| Message ID | 20210921201915.601245-10-git@xen0n.name (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | LoongArch64 port of QEMU TCG | expand |
On 9/21/21 1:18 PM, WANG Xuerui wrote: > + /* Test for PC-relative values that can be loaded faster. */ > + intptr_t pc_offset = val - (uintptr_t)s->code_ptr; This isn't quite right for split r^x code buffer. You should have seen this with --enable-debug-tcg... You need pc_offset = tcg_pcrel_diff(s, (void *)val). > + if (pc_offset == (int32_t)pc_offset) { > + tcg_target_long lo = sextreg(pc_offset, 0, 12); > + tcg_target_long hi = pc_offset - lo; > + tcg_out_opc_pcaddu12i(s, rd, hi >> 12); And... this doesn't quite work, right at the edges. If lo is negative, hi can overflow out of range. There are a number of ways to fix this. One is to extract the pieces and re-assemble to see if it matches. Another is to rearrange the arithmetic just a little and use PCALAU12I. > + tcg_target_long upper = (val >> 12) & 0xfffff; > + tcg_target_long higher = (val >> 32) & 0xfffff; Better to use extract64(val, 12, 20) and extract64(val, 32, 30). r~
Hi Richard, On 9/22/21 12:25, Richard Henderson wrote: > On 9/21/21 1:18 PM, WANG Xuerui wrote: >> + /* Test for PC-relative values that can be loaded faster. */ >> + intptr_t pc_offset = val - (uintptr_t)s->code_ptr; > > This isn't quite right for split r^x code buffer. > You should have seen this with --enable-debug-tcg... > > You need pc_offset = tcg_pcrel_diff(s, (void *)val). Indeed; I just realized TCG debugging isn't fully enabled with --enable-debug only. Will fix in v3. > >> + if (pc_offset == (int32_t)pc_offset) { >> + tcg_target_long lo = sextreg(pc_offset, 0, 12); >> + tcg_target_long hi = pc_offset - lo; >> + tcg_out_opc_pcaddu12i(s, rd, hi >> 12); > > And... this doesn't quite work, right at the edges. If lo is > negative, hi can overflow out of range. There are a number of ways to > fix this. One is to extract the pieces and re-assemble to see if it > matches. Another is to rearrange the arithmetic just a little and use > PCALAU12I. I actually wrote a small test program to test for this, but found no overflow issues here; rather the tcg_out_opc_ori call below has signedness problem (need to mask the low variable, which is signed, with 0xfff to avoid overwriting the opcode field). I think I'll add a tcg_debug_assert here, but keep the logic intact. > >> + tcg_target_long upper = (val >> 12) & 0xfffff; >> + tcg_target_long higher = (val >> 32) & 0xfffff; > > Better to use extract64(val, 12, 20) and extract64(val, 32, 30). Sure; but as the instructions perform sign-extension, thus taking signed operands, sextract64 or the wrapped sextreg will do it. > > > r~
On 9/22/21 8:16 AM, WANG Xuerui wrote: > Hi Richard, > > On 9/22/21 12:25, Richard Henderson wrote: >> On 9/21/21 1:18 PM, WANG Xuerui wrote: >>> + /* Test for PC-relative values that can be loaded faster. */ >>> + intptr_t pc_offset = val - (uintptr_t)s->code_ptr; >> >> This isn't quite right for split r^x code buffer. >> You should have seen this with --enable-debug-tcg... >> >> You need pc_offset = tcg_pcrel_diff(s, (void *)val). > Indeed; I just realized TCG debugging isn't fully enabled with --enable-debug only. Um... it should be.
Hi Richard, On 9/22/21 23:17, Richard Henderson wrote: > On 9/22/21 8:16 AM, WANG Xuerui wrote: >> Hi Richard, >> >> On 9/22/21 12:25, Richard Henderson wrote: >>> On 9/21/21 1:18 PM, WANG Xuerui wrote: >>>> + /* Test for PC-relative values that can be loaded faster. */ >>>> + intptr_t pc_offset = val - (uintptr_t)s->code_ptr; >>> >>> This isn't quite right for split r^x code buffer. >>> You should have seen this with --enable-debug-tcg... >>> >>> You need pc_offset = tcg_pcrel_diff(s, (void *)val). >> Indeed; I just realized TCG debugging isn't fully enabled with >> --enable-debug only. > > Um... it should be. Hmm, maybe I was having the wrong impression, I even grepped for CONFIG_DEBUG_TCG and it showed 1, yet my assertions didn't fire during one of my debugging sessions... Maybe I was just asserting at the wrong place. Never mind though, problems are all solved now.
diff --git a/tcg/loongarch64/tcg-target.c.inc b/tcg/loongarch64/tcg-target.c.inc index 338b772732..e4e7e5e903 100644 --- a/tcg/loongarch64/tcg-target.c.inc +++ b/tcg/loongarch64/tcg-target.c.inc @@ -247,6 +247,93 @@ static void tcg_out_mb(TCGContext *s, TCGArg a0) tcg_out_opc_dbar(s, 0); } +static bool tcg_out_mov(TCGContext *s, TCGType type, TCGReg ret, TCGReg arg) +{ + if (ret == arg) { + return true; + } + switch (type) { + case TCG_TYPE_I32: + case TCG_TYPE_I64: + /* + * Conventional register-register move used in LoongArch is + * `or dst, src, zero`. + */ + tcg_out_opc_or(s, ret, arg, TCG_REG_ZERO); + break; + default: + g_assert_not_reached(); + } + return true; +} + +static void tcg_out_movi(TCGContext *s, TCGType type, TCGReg rd, + tcg_target_long val) +{ + if (type == TCG_TYPE_I32) { + val = (int32_t)val; + } + + /* Single-instruction cases. */ + tcg_target_long low = sextreg(val, 0, 12); + if (low == val) { + /* val fits in simm12: addi.w rd, zero, val */ + tcg_out_opc_addi_w(s, rd, TCG_REG_ZERO, val); + return; + } + if (0x800 <= val && val <= 0xfff) { + /* val fits in uimm12: ori rd, zero, val */ + tcg_out_opc_ori(s, rd, TCG_REG_ZERO, val); + return; + } + + /* Test for PC-relative values that can be loaded faster. */ + intptr_t pc_offset = val - (uintptr_t)s->code_ptr; + if (pc_offset == sextreg(pc_offset, 0, 22) && (pc_offset & 3) == 0) { + tcg_out_opc_pcaddu2i(s, rd, pc_offset >> 2); + return; + } + if (pc_offset == (int32_t)pc_offset) { + tcg_target_long lo = sextreg(pc_offset, 0, 12); + tcg_target_long hi = pc_offset - lo; + tcg_out_opc_pcaddu12i(s, rd, hi >> 12); + tcg_out_opc_addi_d(s, rd, rd, lo); + return; + } + + /* + * Slow path: at most lu12i.w + ori + cu32i.d + cu52i.d. + * + * Chop upper bits into 3 immediate-field-sized segments respectively. + */ + tcg_target_long upper = (val >> 12) & 0xfffff; + tcg_target_long higher = (val >> 32) & 0xfffff; + tcg_target_long top = val >> 52; + + tcg_out_opc_lu12i_w(s, rd, upper); + if (low != 0) { + tcg_out_opc_ori(s, rd, rd, low & 0xfff); + } + + if (sextreg(val, 0, 32) == val) { + /* + * Fits in 32-bits, upper bits are already properly sign-extended by + * lu12i.w. + */ + return; + } + tcg_out_opc_cu32i_d(s, rd, higher); + + if (sextreg(val, 0, 52) == val) { + /* + * Fits in 52-bits, upper bits are already properly sign-extended by + * cu32i.d. + */ + return; + } + tcg_out_opc_cu52i_d(s, rd, rd, top); +} + /* * Entry-points */ @@ -262,6 +349,8 @@ static void tcg_out_op(TCGContext *s, TCGOpcode opc, tcg_out_mb(s, a0); break; + case INDEX_op_mov_i32: /* Always emitted via tcg_out_mov. */ + case INDEX_op_mov_i64: default: g_assert_not_reached(); }
Signed-off-by: WANG Xuerui <git@xen0n.name> --- tcg/loongarch64/tcg-target.c.inc | 89 ++++++++++++++++++++++++++++++++ 1 file changed, 89 insertions(+)