From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from eggs.gnu.org ([2001:4830:134:3::10]:59578) by lists.gnu.org with esmtp (Exim 4.71) (envelope-from ) id 1ddVPa-0003uS-Hj for qemu-devel@nongnu.org; Fri, 04 Aug 2017 01:44:41 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.71) (envelope-from ) id 1ddVPY-0003Eo-Cg for qemu-devel@nongnu.org; Fri, 04 Aug 2017 01:44:38 -0400 Received: from mail-pg0-x244.google.com ([2607:f8b0:400e:c05::244]:34284) by eggs.gnu.org with esmtps (TLS1.0:RSA_AES_128_CBC_SHA1:16) (Exim 4.71) (envelope-from ) id 1ddVPY-0003Cc-0R for qemu-devel@nongnu.org; Fri, 04 Aug 2017 01:44:36 -0400 Received: by mail-pg0-x244.google.com with SMTP id y192so836589pgd.1 for ; Thu, 03 Aug 2017 22:44:35 -0700 (PDT) Received: from bigtime.twiddle.net (97-126-108-236.tukw.qwest.net. [97.126.108.236]) by smtp.gmail.com with ESMTPSA id o14sm1061063pfi.158.2017.08.03.22.44.33 for (version=TLS1_2 cipher=ECDHE-RSA-CHACHA20-POLY1305 bits=256/256); Thu, 03 Aug 2017 22:44:33 -0700 (PDT) Sender: Richard Henderson From: Richard Henderson Date: Thu, 3 Aug 2017 22:44:10 -0700 Message-Id: <20170804054426.10590-8-rth@twiddle.net> In-Reply-To: <20170804054426.10590-1-rth@twiddle.net> References: <20170804054426.10590-1-rth@twiddle.net> Subject: [Qemu-devel] [PATCH for-2.11 07/23] tcg/s390: Use constant pool for movi List-Id: List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: qemu-devel@nongnu.org Split out maybe_out_small_movi for use with other operations that want to add to the constant pool. Signed-off-by: Richard Henderson --- include/elf.h | 3 +- tcg/s390/tcg-target.h | 1 + tcg/s390/tcg-target.inc.c | 130 +++++++++++++++++++++++++++------------------- 3 files changed, 80 insertions(+), 54 deletions(-) diff --git a/include/elf.h b/include/elf.h index cd51434877..e8a515ce3d 100644 --- a/include/elf.h +++ b/include/elf.h @@ -942,8 +942,9 @@ typedef struct { #define R_390_TLS_DTPOFF 55 /* Offset in TLS block. */ #define R_390_TLS_TPOFF 56 /* Negate offset in static TLS block. */ +#define R_390_20 57 /* Keep this the last entry. */ -#define R_390_NUM 57 +#define R_390_NUM 58 /* x86-64 relocation types */ #define R_X86_64_NONE 0 /* No reloc */ diff --git a/tcg/s390/tcg-target.h b/tcg/s390/tcg-target.h index 8c5a30ccf8..18a0efac3e 100644 --- a/tcg/s390/tcg-target.h +++ b/tcg/s390/tcg-target.h @@ -156,5 +156,6 @@ static inline void tb_target_set_jmp_target(uintptr_t tc_ptr, #ifdef CONFIG_SOFTMMU #define TCG_TARGET_NEED_LDST_LABELS #endif +#define TCG_TARGET_NEED_POOL_LABELS #endif diff --git a/tcg/s390/tcg-target.inc.c b/tcg/s390/tcg-target.inc.c index 59c0da0922..29b77ff67f 100644 --- a/tcg/s390/tcg-target.inc.c +++ b/tcg/s390/tcg-target.inc.c @@ -29,6 +29,7 @@ #error "unsupported code generation mode" #endif +#include "tcg-pool.inc.c" #include "elf.h" /* ??? The translation blocks produced by TCG are generally small enough to @@ -361,6 +362,7 @@ static void patch_reloc(tcg_insn_unit *code_ptr, int type, intptr_t value, intptr_t addend) { intptr_t pcrel2; + uint32_t old; value += addend; pcrel2 = (tcg_insn_unit *)value - code_ptr; @@ -374,6 +376,12 @@ static void patch_reloc(tcg_insn_unit *code_ptr, int type, assert(pcrel2 == (int32_t)pcrel2); tcg_patch32(code_ptr, pcrel2); break; + case R_390_20: + assert(value == sextract64(value, 0, 20)); + old = *(uint32_t *)code_ptr & 0xf00000ff; + old |= ((value & 0xfff) << 16) | ((value & 0xff000) >> 4); + tcg_patch32(code_ptr, old); + break; default: g_assert_not_reached(); } @@ -562,14 +570,16 @@ static void tcg_out_mov(TCGContext *s, TCGType type, TCGReg dst, TCGReg src) } } -/* load a register with an immediate value */ -static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, - tcg_target_long sval, bool in_prologue) -{ - static const S390Opcode lli_insns[4] = { - RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH - }; +static const S390Opcode lli_insns[4] = { + RI_LLILL, RI_LLILH, RI_LLIHL, RI_LLIHH +}; +static const S390Opcode ii_insns[4] = { + RI_IILL, RI_IILH, RI_IIHL, RI_IIHH +}; +static bool maybe_out_small_movi(TCGContext *s, TCGType type, + TCGReg ret, tcg_target_long sval) +{ tcg_target_ulong uval = sval; int i; @@ -581,17 +591,37 @@ static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, /* Try all 32-bit insns that can load it in one go. */ if (sval >= -0x8000 && sval < 0x8000) { tcg_out_insn(s, RI, LGHI, ret, sval); - return; + return true; } for (i = 0; i < 4; i++) { tcg_target_long mask = 0xffffull << i*16; if ((uval & mask) == uval) { tcg_out_insn_RI(s, lli_insns[i], ret, uval >> i*16); - return; + return true; } } + return false; +} + +/* load a register with an immediate value */ +static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, + tcg_target_long sval, bool in_prologue) +{ + tcg_target_ulong uval; + + /* Try all 32-bit insns that can load it in one go. */ + if (maybe_out_small_movi(s, type, ret, sval)) { + return; + } + + uval = sval; + if (type == TCG_TYPE_I32) { + uval = (uint32_t)sval; + sval = (int32_t)sval; + } + /* Try all 48-bit insns that can load it in one go. */ if (s390_facilities & FACILITY_EXT_IMM) { if (sval == (int32_t)sval) { @@ -603,7 +633,7 @@ static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, return; } if ((uval & 0xffffffff) == 0) { - tcg_out_insn(s, RIL, LLIHF, ret, uval >> 31 >> 1); + tcg_out_insn(s, RIL, LLIHF, ret, uval >> 32); return; } } @@ -626,55 +656,44 @@ static void tcg_out_movi_int(TCGContext *s, TCGType type, TCGReg ret, } } - /* If extended immediates are not present, then we may have to issue - several instructions to load the low 32 bits. */ - if (!(s390_facilities & FACILITY_EXT_IMM)) { - /* A 32-bit unsigned value can be loaded in 2 insns. And given - that the lli_insns loop above did not succeed, we know that - both insns are required. */ - if (uval <= 0xffffffff) { - tcg_out_insn(s, RI, LLILL, ret, uval); - tcg_out_insn(s, RI, IILH, ret, uval >> 16); - return; - } + /* A 32-bit unsigned value can be loaded in 2 insns. And given + that LLILL, LLIHL, LLILF above did not succeed, we know that + both insns are required. */ + if (uval <= 0xffffffff) { + tcg_out_insn(s, RI, LLILL, ret, uval); + tcg_out_insn(s, RI, IILH, ret, uval >> 16); + return; + } - /* If all high bits are set, the value can be loaded in 2 or 3 insns. - We first want to make sure that all the high bits get set. With - luck the low 16-bits can be considered negative to perform that for - free, otherwise we load an explicit -1. */ - if (sval >> 31 >> 1 == -1) { - if (uval & 0x8000) { - tcg_out_insn(s, RI, LGHI, ret, uval); - } else { - tcg_out_insn(s, RI, LGHI, ret, -1); - tcg_out_insn(s, RI, IILL, ret, uval); - } - tcg_out_insn(s, RI, IILH, ret, uval >> 16); - return; + /* When allowed, stuff it in the constant pool. */ + if (!in_prologue) { + if (USE_REG_TB) { + tcg_out_insn(s, RXY, LG, ret, TCG_REG_TB, TCG_REG_NONE, 0); + new_pool_label(s, sval, R_390_20, s->code_ptr - 2, + -(intptr_t)s->code_gen_ptr); + } else { + tcg_out_insn(s, RIL, LGRL, ret, 0); + new_pool_label(s, sval, R_390_PC32DBL, s->code_ptr - 2, 2); } + return; } - /* If we get here, both the high and low parts have non-zero bits. */ - - /* Recurse to load the lower 32-bits. */ - tcg_out_movi(s, TCG_TYPE_I64, ret, uval & 0xffffffff); - - /* Insert data into the high 32-bits. */ - uval = uval >> 31 >> 1; + /* What's left is for the prologue, loading GUEST_BASE, and because + it failed to match above, is known to be a full 64-bit quantity. + We could try more than this, but it probably wouldn't pay off. */ if (s390_facilities & FACILITY_EXT_IMM) { - if (uval < 0x10000) { - tcg_out_insn(s, RI, IIHL, ret, uval); - } else if ((uval & 0xffff) == 0) { - tcg_out_insn(s, RI, IIHH, ret, uval >> 16); - } else { - tcg_out_insn(s, RIL, IIHF, ret, uval); - } + tcg_out_insn(s, RIL, LLILF, ret, uval); + tcg_out_insn(s, RIL, IIHF, ret, uval >> 32); } else { - if (uval & 0xffff) { - tcg_out_insn(s, RI, IIHL, ret, uval); - } - if (uval & 0xffff0000) { - tcg_out_insn(s, RI, IIHH, ret, uval >> 16); + const S390Opcode *insns = lli_insns; + int i; + + for (i = 0; i < 4; i++) { + uint16_t part = uval >> (16 * i); + if (part) { + tcg_out_insn_RI(s, insns[i], ret, part); + insns = ii_insns; + } } } } @@ -2573,6 +2592,11 @@ static void tcg_target_qemu_prologue(TCGContext *s) tcg_out_insn(s, RR, BCR, S390_CC_ALWAYS, TCG_REG_R14); } +static void tcg_out_nop_fill(tcg_insn_unit *p, int count) +{ + memset(p, 0x07, count * sizeof(tcg_insn_unit)); +} + typedef struct { DebugFrameHeader h; uint8_t fde_def_cfa[4]; -- 2.13.3