Message ID | 061801d86ad8$13976a00$3ac63e00$@nextmovesoftware.com |
---|---|
State | New |
Headers | show |
Series | [take,#2] PR middle-end/98865: Expand X*Y as X&-Y when Y is [0.1]. | expand |
On Wed, May 18, 2022 at 6:55 PM Roger Sayle <roger@nextmovesoftware.com> wrote: > > > The patch is a revised solution for PR middle-end/98865 incorporating > the feedback/suggestions from Richard Biener's review here: > https://gcc.gnu.org/pipermail/gcc-patches/2022-May/593928.html > Most significantly, this patch now performs the transformation/optimization > during RTL expansion, where the target's rtx_costs can be used to determine > whether the original multiplication (that may be possibly be implemented by > a shift or lea) is cheaper than a negation and a bit-wise and. > > Previously the expression (x>>63)*y would be compiled with -O2 as > shrq $63, %rdi > movq %rdi, %rax > imulq %rsi, %rax > > but with this patch now produces: > sarq $63, %rdi > movq %rdi, %rax > andq %rsi, %rax > > Likewise the expression (x>>63)*135 [that appears in a hot-spot of the > Botan AES-128 benchmark] was previously: > > shrq $63, %rdi > leaq (%rdi,%rdi,8), %rdx > movq %rdx, %rax > salq $4, %rax > subq %rdx, %rax > > now becomes: > movq %rdi, %rax > sarq $63, %rax > andl $135, %eax > > > This patch has been tested on x86_64-pc-linux-gnu with make bootstrap and > make -k check, both with and without --target_board=unix{-m32}, with no > new failures. Many thanks to Uros for the speedy review and approval of > my x86_rtx_costs patch that enables this transformation on -m32 using the > correct cost of DImode multiplication. Ok for mainline? Looks good to me Thanks, Richard. > 2022-05-18 Roger Sayle <roger@nextmovesoftware.com> > > gcc/ChangeLog > PR middle-end/98865 > * expr.cc (expand_expr_real_2) [MULT_EXPR]: Expand X*Y as X&Y > when both X and Y are [0, 1], X*Y as X&-Y when Y is [0,1] and > likewise X*Y as -X&Y when X is [0,1] using tree_nonzero_bits. > > gcc/testsuite/ChangeLog > PR middle-end/98865 > * gcc.target/i386/pr98865.c: New test case. > > > Thanks in advance, > Roger > -- >
diff --git a/gcc/expr.cc b/gcc/expr.cc index 1806091..7197996 100644 --- a/gcc/expr.cc +++ b/gcc/expr.cc @@ -9541,6 +9541,38 @@ expand_expr_real_2 (sepops ops, rtx target, machine_mode tmode, } expand_operands (treeop0, treeop1, subtarget, &op0, &op1, EXPAND_NORMAL); + + /* Expand X*Y as X&-Y when Y must be zero or one. */ + if (SCALAR_INT_MODE_P (mode)) + { + bool bit0_p = tree_nonzero_bits (treeop0) == 1; + bool bit1_p = tree_nonzero_bits (treeop1) == 1; + + /* Expand X*Y as X&Y when both X and Y must be zero or one. */ + if (bit0_p && bit1_p) + return REDUCE_BIT_FIELD (expand_and (mode, op0, op1, target)); + + if (bit0_p || bit1_p) + { + bool speed = optimize_insn_for_speed_p (); + int cost = add_cost (speed, mode) + neg_cost (speed, mode); + struct algorithm algorithm; + enum mult_variant variant; + if (CONST_INT_P (op1) + ? !choose_mult_variant (mode, INTVAL (op1), + &algorithm, &variant, cost) + : cost < mul_cost (speed, mode)) + { + target = bit0_p ? expand_and (mode, negate_rtx (mode, op0), + op1, target) + : expand_and (mode, op0, + negate_rtx (mode, op1), + target); + return REDUCE_BIT_FIELD (target); + } + } + } + return REDUCE_BIT_FIELD (expand_mult (mode, op0, op1, target, unsignedp)); case TRUNC_MOD_EXPR: diff --git a/gcc/testsuite/gcc.target/i386/pr98865.c b/gcc/testsuite/gcc.target/i386/pr98865.c new file mode 100644 index 0000000..d047c4b --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/pr98865.c @@ -0,0 +1,54 @@ +/* { dg-do compile } */ +/* { dg-options "-O2 -fdump-tree-optimized" } */ + +#if __SIZEOF_INT__ == 4 +unsigned int foo(unsigned int a, unsigned int b) +{ + return (a >> 31) * b; +} + +int bar(int a, int b) +{ + return -(a >> 31) * b; +} + +int baz(int a, int b) +{ + int c = a >> 31; + int d = -c; + return d * b; +} + +unsigned int pin(int a, unsigned int b) +{ + unsigned int t = a & 1; + return t * b; +} +#endif + +#if __SIZEOF_LONG_LONG__ == 8 +unsigned long long fool(unsigned long long a, unsigned long long b) +{ + return (a >> 63) * b; +} + +long long barl (long long a, long long b) +{ + return -(a >> 63) * b; +} + +long long bazl (long long a, long long b) +{ + long long c = a >> 63; + long long d = -c; + return d * b; +} + +unsigned long long pinl(long long a, unsigned long long b) +{ + unsigned long long t = a & 1; + return t * b; +} +#endif + +/* { dg-final { scan-assembler-not "imul" } } */