Hi Richard,
On 9/20/21 22:36, Richard Henderson wrote:
On 9/20/21 1:04 AM, WANG Xuerui wrote:
+static bool reloc_call(tcg_insn_unit *src_rw, const tcg_insn_unit
*target)
+{
+ const tcg_insn_unit *src_rx = tcg_splitwx_to_rx(src_rw);
+ intptr_t offset = (intptr_t)target - (intptr_t)src_rx;
+ int32_t lo = sextreg(offset, 0, 12);
+ int32_t hi = offset - lo;
+
+ tcg_debug_assert((offset & 2) == 0);
+ if (offset == hi + lo) {
+ hi >>= 12;
+ src_rw[0] |= (hi << 5) & 0x1ffffe0; /* pcaddu12i's Sj20 imm */
+ lo >>= 2;
+ src_rw[1] |= (lo << 10) & 0x3fffc00; /* jirl's Sk16 imm */
+ return true;
+ }
+
+ return false;
+}
This doesn't seem to belong as a "reloc".
Certainly it doesn't seem like something that can simply be allowed to
fail.
Yes, you're right on this; on closer look at the riscv port they
actually reused this logic once (the riscv port drops large constants to
pool, hence need some PC-relative hackery). For LoongArch the only usage
of this code is for generating calls, so I'll just merge this into the
commit doing tcg_out_call, and inline if the resulting code is still
readable.
And it's 1 a.m. here in China, so I'll be processing the other review
comments after getting some sleep. (Today's in the middle of the 3-day
Mid-Autumn Festival holiday here, and that's why I can work on this
hobby project like it's $DAY_JOB!) I'll send the v2 hopefully at
afternoon local time (tomorrow in your timezone).
r~