From 2d977ff983f86b3f7b7c9f1cb5a479af75cd589e Mon Sep 17 00:00:00 2001 From: Jakub Jelinek Date: Tue, 10 Mar 2015 22:03:21 +0100 Subject: [PATCH] re PR target/65368 (_bzhi_u32 intrinsic generates incorrect code when -O1 or above is specified and index is an immediate) PR target/65368 * config/i386/i386.md (bmi2_bzhi_3): Removed define_insn, new define_expand. (*bmi2_bzhi_3, *bmi2_bzhi_3_1): New define_insns. * gcc.target/i386/bmi2-bzhi-2.c: New test. From-SVN: r221335 --- gcc/ChangeLog | 7 +++ gcc/config/i386/i386.md | 44 ++++++++++++-- gcc/testsuite/ChangeLog | 5 ++ gcc/testsuite/gcc.target/i386/bmi2-bzhi-2.c | 67 +++++++++++++++++++++ 4 files changed, 118 insertions(+), 5 deletions(-) create mode 100644 gcc/testsuite/gcc.target/i386/bmi2-bzhi-2.c diff --git a/gcc/ChangeLog b/gcc/ChangeLog index 6802d207a39..b77fb5e99d0 100644 --- a/gcc/ChangeLog +++ b/gcc/ChangeLog @@ -1,3 +1,10 @@ +2015-03-10 Jakub Jelinek + + PR target/65368 + * config/i386/i386.md (bmi2_bzhi_3): Removed define_insn, + new define_expand. + (*bmi2_bzhi_3, *bmi2_bzhi_3_1): New define_insns. + 2015-03-10 Jan Hubicka * ipa-icf.c (sem_function::equals_wpa): Move here some checks diff --git a/gcc/config/i386/i386.md b/gcc/config/i386/i386.md index 8a80415b9a5..1129b935a17 100644 --- a/gcc/config/i386/i386.md +++ b/gcc/config/i386/i386.md @@ -12678,18 +12678,52 @@ (set_attr "mode" "")]) ;; BMI2 instructions. -(define_insn "bmi2_bzhi_3" +(define_expand "bmi2_bzhi_3" + [(parallel + [(set (match_operand:SWI48 0 "register_operand") + (zero_extract:SWI48 + (match_operand:SWI48 1 "nonimmediate_operand") + (umin:SWI48 + (and:SWI48 (match_operand:SWI48 2 "register_operand") + (const_int 255)) + (match_dup 3)) + (const_int 0))) + (clobber (reg:CC FLAGS_REG))])] + "TARGET_BMI2" + "operands[3] = GEN_INT ( * BITS_PER_UNIT);") + +(define_insn "*bmi2_bzhi_3" [(set (match_operand:SWI48 0 "register_operand" "=r") - (and:SWI48 (lshiftrt:SWI48 (const_int -1) - (match_operand:SWI48 2 "register_operand" "r")) - (match_operand:SWI48 1 "nonimmediate_operand" "rm"))) + (zero_extract:SWI48 + (match_operand:SWI48 1 "nonimmediate_operand" "rm") + (umin:SWI48 + (and:SWI48 (match_operand:SWI48 2 "register_operand" "r") + (const_int 255)) + (match_operand:SWI48 3 "const_int_operand" "n")) + (const_int 0))) (clobber (reg:CC FLAGS_REG))] - "TARGET_BMI2" + "TARGET_BMI2 && INTVAL (operands[3]) == * BITS_PER_UNIT" "bzhi\t{%2, %1, %0|%0, %1, %2}" [(set_attr "type" "bitmanip") (set_attr "prefix" "vex") (set_attr "mode" "")]) +(define_mode_attr k [(SI "k") (DI "q")]) +(define_insn "*bmi2_bzhi_3_1" + [(set (match_operand:SWI48 0 "register_operand" "=r") + (zero_extract:SWI48 + (match_operand:SWI48 1 "nonimmediate_operand" "rm") + (umin:SWI48 + (zero_extend:SWI48 (match_operand:QI 2 "register_operand" "r")) + (match_operand:SWI48 3 "const_int_operand" "n")) + (const_int 0))) + (clobber (reg:CC FLAGS_REG))] + "TARGET_BMI2 && INTVAL (operands[3]) == * BITS_PER_UNIT" + "bzhi\t{%2, %1, %0|%0, %1, %2}" + [(set_attr "type" "bitmanip") + (set_attr "prefix" "vex") + (set_attr "mode" "")]) + (define_insn "bmi2_pdep_3" [(set (match_operand:SWI48 0 "register_operand" "=r") (unspec:SWI48 [(match_operand:SWI48 1 "register_operand" "r") diff --git a/gcc/testsuite/ChangeLog b/gcc/testsuite/ChangeLog index 5acef589ff2..2a201e18ff6 100644 --- a/gcc/testsuite/ChangeLog +++ b/gcc/testsuite/ChangeLog @@ -1,3 +1,8 @@ +2015-03-10 Jakub Jelinek + + PR target/65368 + * gcc.target/i386/bmi2-bzhi-2.c: New test. + 2015-03-10 Paul Thomas PR fortran/65024 diff --git a/gcc/testsuite/gcc.target/i386/bmi2-bzhi-2.c b/gcc/testsuite/gcc.target/i386/bmi2-bzhi-2.c new file mode 100644 index 00000000000..34579d52695 --- /dev/null +++ b/gcc/testsuite/gcc.target/i386/bmi2-bzhi-2.c @@ -0,0 +1,67 @@ +/* PR target/65368 */ +/* { dg-do assemble { target bmi2 } } */ +/* { dg-options "-O2 -mbmi2" } */ + +#include +#include "bmi2-check.h" + +unsigned int a; +unsigned long long b; + +#define A __attribute__((noinline, noclone)) + +A unsigned int f1 (void) { return _bzhi_u32 (a, 0); } +A unsigned int f2 (unsigned int x) { return _bzhi_u32 (x, 0); } +A unsigned int f3 (void) { return _bzhi_u32 (a, 5); } +A unsigned int f4 (unsigned int x) { return _bzhi_u32 (x, 5); } +A unsigned int f5 (void) { return _bzhi_u32 (a, 31); } +A unsigned int f6 (unsigned int x) { return _bzhi_u32 (x, 31); } +A unsigned int f7 (void) { return _bzhi_u32 (a, 32); } +A unsigned int f8 (unsigned int x) { return _bzhi_u32 (x, 32); } +A unsigned int f9 (void) { return _bzhi_u32 (a, 37); } +A unsigned int f10 (unsigned int x) { return _bzhi_u32 (x, 37); } +A unsigned int f11 (void) { return _bzhi_u32 (a, 257); } +A unsigned int f12 (unsigned int x) { return _bzhi_u32 (x, 257); } +A unsigned int f13 (void) { return _bzhi_u32 (a, 289); } +A unsigned int f14 (unsigned int x) { return _bzhi_u32 (x, 289); } +#ifdef __x86_64__ +A unsigned long long f21 (void) { return _bzhi_u64 (b, 0); } +A unsigned long long f22 (unsigned long long x) { return _bzhi_u64 (x, 0); } +A unsigned long long f23 (void) { return _bzhi_u64 (b, 5); } +A unsigned long long f24 (unsigned long long x) { return _bzhi_u64 (x, 5); } +A unsigned long long f25 (void) { return _bzhi_u64 (b, 63); } +A unsigned long long f26 (unsigned long long x) { return _bzhi_u64 (x, 63); } +A unsigned long long f27 (void) { return _bzhi_u64 (b, 64); } +A unsigned long long f28 (unsigned long long x) { return _bzhi_u64 (x, 64); } +A unsigned long long f29 (void) { return _bzhi_u64 (b, 69); } +A unsigned long long f30 (unsigned long long x) { return _bzhi_u64 (x, 69); } +A unsigned long long f31 (void) { return _bzhi_u64 (b, 257); } +A unsigned long long f32 (unsigned long long x) { return _bzhi_u64 (x, 257); } +A unsigned long long f33 (void) { return _bzhi_u64 (b, 321); } +A unsigned long long f34 (unsigned long long x) { return _bzhi_u64 (x, 321); } +#endif + +static void +bmi2_test () +{ + a = -1U; + b = -1ULL; + if (f1 () != 0 || f2 (-1U) != 0 + || f3 () != 0x1f || f4 (-1U) != 0x1f + || f5 () != 0x7fffffffU || f6 (-1U) != 0x7fffffffU + || f7 () != -1U || f8 (-1U) != -1U + || f9 () != -1U || f10 (-1U) != -1U + || f11 () != 1 || f12 (-1U) != 1 + || f13 () != -1U || f14 (-1U) != -1U) + abort (); +#ifdef __x86_64__ + if (f21 () != 0 || f22 (-1ULL) != 0 + || f23 () != 0x1f || f24 (-1ULL) != 0x1f + || f25 () != 0x7fffffffffffffffULL || f26 (-1ULL) != 0x7fffffffffffffffULL + || f27 () != -1ULL || f28 (-1ULL) != -1ULL + || f29 () != -1ULL || f30 (-1ULL) != -1ULL + || f31 () != 1 || f32 (-1ULL) != 1 + || f33 () != -1ULL || f34 (-1ULL) != -1ULL) + abort (); +#endif +} -- 2.30.2