, b * 3, b * 9, b * 27, b * 81, b * 243, b * 729, b * 2187 }; > + v8si init = __extension__(v8si) { b, b * 4, b * 16, b * 64, b * 256, b * 1024, b * 4096, b * 16384 }; > > for (int i = 0; i != N / 8; i++) > { > memcpy (epi32_exp + i * 8, &init, 32); > - init *= 6561; > + init *= 65536; > } > > foo_mul (epi32_dst, b); > @@ -32,11 +32,11 @@ avx2_test (void) > if (__builtin_memcmp (epi32_dst, epi32_exp, 39 * 4) != 0) > __builtin_abort (); > > - init = __extension__(v8si) { 1, 3, 9, 27, 81, 243, 729, 2187 }; > + init = __extension__(v8si) { 1, 4, 16, 64, 256, 1024, 4096, 16384 }; > for (int i = 0; i != N / 8; i++) > { > memcpy (epi32_exp + i * 8, &init, 32); > - init *= 6561; > + init *= 65536; > } > > foo_mul_const (epi32_dst); > diff --git a/gcc/testsuite/gcc.target/i386/pr111820-1.c b/gcc/testsuite/gcc.target/i386/pr111820-1.c > new file mode 100644 > index 00000000000..50e960c39d4 > --- /dev/null > +++ b/gcc/testsuite/gcc.target/i386/pr111820-1.c > @@ -0,0 +1,16 @@ > +/* { dg-do compile } */ > +/* { dg-options "-O3 -mavx2 -fno-tree-vrp -Wno-aggressive-loop-optimizations -fdump-tree-vect-details" } */ > +/* { dg-final { scan-tree-dump "Avoid compile time hog on vect_peel_nonlinear_iv_init for nonlinear induction vec_step_op_mul when iteration count is too big" "vect" } } */ > + > +int r; > +int r_0; > + > +void f1 (void) > +{ > + int n = 0; > + while (-- n) > + { > + r_0 += r; > + r *= 3; > + } > +} > diff --git a/gcc/testsuite/gcc.target/i386/pr111820-2.c b/gcc/testsuite/gcc.target/i386/pr111820-2.c > new file mode 100644 > index 00000000000..dbeceb228c3 > --- /dev/null > +++ b/gcc/testsuite/gcc.target/i386/pr111820-2.c > @@ -0,0 +1,16 @@ > +/* { dg-do compile } */ > +/* { dg-options "-O3 -mavx2 -fno-tree-vrp -fdump-tree-vect-details -Wno-aggressive-loop-optimizations" } */ > +/* { dg-final { scan-tree-dump "LOOP VECTORIZED" "vect" } } */ > + > +int r; > +int r_0; > + > +void f (void) > +{ > + int n = 0; > + while (-- n) > + { > + r_0 += r ; > + r *= 2; > + } > +} > diff --git a/gcc/testsuite/gcc.target/i386/pr111820-3.c b/gcc/testsuite/gcc.target/i386/pr111820-3.c > new file mode 100644 > index 00000000000..b778f517663 > --- /dev/null > +++ b/gcc/testsuite/gcc.target/i386/pr111820-3.c > @@ -0,0 +1,16 @@ > +/* { dg-do compile } */ > +/* { dg-options "-O3 -mavx2 -fno-tree-vrp -fdump-tree-vect-details -Wno-aggressive-loop-optimizations" } */ > +/* { dg-final { scan-tree-dump "LOOP VECTORIZED" "vect" } } */ > + > +int r; > +int r_0; > + > +void f (void) > +{ > + int n = 14; > + while (-- n) > + { > + r_0 += r ; > + r *= 3; > + } > +} > diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc > index 2608c286e5d..a530088b61d 100644 > --- a/gcc/tree-vect-loop-manip.cc > +++ b/gcc/tree-vect-loop-manip.cc > @@ -1783,8 +1783,10 @@ iv_phi_p (stmt_vec_info stmt_info) > /* Return true if vectorizer can peel for nonlinear iv. */ > static bool > vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, > - enum vect_induction_op_type induction_type) > + stmt_vec_info stmt_info) > { > + enum vect_induction_op_type induction_type > + = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (stmt_info); > tree niters_skip; > /* Init_expr will be update by vect_update_ivs_after_vectorizer, > if niters or vf is unkown: > @@ -1805,11 +1807,31 @@ vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, > return false; > } > > + /* Avoid compile time hog on vect_peel_nonlinear_iv_init. */ > + if (induction_type == vect_step_op_mul) > + { > + tree step_expr = STMT_VINFO_LOOP_PHI_EVOLUTION_PART (stmt_info); > + tree type = TREE_TYPE (step_expr); > + > + if (wi::exact_log2 (wi::to_wide (step_expr)) == -1 > + && LOOP_VINFO_INT_NITERS(loop_vinfo) >= TYPE_PRECISION (type)) So with pow being available this limit shouldn't be necessary any more and the testcase adjustment can be avoided? > + { > + if (dump_enabled_p ()) > + dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, > + "Avoid compile time hog on" > + " vect_peel_nonlinear_iv_init" > + " for nonlinear induction vec_step_op_mul" > + " when iteration count is too big.\n"); > + return false; > + } > + } > + > /* Also doens't support peel for neg when niter is variable. > ??? generate something like niter_expr & 1 ? init_expr : -init_expr? */ > niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); > if ((niters_skip != NULL_TREE > - && TREE_CODE (niters_skip) != INTEGER_CST) > + && (TREE_CODE (niters_skip) != INTEGER_CST > + || (HOST_WIDE_INT) TREE_INT_CST_LOW (niters_skip) < 0)) > || (!vect_use_loop_mask_for_alignment_p (loop_vinfo) > && LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0)) > { > @@ -1870,7 +1892,7 @@ vect_can_advance_ivs_p (loop_vec_info loop_vinfo) > induction_type = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (phi_info); > if (induction_type != vect_step_op_add) > { > - if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, induction_type)) > + if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, phi_info)) > return false; > > continue; > diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc > index 89bdcaa0910..f0dbba50786 100644 > --- a/gcc/tree-vect-loop.cc > +++ b/gcc/tree-vect-loop.cc > @@ -9134,11 +9134,35 @@ vect_peel_nonlinear_iv_init (gimple_seq* stmts, tree init_expr, > init_expr = gimple_convert (stmts, utype, init_expr); > unsigned skipn = TREE_INT_CST_LOW (skip_niters); > wide_int begin = wi::to_wide (step_expr); > - for (unsigned i = 0; i != skipn - 1; i++) > - begin = wi::mul (begin, wi::to_wide (step_expr)); > - tree mult_expr = wide_int_to_tree (utype, begin); > - init_expr = gimple_build (stmts, MULT_EXPR, utype, init_expr, mult_expr); > - init_expr = gimple_convert (stmts, type, init_expr); > + int pow2_step = wi::exact_log2 (begin); > + /* Optimize init_expr * pow (step_expr, skipn) to > + init_expr << (log2 (step_expr) * skipn). */ > + if (pow2_step != -1) and to avoid undefined behavior with too large shift just go the gmp way unconditionally. > + { > + if (skipn >= TYPE_PRECISION (type) > + || skipn > (UINT_MAX / (unsigned) pow2_step) > + || skipn * (unsigned) pow2_step >= TYPE_PRECISION (type)) > + init_expr = build_zero_cst (type); > + else > + { > + tree lshc = build_int_cst (utype, skipn * (unsigned) pow2_step); > + init_expr = gimple_build (stmts, LSHIFT_EXPR, utype, > + init_expr, lshc); > + } > + } > + /* Any better way for init_expr * pow (step_expr, skipn)???. */ this comment is now resolved I think. > + else > + { > + gcc_assert (skipn < TYPE_PRECISION (type)); > + auto_mpz base, exp; > + wi::to_mpz (begin, base, TYPE_SIGN (type)); > + mpz_pow_ui (exp, base, skipn); mpz_pow_ui uses unsigned long while i think we constrain known niters to uint64 - so I suggest to use mpz_powm instead (limiting to a possibly host specific limit - unsigned long - is unfortunately a no-go). Otherwise looks good to me. Thanks, Richard. > + begin = wi::from_mpz (type, exp, TYPE_SIGN (type)); > + tree mult_expr = wide_int_to_tree (utype, begin); > + init_expr = gimple_build (stmts, MULT_EXPR, utype, > + init_expr, mult_expr); > + } > + init_expr = gimple_convert (stmts, type, init_expr); > } > break; > > -- > 2.31.1 >
diff --git a/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c b/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c index 640c34fd959..913d7737dcd 100644 --- a/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c +++ b/gcc/testsuite/gcc.target/i386/pr103144-mul-1.c @@ -11,7 +11,7 @@ foo_mul (int* a, int b) for (int i = 0; i != N; i++) { a[i] = b; - b *= 3; + b *= 4; } } @@ -23,7 +23,7 @@ foo_mul_const (int* a) for (int i = 0; i != N; i++) { a[i] = b; - b *= 3; + b *= 4; } } @@ -34,7 +34,7 @@ foo_mul_peel (int* a, int b) for (int i = 0; i != 39; i++) { a[i] = b; - b *= 3; + b *= 4; } } @@ -46,6 +46,6 @@ foo_mul_peel_const (int* a) for (int i = 0; i != 39; i++) { a[i] = b; - b *= 3; + b *= 4; } } diff --git a/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c b/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c index 39fdea3a69d..b2ff186e335 100644 --- a/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c +++ b/gcc/testsuite/gcc.target/i386/pr103144-mul-2.c @@ -16,12 +16,12 @@ avx2_test (void) __builtin_memset (epi32_exp, 0, N * sizeof (int)); int b = 8; - v8si init = __extension__(v8si) { b, b * 3, b * 9, b * 27, b * 81, b * 243, b * 729, b * 2187 }; + v8si init = __extension__(v8si) { b, b * 4, b * 16, b * 64, b * 256, b * 1024, b * 4096, b * 16384 }; for (int i = 0; i != N / 8; i++) { memcpy (epi32_exp + i * 8, &init, 32); - init *= 6561; + init *= 65536; } foo_mul (epi32_dst, b); @@ -32,11 +32,11 @@ avx2_test (void) if (__builtin_memcmp (epi32_dst, epi32_exp, 39 * 4) != 0) __builtin_abort (); - init = __extension__(v8si) { 1, 3, 9, 27, 81, 243, 729, 2187 }; + init = __extension__(v8si) { 1, 4, 16, 64, 256, 1024, 4096, 16384 }; for (int i = 0; i != N / 8; i++) { memcpy (epi32_exp + i * 8, &init, 32); - init *= 6561; + init *= 65536; } foo_mul_const (epi32_dst); diff --git a/gcc/testsuite/gcc.target/i386/pr111820-1.c b/gcc/testsuite/gcc.target/i386/pr111820-1.c new file mode 100644 index 00000000000..50e960c39d4 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr111820-1.c @@ -0,0 +1,16 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -mavx2 -fno-tree-vrp -Wno-aggressive-loop-optimizations -fdump-tree-vect-details" } */ +/* { dg-final { scan-tree-dump "Avoid compile time hog on vect_peel_nonlinear_iv_init for nonlinear induction vec_step_op_mul when iteration count is too big" "vect" } } */ + +int r; +int r_0; + +void f1 (void) +{ + int n = 0; + while (-- n) + { + r_0 += r; + r *= 3; + } +} diff --git a/gcc/testsuite/gcc.target/i386/pr111820-2.c b/gcc/testsuite/gcc.target/i386/pr111820-2.c new file mode 100644 index 00000000000..dbeceb228c3 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr111820-2.c @@ -0,0 +1,16 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -mavx2 -fno-tree-vrp -fdump-tree-vect-details -Wno-aggressive-loop-optimizations" } */ +/* { dg-final { scan-tree-dump "LOOP VECTORIZED" "vect" } } */ + +int r; +int r_0; + +void f (void) +{ + int n = 0; + while (-- n) + { + r_0 += r ; + r *= 2; + } +} diff --git a/gcc/testsuite/gcc.target/i386/pr111820-3.c b/gcc/testsuite/gcc.target/i386/pr111820-3.c new file mode 100644 index 00000000000..b778f517663 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr111820-3.c @@ -0,0 +1,16 @@ +/* { dg-do compile } */ +/* { dg-options "-O3 -mavx2 -fno-tree-vrp -fdump-tree-vect-details -Wno-aggressive-loop-optimizations" } */ +/* { dg-final { scan-tree-dump "LOOP VECTORIZED" "vect" } } */ + +int r; +int r_0; + +void f (void) +{ + int n = 14; + while (-- n) + { + r_0 += r ; + r *= 3; + } +} diff --git a/gcc/tree-vect-loop-manip.cc b/gcc/tree-vect-loop-manip.cc index 2608c286e5d..a530088b61d 100644 --- a/gcc/tree-vect-loop-manip.cc +++ b/gcc/tree-vect-loop-manip.cc @@ -1783,8 +1783,10 @@ iv_phi_p (stmt_vec_info stmt_info) /* Return true if vectorizer can peel for nonlinear iv. */ static bool vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, - enum vect_induction_op_type induction_type) + stmt_vec_info stmt_info) { + enum vect_induction_op_type induction_type + = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (stmt_info); tree niters_skip; /* Init_expr will be update by vect_update_ivs_after_vectorizer, if niters or vf is unkown: @@ -1805,11 +1807,31 @@ vect_can_peel_nonlinear_iv_p (loop_vec_info loop_vinfo, return false; } + /* Avoid compile time hog on vect_peel_nonlinear_iv_init. */ + if (induction_type == vect_step_op_mul) + { + tree step_expr = STMT_VINFO_LOOP_PHI_EVOLUTION_PART (stmt_info); + tree type = TREE_TYPE (step_expr); + + if (wi::exact_log2 (wi::to_wide (step_expr)) == -1 + && LOOP_VINFO_INT_NITERS(loop_vinfo) >= TYPE_PRECISION (type)) + { + if (dump_enabled_p ()) + dump_printf_loc (MSG_MISSED_OPTIMIZATION, vect_location, + "Avoid compile time hog on" + " vect_peel_nonlinear_iv_init" + " for nonlinear induction vec_step_op_mul" + " when iteration count is too big.\n"); + return false; + } + } + /* Also doens't support peel for neg when niter is variable. ??? generate something like niter_expr & 1 ? init_expr : -init_expr? */ niters_skip = LOOP_VINFO_MASK_SKIP_NITERS (loop_vinfo); if ((niters_skip != NULL_TREE - && TREE_CODE (niters_skip) != INTEGER_CST) + && (TREE_CODE (niters_skip) != INTEGER_CST + || (HOST_WIDE_INT) TREE_INT_CST_LOW (niters_skip) < 0)) || (!vect_use_loop_mask_for_alignment_p (loop_vinfo) && LOOP_VINFO_PEELING_FOR_ALIGNMENT (loop_vinfo) < 0)) { @@ -1870,7 +1892,7 @@ vect_can_advance_ivs_p (loop_vec_info loop_vinfo) induction_type = STMT_VINFO_LOOP_PHI_EVOLUTION_TYPE (phi_info); if (induction_type != vect_step_op_add) { - if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, induction_type)) + if (!vect_can_peel_nonlinear_iv_p (loop_vinfo, phi_info)) return false; continue; diff --git a/gcc/tree-vect-loop.cc b/gcc/tree-vect-loop.cc index 89bdcaa0910..f0dbba50786 100644 --- a/gcc/tree-vect-loop.cc +++ b/gcc/tree-vect-loop.cc @@ -9134,11 +9134,35 @@ vect_peel_nonlinear_iv_init (gimple_seq* stmts, tree init_expr, init_expr = gimple_convert (stmts, utype, init_expr); unsigned skipn = TREE_INT_CST_LOW (skip_niters); wide_int begin = wi::to_wide (step_expr); - for (unsigned i = 0; i != skipn - 1; i++) - begin = wi::mul (begin, wi::to_wide (step_expr)); - tree mult_expr = wide_int_to_tree (utype, begin); - init_expr = gimple_build (stmts, MULT_EXPR, utype, init_expr, mult_expr); - init_expr = gimple_convert (stmts, type, init_expr); + int pow2_step = wi::exact_log2 (begin); + /* Optimize init_expr * pow (step_expr, skipn) to + init_expr << (log2 (step_expr) * skipn). */ + if (pow2_step != -1) + { + if (skipn >= TYPE_PRECISION (type) + || skipn > (UINT_MAX / (unsigned) pow2_step) + || skipn * (unsigned) pow2_step >= TYPE_PRECISION (type)) + init_expr = build_zero_cst (type); + else + { + tree lshc = build_int_cst (utype, skipn * (unsigned) pow2_step); + init_expr = gimple_build (stmts, LSHIFT_EXPR, utype, + init_expr, lshc); + } + } + /* Any better way for init_expr * pow (step_expr, skipn)???. */ + else + { + gcc_assert (skipn < TYPE_PRECISION (type)); + auto_mpz base, exp; + wi::to_mpz (begin, base, TYPE_SIGN (type)); + mpz_pow_ui (exp, base, skipn); + begin = wi::from_mpz (type, exp, TYPE_SIGN (type)); + tree mult_expr = wide_int_to_tree (utype, begin); + init_expr = gimple_build (stmts, MULT_EXPR, utype, + init_expr, mult_expr); + } + init_expr = gimple_convert (stmts, type, init_expr); } break;