From 78be79d534fbcf2799c2c4e0b2c768134c9990d4 Mon Sep 17 00:00:00 2001 From: Bill Schmidt Date: Tue, 24 May 2011 18:02:22 +0000 Subject: re PR tree-optimization/46728 (GCC does not generate fmadd for pow (x, 0.75)+y on powerpc) 2011-05-24 Bill Schmidt PR tree-optimization/46728 * tree-ssa-math-opts.c (powi_table): New. (powi_lookup_cost): New. (powi_cost): New. (powi_as_mults_1): New. (powi_as_mults): New. (gimple_expand_builtin_powi): New. (execute_cse_sincos): Add switch case for BUILT_IN_POWI. (gate_cse_sincos): Remove sincos/cexp restriction. From-SVN: r174129 --- gcc/ChangeLog | 12 +++ gcc/tree-ssa-math-opts.c | 263 +++++++++++++++++++++++++++++++++++++++++++++-- 2 files changed, 269 insertions(+), 6 deletions(-) diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 925c24a..1565823 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,15 @@ +2011-05-24 Bill Schmidt + + PR tree-optimization/46728 + * tree-ssa-math-opts.c (powi_table): New. + (powi_lookup_cost): New. + (powi_cost): New. + (powi_as_mults_1): New. + (powi_as_mults): New. + (gimple_expand_builtin_powi): New. + (execute_cse_sincos): Add switch case for BUILT_IN_POWI. + (gate_cse_sincos): Remove sincos/cexp restriction. + 2011-05-24 Rainer Orth PR target/3746 diff --git a/gcc/tree-ssa-math-opts.c b/gcc/tree-ssa-math-opts.c index b9f631e..26cdc69 100644 --- a/gcc/tree-ssa-math-opts.c +++ b/gcc/tree-ssa-math-opts.c @@ -795,8 +795,238 @@ execute_cse_sincos_1 (tree name) return cfg_changed; } +/* To evaluate powi(x,n), the floating point value x raised to the + constant integer exponent n, we use a hybrid algorithm that + combines the "window method" with look-up tables. For an + introduction to exponentiation algorithms and "addition chains", + see section 4.6.3, "Evaluation of Powers" of Donald E. Knuth, + "Seminumerical Algorithms", Vol. 2, "The Art of Computer Programming", + 3rd Edition, 1998, and Daniel M. Gordon, "A Survey of Fast Exponentiation + Methods", Journal of Algorithms, Vol. 27, pp. 129-146, 1998. */ + +/* Provide a default value for POWI_MAX_MULTS, the maximum number of + multiplications to inline before calling the system library's pow + function. powi(x,n) requires at worst 2*bits(n)-2 multiplications, + so this default never requires calling pow, powf or powl. */ + +#ifndef POWI_MAX_MULTS +#define POWI_MAX_MULTS (2*HOST_BITS_PER_WIDE_INT-2) +#endif + +/* The size of the "optimal power tree" lookup table. All + exponents less than this value are simply looked up in the + powi_table below. This threshold is also used to size the + cache of pseudo registers that hold intermediate results. */ +#define POWI_TABLE_SIZE 256 + +/* The size, in bits of the window, used in the "window method" + exponentiation algorithm. This is equivalent to a radix of + (1<= POWI_TABLE_SIZE) + { + if (val & 1) + { + digit = val & ((1 << POWI_WINDOW_SIZE) - 1); + result += powi_lookup_cost (digit, cache) + + POWI_WINDOW_SIZE + 1; + val >>= POWI_WINDOW_SIZE; + } + else + { + val >>= 1; + result++; + } + } + + return result + powi_lookup_cost (val, cache); +} + +/* Recursive subroutine of powi_as_mults. This function takes the + array, CACHE, of already calculated exponents and an exponent N and + returns a tree that corresponds to CACHE[1]**N, with type TYPE. */ + +static tree +powi_as_mults_1 (gimple_stmt_iterator *gsi, location_t loc, tree type, + HOST_WIDE_INT n, tree *cache, tree target) +{ + tree op0, op1, ssa_target; + unsigned HOST_WIDE_INT digit; + gimple mult_stmt; + + if (n < POWI_TABLE_SIZE && cache[n]) + return cache[n]; + + ssa_target = make_ssa_name (target, NULL); + + if (n < POWI_TABLE_SIZE) + { + cache[n] = ssa_target; + op0 = powi_as_mults_1 (gsi, loc, type, n - powi_table[n], cache, target); + op1 = powi_as_mults_1 (gsi, loc, type, powi_table[n], cache, target); + } + else if (n & 1) + { + digit = n & ((1 << POWI_WINDOW_SIZE) - 1); + op0 = powi_as_mults_1 (gsi, loc, type, n - digit, cache, target); + op1 = powi_as_mults_1 (gsi, loc, type, digit, cache, target); + } + else + { + op0 = powi_as_mults_1 (gsi, loc, type, n >> 1, cache, target); + op1 = op0; + } + + mult_stmt = gimple_build_assign_with_ops (MULT_EXPR, ssa_target, op0, op1); + gsi_insert_before (gsi, mult_stmt, GSI_SAME_STMT); + + return ssa_target; +} + +/* Convert ARG0**N to a tree of multiplications of ARG0 with itself. + This function needs to be kept in sync with powi_cost above. */ + +static tree +powi_as_mults (gimple_stmt_iterator *gsi, location_t loc, + tree arg0, HOST_WIDE_INT n) +{ + tree cache[POWI_TABLE_SIZE], result, type = TREE_TYPE (arg0), target; + gimple div_stmt; + + if (n == 0) + return build_real (type, dconst1); + + memset (cache, 0, sizeof (cache)); + cache[1] = arg0; + + target = create_tmp_var (type, "powmult"); + add_referenced_var (target); + + result = powi_as_mults_1 (gsi, loc, type, (n < 0) ? -n : n, cache, target); + + if (n >= 0) + return result; + + /* If the original exponent was negative, reciprocate the result. */ + target = make_ssa_name (target, NULL); + div_stmt = gimple_build_assign_with_ops (RDIV_EXPR, target, + build_real (type, dconst1), + result); + gsi_insert_before (gsi, div_stmt, GSI_SAME_STMT); + + return target; +} + +/* ARG0 and N are the two arguments to a powi builtin in GSI with + location info LOC. If the arguments are appropriate, create an + equivalent sequence of statements prior to GSI using an optimal + number of multiplications, and return an expession holding the + result. */ + +static tree +gimple_expand_builtin_powi (gimple_stmt_iterator *gsi, location_t loc, + tree arg0, HOST_WIDE_INT n) +{ + /* Avoid largest negative number. */ + if (n != -n + && ((n >= -1 && n <= 2) + || (optimize_function_for_speed_p (cfun) + && powi_cost (n) <= POWI_MAX_MULTS))) + return powi_as_mults (gsi, loc, arg0, n); + + return NULL_TREE; +} + /* Go through all calls to sin, cos and cexpi and call execute_cse_sincos_1 - on the SSA_NAME argument of each of them. */ + on the SSA_NAME argument of each of them. Also expand powi(x,n) into + an optimal number of multiplies, when n is a constant. */ static unsigned int execute_cse_sincos (void) @@ -821,7 +1051,9 @@ execute_cse_sincos (void) && (fndecl = gimple_call_fndecl (stmt)) && DECL_BUILT_IN_CLASS (fndecl) == BUILT_IN_NORMAL) { - tree arg; + tree arg, arg0, arg1, result; + HOST_WIDE_INT n; + location_t loc; switch (DECL_FUNCTION_CODE (fndecl)) { @@ -833,6 +1065,26 @@ execute_cse_sincos (void) cfg_changed |= execute_cse_sincos_1 (arg); break; + CASE_FLT_FN (BUILT_IN_POWI): + arg0 = gimple_call_arg (stmt, 0); + arg1 = gimple_call_arg (stmt, 1); + if (!host_integerp (arg1, 0)) + break; + + n = TREE_INT_CST_LOW (arg1); + loc = gimple_location (stmt); + result = gimple_expand_builtin_powi (&gsi, loc, arg0, n); + + if (result) + { + tree lhs = gimple_get_lhs (stmt); + gimple new_stmt = gimple_build_assign (lhs, result); + gimple_set_location (new_stmt, loc); + unlink_stmt_vdef (stmt); + gsi_replace (&gsi, new_stmt, true); + } + break; + default:; } } @@ -849,10 +1101,9 @@ execute_cse_sincos (void) static bool gate_cse_sincos (void) { - /* Make sure we have either sincos or cexp. */ - return (TARGET_HAS_SINCOS - || TARGET_C99_FUNCTIONS) - && optimize; + /* We no longer require either sincos or cexp, since powi expansion + piggybacks on this pass. */ + return optimize; } struct gimple_opt_pass pass_cse_sincos = -- cgit v1.1