[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
[Qemu-devel] [PATCH 57/57] target-i386: Implement tzcnt and fix lzcnt
From: |
Richard Henderson |
Subject: |
[Qemu-devel] [PATCH 57/57] target-i386: Implement tzcnt and fix lzcnt |
Date: |
Wed, 23 Jan 2013 20:03:41 -0800 |
We weren't computing flags for lzcnt at all.
Signed-off-by: Richard Henderson <address@hidden>
---
target-i386/helper.h | 5 ++---
target-i386/int_helper.c | 11 +++--------
target-i386/translate.c | 50 ++++++++++++++++++++++++++++++++----------------
3 files changed, 38 insertions(+), 28 deletions(-)
diff --git a/target-i386/helper.h b/target-i386/helper.h
index ddf6936..2b3e2ae 100644
--- a/target-i386/helper.h
+++ b/target-i386/helper.h
@@ -194,9 +194,8 @@ DEF_HELPER_3(frstor, void, env, tl, int)
DEF_HELPER_3(fxsave, void, env, tl, int)
DEF_HELPER_3(fxrstor, void, env, tl, int)
-DEF_HELPER_FLAGS_1(bsf, TCG_CALL_NO_RWG_SE, tl, tl)
-DEF_HELPER_FLAGS_1(bsr, TCG_CALL_NO_RWG_SE, tl, tl)
-DEF_HELPER_FLAGS_2(lzcnt, TCG_CALL_NO_RWG_SE, tl, tl, int)
+DEF_HELPER_FLAGS_1(clz, TCG_CALL_NO_RWG_SE, tl, tl)
+DEF_HELPER_FLAGS_1(ctz, TCG_CALL_NO_RWG_SE, tl, tl)
DEF_HELPER_FLAGS_2(pdep, TCG_CALL_NO_RWG_SE, tl, tl, tl)
DEF_HELPER_FLAGS_2(pext, TCG_CALL_NO_RWG_SE, tl, tl, tl)
diff --git a/target-i386/int_helper.c b/target-i386/int_helper.c
index 7bec4eb..3b56075 100644
--- a/target-i386/int_helper.c
+++ b/target-i386/int_helper.c
@@ -456,19 +456,14 @@ void helper_idivq_EAX(CPUX86State *env, target_ulong t0)
#endif
/* bit operations */
-target_ulong helper_bsf(target_ulong t0)
+target_ulong helper_ctz(target_ulong t0)
{
return ctztl(t0);
}
-target_ulong helper_lzcnt(target_ulong t0, int wordsize)
+target_ulong helper_clz(target_ulong t0)
{
- return clztl(t0) - (TARGET_LONG_BITS - wordsize);
-}
-
-target_ulong helper_bsr(target_ulong t0)
-{
- return clztl(t0) ^ (TARGET_LONG_BITS - 1);
+ return clztl(t0);
}
target_ulong helper_pdep(target_ulong src, target_ulong mask)
diff --git a/target-i386/translate.c b/target-i386/translate.c
index a9bc21d..c049e42 100644
--- a/target-i386/translate.c
+++ b/target-i386/translate.c
@@ -7151,33 +7151,49 @@ static target_ulong disas_insn(CPUX86State *env,
DisasContext *s,
tcg_gen_movi_tl(cpu_cc_dst, 0);
}
break;
- case 0x1bc: /* bsf */
- case 0x1bd: /* bsr */
+ case 0x1bc: /* bsf / tzcnt */
+ case 0x1bd: /* bsr / lzcnt */
ot = dflag + OT_WORD;
modrm = cpu_ldub_code(env, s->pc++);
reg = ((modrm >> 3) & 7) | rex_r;
gen_ldst_modrm(env, s, modrm, ot, OR_TMP0, 0);
gen_extu(ot, cpu_T[0]);
- if ((b & 1) && (prefixes & PREFIX_REPZ) &&
- (s->cpuid_ext3_features & CPUID_EXT3_ABM)) {
- switch (ot) {
- case OT_WORD:
- gen_helper_lzcnt(cpu_T[0], cpu_T[0], tcg_const_i32(16));
- break;
- case OT_LONG:
- gen_helper_lzcnt(cpu_T[0], cpu_T[0], tcg_const_i32(32));
- break;
- case OT_QUAD:
- gen_helper_lzcnt(cpu_T[0], cpu_T[0], tcg_const_i32(64));
- break;
- }
+
+ /* Note that lzcnt and tzcnt are in different extensions. */
+ if ((prefixes & PREFIX_REPZ)
+ && (b & 1
+ ? s->cpuid_ext3_features & CPUID_EXT3_ABM
+ : s->cpuid_7_0_ebx_features & CPUID_7_0_EBX_BMI1)) {
+ int size = 8 << ot;
+ tcg_gen_mov_tl(cpu_cc_src, cpu_T[0]);
+ if (b & 1) {
+ /* For lzcnt, reduce the target_ulong result by the
+ number of zeros that we expect to find at the top. */
+ gen_helper_clz(cpu_T[0], cpu_T[0]);
+ tcg_gen_subi_tl(cpu_T[0], cpu_T[0], TARGET_LONG_BITS - size);
+ } else {
+ /* For tzcnt, a zero input must return the operand size:
+ force all bits outside the operand size to 1. */
+ target_ulong mask = (target_ulong)-2 << (size - 1);
+ tcg_gen_ori_tl(cpu_T[0], cpu_T[0], mask);
+ gen_helper_ctz(cpu_T[0], cpu_T[0]);
+ }
+ /* For lzcnt/tzcnt, C and Z bits are defined and are
+ related to the result. */
+ gen_op_update1_cc();
+ set_cc_op(s, CC_OP_BMILGB + ot);
} else {
+ /* For bsr/bsf, only the Z bit is defined and its related
+ to the input and not the result. */
tcg_gen_mov_tl(cpu_cc_dst, cpu_T[0]);
set_cc_op(s, CC_OP_LOGICB + ot);
if (b & 1) {
- gen_helper_bsr(cpu_T[0], cpu_T[0]);
+ /* For bsr, return the bit index of the first 1 bit,
+ not the count of leading zeros. */
+ gen_helper_clz(cpu_T[0], cpu_T[0]);
+ tcg_gen_xori_tl(cpu_T[0], cpu_T[0], TARGET_LONG_BITS - 1);
} else {
- gen_helper_bsf(cpu_T[0], cpu_T[0]);
+ gen_helper_ctz(cpu_T[0], cpu_T[0]);
}
}
gen_op_mov_reg_T0(ot, reg);
--
1.7.11.7
- [Qemu-devel] [PATCH 23/57] target-i386: convert gen_compute_eflags_c to TCG, (continued)
- [Qemu-devel] [PATCH 23/57] target-i386: convert gen_compute_eflags_c to TCG, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 28/57] target-i386: introduce CCPrepare, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 30/57] target-i386: use CCPrepare to generate conditional jumps, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 44/57] target-i386: Decode the VEX prefixes, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 43/57] target-i386: Tidy prefix parsing, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 42/57] target-i386: Make helper_cc_compute_all const, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 53/57] target-i386: Implement RORX, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 55/57] target-i386: Use clz/ctz for bsf/bsr helpers, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 54/57] target-i386: Implement ADX extension, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 56/57] target-i386: Simplify bsf/bsr flags computation, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 57/57] target-i386: Implement tzcnt and fix lzcnt,
Richard Henderson <=
- [Qemu-devel] [PATCH 33/57] target-i386: introduce gen_cmovcc1, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 31/57] target-i386: inline gen_prepare_cc_slow, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 40/57] target-i386: Use CC_SRC2 for ADC and SBB, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 35/57] target-i386: kill cpu_T3, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 49/57] target-i386: Implement BZHI, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 39/57] target-i386: optimize flags checking after sub using CC_SRC2, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 36/57] target-i386: use gen_op for cmps/scas, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 48/57] target-i386: Implement BLSR, BLSMSK, BLSI, Richard Henderson, 2013/01/23
- [Qemu-devel] [PATCH 46/57] target-i386: Implement ANDN, Richard Henderson, 2013/01/23