From 627227636127c44df0e01ab8fd9fae3f731fa8b0 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Sat, 22 Oct 2022 19:58:21 +1000 Subject: tcg/riscv: Fix range matched by TCG_CT_CONST_M12 We were matching a signed 13-bit range, not a 12-bit range. Expand the commentary within the function and be explicit about all of the ranges. Reported-by: LIU Zhiwei Signed-off-by: Richard Henderson Reviewed-by: LIU Zhiwei Reviewed-by: Alistair Francis Message-Id: <20221022095821.2441874-1-richard.henderson@linaro.org> Signed-off-by: Alistair Francis --- tcg/riscv/tcg-target.c.inc | 19 ++++++++++++++++--- 1 file changed, 16 insertions(+), 3 deletions(-) (limited to 'tcg/riscv/tcg-target.c.inc') diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc index 81a83e4..1911978 100644 --- a/tcg/riscv/tcg-target.c.inc +++ b/tcg/riscv/tcg-target.c.inc @@ -154,13 +154,26 @@ static bool tcg_target_const_match(int64_t val, TCGType type, int ct) if ((ct & TCG_CT_CONST_ZERO) && val == 0) { return 1; } - if ((ct & TCG_CT_CONST_S12) && val == sextreg(val, 0, 12)) { + /* + * Sign extended from 12 bits: [-0x800, 0x7ff]. + * Used for most arithmetic, as this is the isa field. + */ + if ((ct & TCG_CT_CONST_S12) && val >= -0x800 && val <= 0x7ff) { return 1; } - if ((ct & TCG_CT_CONST_N12) && -val == sextreg(-val, 0, 12)) { + /* + * Sign extended from 12 bits, negated: [-0x7ff, 0x800]. + * Used for subtraction, where a constant must be handled by ADDI. + */ + if ((ct & TCG_CT_CONST_N12) && val >= -0x7ff && val <= 0x800) { return 1; } - if ((ct & TCG_CT_CONST_M12) && val >= -0xfff && val <= 0xfff) { + /* + * Sign extended from 12 bits, +/- matching: [-0x7ff, 0x7ff]. + * Used by addsub2, which may need the negative operation, + * and requires the modified constant to be representable. + */ + if ((ct & TCG_CT_CONST_M12) && val >= -0x7ff && val <= 0x7ff) { return 1; } return 0; -- cgit v1.1 From 9b246685b3dbbf21800e3a9a09f8bed384a1fb37 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Fri, 21 Oct 2022 09:38:36 +1000 Subject: tcg/riscv: Fix reg overlap case in tcg_out_addsub2 There was a typo using opc_addi instead of opc_add with the two registers. While we're at it, simplify the gating test to al == bl to improve dynamic scheduling even when the output register does not overlap the inputs. Reported-by: LIU Zhiwei Signed-off-by: Richard Henderson Reviewed-by: Alistair Francis Message-Id: <20221020233836.2341671-1-richard.henderson@linaro.org> Signed-off-by: Alistair Francis --- tcg/riscv/tcg-target.c.inc | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) (limited to 'tcg/riscv/tcg-target.c.inc') diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc index 1911978..2a84c57 100644 --- a/tcg/riscv/tcg-target.c.inc +++ b/tcg/riscv/tcg-target.c.inc @@ -700,9 +700,15 @@ static void tcg_out_addsub2(TCGContext *s, if (cbl) { tcg_out_opc_imm(s, opc_addi, rl, al, bl); tcg_out_opc_imm(s, OPC_SLTIU, TCG_REG_TMP0, rl, bl); - } else if (rl == al && rl == bl) { + } else if (al == bl) { + /* + * If the input regs overlap, this is a simple doubling + * and carry-out is the input msb. This special case is + * required when the output reg overlaps the input, + * but we might as well use it always. + */ tcg_out_opc_imm(s, OPC_SLTI, TCG_REG_TMP0, al, 0); - tcg_out_opc_reg(s, opc_addi, rl, al, bl); + tcg_out_opc_reg(s, opc_add, rl, al, al); } else { tcg_out_opc_reg(s, opc_add, rl, al, bl); tcg_out_opc_reg(s, OPC_SLTU, TCG_REG_TMP0, -- cgit v1.1 From 2e3a933abb8bd965db99375bfc341d2d46480995 Mon Sep 17 00:00:00 2001 From: Richard Henderson Date: Mon, 24 Oct 2022 09:33:37 +1000 Subject: tcg/riscv: Fix base register for user-only qemu_ld/st When guest_base != 0, we were not coordinating the usage of TCG_REG_TMP0 as base properly, leading to a previous zero-extend of the input address being discarded. Shuffle the alignment check to the front, because that does not depend on the zero-extend, and it keeps the register usage clear. Set base after each step of the address arithmetic instead of before. Return the base register used from tcg_out_tlb_load, so as to keep that register choice localized to that function. Reported-by: LIU Zhiwei Signed-off-by: Richard Henderson Reviewed-by: LIU Zhiwei Reviewed-by: Alistair Francis Message-Id: <20221023233337.2846860-1-richard.henderson@linaro.org> Signed-off-by: Alistair Francis --- tcg/riscv/tcg-target.c.inc | 39 ++++++++++++++++++++++----------------- 1 file changed, 22 insertions(+), 17 deletions(-) (limited to 'tcg/riscv/tcg-target.c.inc') diff --git a/tcg/riscv/tcg-target.c.inc b/tcg/riscv/tcg-target.c.inc index 2a84c57..e3b6080 100644 --- a/tcg/riscv/tcg-target.c.inc +++ b/tcg/riscv/tcg-target.c.inc @@ -923,9 +923,9 @@ static void tcg_out_goto(TCGContext *s, const tcg_insn_unit *target) tcg_debug_assert(ok); } -static void tcg_out_tlb_load(TCGContext *s, TCGReg addrl, - TCGReg addrh, MemOpIdx oi, - tcg_insn_unit **label_ptr, bool is_load) +static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addrl, + TCGReg addrh, MemOpIdx oi, + tcg_insn_unit **label_ptr, bool is_load) { MemOp opc = get_memop(oi); unsigned s_bits = opc & MO_SIZE; @@ -975,6 +975,7 @@ static void tcg_out_tlb_load(TCGContext *s, TCGReg addrl, addrl = TCG_REG_TMP0; } tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_REG_TMP2, addrl); + return TCG_REG_TMP0; } static void add_qemu_ldst_label(TCGContext *s, int is_ld, MemOpIdx oi, @@ -1177,7 +1178,7 @@ static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64) #else unsigned a_bits; #endif - TCGReg base = TCG_REG_TMP0; + TCGReg base; data_regl = *args++; data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0); @@ -1187,23 +1188,25 @@ static void tcg_out_qemu_ld(TCGContext *s, const TCGArg *args, bool is_64) opc = get_memop(oi); #if defined(CONFIG_SOFTMMU) - tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 1); + base = tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 1); tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64); add_qemu_ldst_label(s, 1, oi, (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32), data_regl, data_regh, addr_regl, addr_regh, s->code_ptr, label_ptr); #else - if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) { - tcg_out_ext32u(s, base, addr_regl); - addr_regl = base; - } a_bits = get_alignment_bits(opc); if (a_bits) { tcg_out_test_alignment(s, true, addr_regl, a_bits); } + base = addr_regl; + if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) { + tcg_out_ext32u(s, TCG_REG_TMP0, base); + base = TCG_REG_TMP0; + } if (guest_base != 0) { - tcg_out_opc_reg(s, OPC_ADD, base, TCG_GUEST_BASE_REG, addr_regl); + tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_GUEST_BASE_REG, base); + base = TCG_REG_TMP0; } tcg_out_qemu_ld_direct(s, data_regl, data_regh, base, opc, is_64); #endif @@ -1249,7 +1252,7 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64) #else unsigned a_bits; #endif - TCGReg base = TCG_REG_TMP0; + TCGReg base; data_regl = *args++; data_regh = (TCG_TARGET_REG_BITS == 32 && is_64 ? *args++ : 0); @@ -1259,23 +1262,25 @@ static void tcg_out_qemu_st(TCGContext *s, const TCGArg *args, bool is_64) opc = get_memop(oi); #if defined(CONFIG_SOFTMMU) - tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 0); + base = tcg_out_tlb_load(s, addr_regl, addr_regh, oi, label_ptr, 0); tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc); add_qemu_ldst_label(s, 0, oi, (is_64 ? TCG_TYPE_I64 : TCG_TYPE_I32), data_regl, data_regh, addr_regl, addr_regh, s->code_ptr, label_ptr); #else - if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) { - tcg_out_ext32u(s, base, addr_regl); - addr_regl = base; - } a_bits = get_alignment_bits(opc); if (a_bits) { tcg_out_test_alignment(s, false, addr_regl, a_bits); } + base = addr_regl; + if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) { + tcg_out_ext32u(s, TCG_REG_TMP0, base); + base = TCG_REG_TMP0; + } if (guest_base != 0) { - tcg_out_opc_reg(s, OPC_ADD, base, TCG_GUEST_BASE_REG, addr_regl); + tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP0, TCG_GUEST_BASE_REG, base); + base = TCG_REG_TMP0; } tcg_out_qemu_st_direct(s, data_regl, data_regh, base, opc); #endif -- cgit v1.1