Got it, will have a try for the suggestion. Pan
From: juzhe.zh...@rivai.ai <juzhe.zh...@rivai.ai> Sent: Thursday, November 30, 2023 3:54 PM To: Li, Pan2 <pan2...@intel.com>; gcc-patches <gcc-patches@gcc.gnu.org> Cc: Wang, Yanzhang <yanzhang.w...@intel.com>; kito.cheng <kito.ch...@gmail.com> Subject: Re: RE: [PATCH v1] RISC-V: Bugfix for legitimize move when get vec mode in zve32f I see. Is it possible to block this situation in other place ? that is, make the situation you said never reach here. Or handle this situation with this following approach: if (get_vector_mode doesn't exist such mode) 1. extract V2SI -> lowpart of the DI reg. 2. extract V2SI -> highpart of the DI reg. 3. scalar move DI reg to DF reg. This patch just bypass the situation (return false) you mention is quite unsafe here. As I remembered, I add this code to handle some ICE due to "return false" here. ________________________________ juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai> From: Li, Pan2<mailto:pan2...@intel.com> Date: 2023-11-30 15:42 To: juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai>; gcc-patches<mailto:gcc-patches@gcc.gnu.org> CC: Wang, Yanzhang<mailto:yanzhang.w...@intel.com>; kito.cheng<mailto:kito.ch...@gmail.com> Subject: RE: RE: [PATCH v1] RISC-V: Bugfix for legitimize move when get vec mode in zve32f Take this file riscv-gnu-toolchain/newlib/newlib/libc/stdlib/mprec.c for example, the arguments and/or related local variables list as below riscv_legitimize_move mode = E_DFmode dest = (reg:DF 144 [ <retval> ]) src = (subreg:DF (reg:V2SI 170) 0) nunits = 1 smode = {m_mode = E_DFmode} Pan From: juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai> <juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai>> Sent: Thursday, November 30, 2023 3:33 PM To: Li, Pan2 <pan2...@intel.com<mailto:pan2...@intel.com>>; gcc-patches <gcc-patches@gcc.gnu.org<mailto:gcc-patches@gcc.gnu.org>> Cc: Wang, Yanzhang <yanzhang.w...@intel.com<mailto:yanzhang.w...@intel.com>>; kito.cheng <kito.ch...@gmail.com<mailto:kito.ch...@gmail.com>> Subject: Re: RE: [PATCH v1] RISC-V: Bugfix for legitimize move when get vec mode in zve32f What it the RTX of the operand ? ________________________________ juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai> From: Li, Pan2<mailto:pan2...@intel.com> Date: 2023-11-30 15:31 To: juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai>; gcc-patches<mailto:gcc-patches@gcc.gnu.org> CC: Wang, Yanzhang<mailto:yanzhang.w...@intel.com>; kito.cheng<mailto:kito.ch...@gmail.com> Subject: RE: [PATCH v1] RISC-V: Bugfix for legitimize move when get vec mode in zve32f > Why does get_vector_mode doesn't exist a vector mode ? Because we set the zve32f here, but try to get_vect_mode with E_V1DFmode. According to the ISA, FP64 is not support when zve32F. Pan From: juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai> <juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai>> Sent: Thursday, November 30, 2023 3:24 PM To: Li, Pan2 <pan2...@intel.com<mailto:pan2...@intel.com>>; gcc-patches <gcc-patches@gcc.gnu.org<mailto:gcc-patches@gcc.gnu.org>> Cc: Li, Pan2 <pan2...@intel.com<mailto:pan2...@intel.com>>; Wang, Yanzhang <yanzhang.w...@intel.com<mailto:yanzhang.w...@intel.com>>; kito.cheng <kito.ch...@gmail.com<mailto:kito.ch...@gmail.com>> Subject: Re: [PATCH v1] RISC-V: Bugfix for legitimize move when get vec mode in zve32f Why does get_vector_mode doesn't exist a vector mode ? It must exist a vector mode, otherwise, it will cause ICE in other situations. ________________________________ juzhe.zh...@rivai.ai<mailto:juzhe.zh...@rivai.ai> From: pan2.li<mailto:pan2...@intel.com> Date: 2023-11-30 15:21 To: gcc-patches<mailto:gcc-patches@gcc.gnu.org> CC: juzhe.zhong<mailto:juzhe.zh...@rivai.ai>; pan2.li<mailto:pan2...@intel.com>; yanzhang.wang<mailto:yanzhang.w...@intel.com>; kito.cheng<mailto:kito.ch...@gmail.com> Subject: [PATCH v1] RISC-V: Bugfix for legitimize move when get vec mode in zve32f From: Pan Li <pan2...@intel.com<mailto:pan2...@intel.com>> When require mode after get_vec_mode in riscv_legitimize_move, there will be precondition that the mode is exists. Or we will have E_VOIDMode and of course have ICE when required. Typically we should first check the mode exists or not before require, or more friendly like leverage exist (U *mode) to get the expected mode if exists and unchanged if not. This patch would like to fix this by exist (U *mode) for requiring a mode after get_vec_mode. PR target/112743 gcc/ChangeLog: * config/riscv/riscv.cc (riscv_legitimize_move): Take the exist (U *mode) instead of directly require (). gcc/testsuite/ChangeLog: * gcc.target/riscv/rvv/base/pr112743-2.c: New test. Signed-off-by: Pan Li <pan2...@intel.com<mailto:pan2...@intel.com>> --- gcc/config/riscv/riscv.cc | 47 ++++++++++------- .../gcc.target/riscv/rvv/base/pr112743-2.c | 52 +++++++++++++++++++ 2 files changed, 79 insertions(+), 20 deletions(-) create mode 100644 gcc/testsuite/gcc.target/riscv/rvv/base/pr112743-2.c diff --git a/gcc/config/riscv/riscv.cc b/gcc/config/riscv/riscv.cc index a4fc858fb50..19413b2c976 100644 --- a/gcc/config/riscv/riscv.cc +++ b/gcc/config/riscv/riscv.cc @@ -2615,32 +2615,39 @@ riscv_legitimize_move (machine_mode mode, rtx dest, rtx src) smode = SImode; nunits = nunits * 2; } - vmode = riscv_vector::get_vector_mode (smode, nunits).require (); - rtx v = gen_lowpart (vmode, SUBREG_REG (src)); - for (unsigned int i = 0; i < num; i++) + opt_machine_mode opt_mode = riscv_vector::get_vector_mode (smode, nunits); + + if (opt_mode.exists (&vmode)) { - rtx result; - if (num == 1) - result = dest; - else if (i == 0) - result = gen_lowpart (smode, dest); - else - result = gen_reg_rtx (smode); - riscv_vector::emit_vec_extract (result, v, index + i); + rtx v = gen_lowpart (vmode, SUBREG_REG (src)); - if (i == 1) + for (unsigned int i = 0; i < num; i++) { - rtx tmp - = expand_binop (Pmode, ashl_optab, gen_lowpart (Pmode, result), - gen_int_mode (32, Pmode), NULL_RTX, 0, - OPTAB_DIRECT); - rtx tmp2 = expand_binop (Pmode, ior_optab, tmp, dest, NULL_RTX, 0, - OPTAB_DIRECT); - emit_move_insn (dest, tmp2); + rtx result; + if (num == 1) + result = dest; + else if (i == 0) + result = gen_lowpart (smode, dest); + else + result = gen_reg_rtx (smode); + + riscv_vector::emit_vec_extract (result, v, index + i); + + if (i == 1) + { + rtx tmp = expand_binop (Pmode, ashl_optab, + gen_lowpart (Pmode, result), + gen_int_mode (32, Pmode), NULL_RTX, 0, + OPTAB_DIRECT); + rtx tmp2 = expand_binop (Pmode, ior_optab, tmp, dest, + NULL_RTX, 0, + OPTAB_DIRECT); + emit_move_insn (dest, tmp2); + } } + return true; } - return true; } /* Expand (set (reg:QI target) (mem:QI (address))) diff --git a/gcc/testsuite/gcc.target/riscv/rvv/base/pr112743-2.c b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112743-2.c new file mode 100644 index 00000000000..fdb35fd70f2 --- /dev/null +++ b/gcc/testsuite/gcc.target/riscv/rvv/base/pr112743-2.c @@ -0,0 +1,52 @@ +/* Test that we do not have ice when compile */ +/* { dg-do compile } */ +/* { dg-options "-march=rv64gc_zve32f_zvfh_zfh -mabi=lp64 -O2" } */ + +#include <sys/types.h> + +union double_union +{ + double d; + __uint32_t i[2]; +}; + +#define word0(x) (x.i[1]) +#define word1(x) (x.i[0]) + +#define P 53 +#define Exp_shift 20 +#define Exp_msk1 ((__uint32_t)0x100000L) +#define Exp_mask ((__uint32_t)0x7ff00000L) + +double ulp (double _x) +{ + union double_union x, a; + register int L; + + x.d = _x; + L = (word0 (x) & Exp_mask) - (P - 1) * Exp_msk1; + + if (L > 0) + { + L |= Exp_msk1 >> 4; + word0 (a) = L; + word1 (a) = 0; + } + else + { + L = -L >> Exp_shift; + if (L < Exp_shift) + { + word0 (a) = 0x80000 >> L; + word1 (a) = 0; + } + else + { + word0 (a) = 0; + L -= Exp_shift; + word1 (a) = L >= 31 ? 1 : 1 << (31 - L); + } + } + + return a.d; +} -- 2.34.1
