mirrors/qemu
0
0
Fork 0
mirror of https://github.com/qemu/qemu.git synced 2024-12-05 09:43:44 +08:00
Code Issues Packages Projects Releases Wiki Activity

Backend vector enhancements

Browse Source 
Dynamic tlb resizing
 -----BEGIN PGP SIGNATURE-----
 
 iQEcBAABAgAGBQJcTyZfAAoJEGTfOOivfiFf/XIH/2uG8YTamq97ZMALuzSMUD1O
 RApi8FRghk4M1SMrZv9KAnR3IcYl8Y8Qjlj7LDytD4axVG+1PdPsOwIiVThd3a0d
 yYB510vCr1nBi0d7an70Ks2n5v0pCm/Q5/WK00F03Swg/eeUGVjiyVhXUQDAdJ8M
 wI8Qi2eIF7Y2Pin+kXvEvHQwkrCYoRV5V8c+gW7DfuPM9rfjZ2ieAWisUeRkWJuT
 QVwVEjbAts1RH1JLe7M4DZYaaHoHjjhssG4WUWVt5CVtZBnb10raoRZYR69bNT+w
 f3LTvpY2Ga0K+rQJa90hWig5dbpgUQ2nOBCU0B6/Ee/SRxo74HQEIzhKM8TjMYw=
 =NT5a
 -----END PGP SIGNATURE-----

Merge remote-tracking branch 'remotes/rth/tags/pull-tcg-20190128' into staging

Backend vector enhancements
Dynamic tlb resizing

# gpg: Signature made Mon 28 Jan 2019 15:57:19 GMT
# gpg:                using RSA key 64DF38E8AF7E215F
# gpg: Good signature from "Richard Henderson <richard.henderson@linaro.org>" [full]
# Primary key fingerprint: 7A48 1E78 868B 4DB6 A85A  05C0 64DF 38E8 AF7E 215F

* remotes/rth/tags/pull-tcg-20190128: (23 commits)
  cputlb: Remove static tlb sizing
  tcg/tci: enable dynamic TLB sizing
  tcg/mips: enable dynamic TLB sizing
  tcg/mips: Fix tcg_out_qemu_ld_slow_path
  tcg/arm: enable dynamic TLB sizing
  tcg/riscv: enable dynamic TLB sizing
  tcg/s390: enable dynamic TLB sizing
  tcg/sparc: enable dynamic TLB sizing
  tcg/ppc: enable dynamic TLB sizing
  tcg/aarch64: enable dynamic TLB sizing
  tcg/i386: enable dynamic TLB sizing
  tcg: introduce dynamic TLB sizing
  cputlb: do not evict empty entries to the vtlb
  tcg/aarch64: Implement vector minmax arithmetic
  tcg/aarch64: Implement vector saturating arithmetic
  tcg/i386: Implement vector minmax arithmetic
  tcg/i386: Implement vector saturating arithmetic
  tcg/i386: Split subroutines out of tcg_expand_vec_op
  tcg: Add opcodes for vector minmax arithmetic
  tcg: Add opcodes for vector saturated arithmetic
  ...

Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
This commit is contained in:
Peter Maydell 2019-01-28 16:26:47 +00:00
commit 3a183e330d
23 changed files with 1714 additions and 630 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

192
accel/tcg/cputlb.c
View File

@ -74,6 +74,166 @@ QEMU_BUILD_BUG_ON(sizeof(target_ulong) > sizeof(run_on_cpu_data));
QEMU_BUILD_BUG_ON(NB_MMU_MODES > 16);
#define ALL_MMUIDX_BITS ((1 << NB_MMU_MODES) - 1)
static inline size_t sizeof_tlb(CPUArchState *env, uintptr_t mmu_idx)
{
return env->tlb_mask[mmu_idx] + (1 << CPU_TLB_ENTRY_BITS);
}
static void tlb_window_reset(CPUTLBWindow *window, int64_t ns,
size_t max_entries)
{
window->begin_ns = ns;
window->max_entries = max_entries;
}
static void tlb_dyn_init(CPUArchState *env)
{
int i;
for (i = 0; i < NB_MMU_MODES; i++) {
CPUTLBDesc *desc = &env->tlb_d[i];
size_t n_entries = 1 << CPU_TLB_DYN_DEFAULT_BITS;
tlb_window_reset(&desc->window, get_clock_realtime(), 0);
desc->n_used_entries = 0;
env->tlb_mask[i] = (n_entries - 1) << CPU_TLB_ENTRY_BITS;
env->tlb_table[i] = g_new(CPUTLBEntry, n_entries);
env->iotlb[i] = g_new(CPUIOTLBEntry, n_entries);
}
}
/**
* tlb_mmu_resize_locked() - perform TLB resize bookkeeping; resize if necessary
* @env: CPU that owns the TLB
* @mmu_idx: MMU index of the TLB
*
* Called with tlb_lock_held.
*
* We have two main constraints when resizing a TLB: (1) we only resize it
* on a TLB flush (otherwise we'd have to take a perf hit by either rehashing
* the array or unnecessarily flushing it), which means we do not control how
* frequently the resizing can occur; (2) we don't have access to the guest's
* future scheduling decisions, and therefore have to decide the magnitude of
* the resize based on past observations.
*
* In general, a memory-hungry process can benefit greatly from an appropriately
* sized TLB, since a guest TLB miss is very expensive. This doesn't mean that
* we just have to make the TLB as large as possible; while an oversized TLB
* results in minimal TLB miss rates, it also takes longer to be flushed
* (flushes can be _very_ frequent), and the reduced locality can also hurt
* performance.
*
* To achieve near-optimal performance for all kinds of workloads, we:
*
* 1. Aggressively increase the size of the TLB when the use rate of the
* TLB being flushed is high, since it is likely that in the near future this
* memory-hungry process will execute again, and its memory hungriness will
* probably be similar.
*
* 2. Slowly reduce the size of the TLB as the use rate declines over a
* reasonably large time window. The rationale is that if in such a time window
* we have not observed a high TLB use rate, it is likely that we won't observe
* it in the near future. In that case, once a time window expires we downsize
* the TLB to match the maximum use rate observed in the window.
*
* 3. Try to keep the maximum use rate in a time window in the 30-70% range,
* since in that range performance is likely near-optimal. Recall that the TLB
* is direct mapped, so we want the use rate to be low (or at least not too
* high), since otherwise we are likely to have a significant amount of
* conflict misses.
*/
static void tlb_mmu_resize_locked(CPUArchState *env, int mmu_idx)
{
CPUTLBDesc *desc = &env->tlb_d[mmu_idx];
size_t old_size = tlb_n_entries(env, mmu_idx);
size_t rate;
size_t new_size = old_size;
int64_t now = get_clock_realtime();
int64_t window_len_ms = 100;
int64_t window_len_ns = window_len_ms * 1000 * 1000;
bool window_expired = now > desc->window.begin_ns + window_len_ns;
if (desc->n_used_entries > desc->window.max_entries) {
desc->window.max_entries = desc->n_used_entries;
}
rate = desc->window.max_entries * 100 / old_size;
if (rate > 70) {
new_size = MIN(old_size << 1, 1 << CPU_TLB_DYN_MAX_BITS);
} else if (rate < 30 && window_expired) {
size_t ceil = pow2ceil(desc->window.max_entries);
size_t expected_rate = desc->window.max_entries * 100 / ceil;
/*
* Avoid undersizing when the max number of entries seen is just below
* a pow2. For instance, if max_entries == 1025, the expected use rate
* would be 1025/2048==50%. However, if max_entries == 1023, we'd get
* 1023/1024==99.9% use rate, so we'd likely end up doubling the size
* later. Thus, make sure that the expected use rate remains below 70%.
* (and since we double the size, that means the lowest rate we'd
* expect to get is 35%, which is still in the 30-70% range where
* we consider that the size is appropriate.)
*/
if (expected_rate > 70) {
ceil *= 2;
}
new_size = MAX(ceil, 1 << CPU_TLB_DYN_MIN_BITS);
}
if (new_size == old_size) {
if (window_expired) {
tlb_window_reset(&desc->window, now, desc->n_used_entries);
}
return;
}
g_free(env->tlb_table[mmu_idx]);
g_free(env->iotlb[mmu_idx]);
tlb_window_reset(&desc->window, now, 0);
/* desc->n_used_entries is cleared by the caller */
env->tlb_mask[mmu_idx] = (new_size - 1) << CPU_TLB_ENTRY_BITS;
env->tlb_table[mmu_idx] = g_try_new(CPUTLBEntry, new_size);
env->iotlb[mmu_idx] = g_try_new(CPUIOTLBEntry, new_size);
/*
* If the allocations fail, try smaller sizes. We just freed some
* memory, so going back to half of new_size has a good chance of working.
* Increased memory pressure elsewhere in the system might cause the
* allocations to fail though, so we progressively reduce the allocation
* size, aborting if we cannot even allocate the smallest TLB we support.
*/
while (env->tlb_table[mmu_idx] == NULL || env->iotlb[mmu_idx] == NULL) {
if (new_size == (1 << CPU_TLB_DYN_MIN_BITS)) {
error_report("%s: %s", __func__, strerror(errno));
abort();
}
new_size = MAX(new_size >> 1, 1 << CPU_TLB_DYN_MIN_BITS);
env->tlb_mask[mmu_idx] = (new_size - 1) << CPU_TLB_ENTRY_BITS;
g_free(env->tlb_table[mmu_idx]);
g_free(env->iotlb[mmu_idx]);
env->tlb_table[mmu_idx] = g_try_new(CPUTLBEntry, new_size);
env->iotlb[mmu_idx] = g_try_new(CPUIOTLBEntry, new_size);
}
}
static inline void tlb_table_flush_by_mmuidx(CPUArchState *env, int mmu_idx)
{
tlb_mmu_resize_locked(env, mmu_idx);
memset(env->tlb_table[mmu_idx], -1, sizeof_tlb(env, mmu_idx));
env->tlb_d[mmu_idx].n_used_entries = 0;
}
static inline void tlb_n_used_entries_inc(CPUArchState *env, uintptr_t mmu_idx)
{
env->tlb_d[mmu_idx].n_used_entries++;
}
static inline void tlb_n_used_entries_dec(CPUArchState *env, uintptr_t mmu_idx)
{
env->tlb_d[mmu_idx].n_used_entries--;
}
void tlb_init(CPUState *cpu)
{
CPUArchState *env = cpu->env_ptr;
@ -82,6 +242,8 @@ void tlb_init(CPUState *cpu)
/* Ensure that cpu_reset performs a full flush. */
env->tlb_c.dirty = ALL_MMUIDX_BITS;
tlb_dyn_init(env);
}
/* flush_all_helper: run fn across all cpus
@ -122,7 +284,7 @@ void tlb_flush_counts(size_t *pfull, size_t *ppart, size_t *pelide)
static void tlb_flush_one_mmuidx_locked(CPUArchState *env, int mmu_idx)
{
memset(env->tlb_table[mmu_idx], -1, sizeof(env->tlb_table[0]));
tlb_table_flush_by_mmuidx(env, mmu_idx);
memset(env->tlb_v_table[mmu_idx], -1, sizeof(env->tlb_v_table[0]));
env->tlb_d[mmu_idx].large_page_addr = -1;
env->tlb_d[mmu_idx].large_page_mask = -1;
@ -224,13 +386,24 @@ static inline bool tlb_hit_page_anyprot(CPUTLBEntry *tlb_entry,
tlb_hit_page(tlb_entry->addr_code, page);
}
/**
* tlb_entry_is_empty - return true if the entry is not in use
* @te: pointer to CPUTLBEntry
*/
static inline bool tlb_entry_is_empty(const CPUTLBEntry *te)
{
return te->addr_read == -1 && te->addr_write == -1 && te->addr_code == -1;
}
/* Called with tlb_c.lock held */
static inline void tlb_flush_entry_locked(CPUTLBEntry *tlb_entry,
static inline bool tlb_flush_entry_locked(CPUTLBEntry *tlb_entry,
target_ulong page)
{
if (tlb_hit_page_anyprot(tlb_entry, page)) {
memset(tlb_entry, -1, sizeof(*tlb_entry));
return true;
}
return false;
}
/* Called with tlb_c.lock held */
@ -241,7 +414,9 @@ static inline void tlb_flush_vtlb_page_locked(CPUArchState *env, int mmu_idx,
assert_cpu_is_self(ENV_GET_CPU(env));
for (k = 0; k < CPU_VTLB_SIZE; k++) {
tlb_flush_entry_locked(&env->tlb_v_table[mmu_idx][k], page);
if (tlb_flush_entry_locked(&env->tlb_v_table[mmu_idx][k], page)) {
tlb_n_used_entries_dec(env, mmu_idx);
}
}
}
@ -258,7 +433,9 @@ static void tlb_flush_page_locked(CPUArchState *env, int midx,
midx, lp_addr, lp_mask);
tlb_flush_one_mmuidx_locked(env, midx);
} else {
tlb_flush_entry_locked(tlb_entry(env, midx, page), page);
if (tlb_flush_entry_locked(tlb_entry(env, midx, page), page)) {
tlb_n_used_entries_dec(env, midx);
}
tlb_flush_vtlb_page_locked(env, midx, page);
}
}
@ -435,8 +612,9 @@ void tlb_reset_dirty(CPUState *cpu, ram_addr_t start1, ram_addr_t length)
qemu_spin_lock(&env->tlb_c.lock);
for (mmu_idx = 0; mmu_idx < NB_MMU_MODES; mmu_idx++) {
unsigned int i;
unsigned int n = tlb_n_entries(env, mmu_idx);
for (i = 0; i < CPU_TLB_SIZE; i++) {
for (i = 0; i < n; i++) {
tlb_reset_dirty_range_locked(&env->tlb_table[mmu_idx][i], start1,
length);
}
@ -591,13 +769,14 @@ void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
* Only evict the old entry to the victim tlb if it's for a
* different page; otherwise just overwrite the stale data.
*/
if (!tlb_hit_page_anyprot(te, vaddr_page)) {
if (!tlb_hit_page_anyprot(te, vaddr_page) && !tlb_entry_is_empty(te)) {
unsigned vidx = env->tlb_d[mmu_idx].vindex++ % CPU_VTLB_SIZE;
CPUTLBEntry *tv = &env->tlb_v_table[mmu_idx][vidx];
/* Evict the old entry into the victim tlb. */
copy_tlb_helper_locked(tv, te);
env->iotlb_v[mmu_idx][vidx] = env->iotlb[mmu_idx][index];
tlb_n_used_entries_dec(env, mmu_idx);
}
/* refill the tlb */
@ -649,6 +828,7 @@ void tlb_set_page_with_attrs(CPUState *cpu, target_ulong vaddr,
}
copy_tlb_helper_locked(te, &tn);
tlb_n_used_entries_inc(env, mmu_idx);
qemu_spin_unlock(&env->tlb_c.lock);
}

257
accel/tcg/tcg-runtime-gvec.c
View File

@ -512,6 +512,39 @@ void HELPER(gvec_orc)(void *d, void *a, void *b, uint32_t desc)
clear_high(d, oprsz, desc);
}
void HELPER(gvec_nand)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(vec64)) {
*(vec64 *)(d + i) = ~(*(vec64 *)(a + i) & *(vec64 *)(b + i));
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_nor)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(vec64)) {
*(vec64 *)(d + i) = ~(*(vec64 *)(a + i) | *(vec64 *)(b + i));
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_eqv)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(vec64)) {
*(vec64 *)(d + i) = ~(*(vec64 *)(a + i) ^ *(vec64 *)(b + i));
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_ands)(void *d, void *a, uint64_t b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
@ -995,3 +1028,227 @@ void HELPER(gvec_ussub64)(void *d, void *a, void *b, uint32_t desc)
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smin8)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int8_t)) {
int8_t aa = *(int8_t *)(a + i);
int8_t bb = *(int8_t *)(b + i);
int8_t dd = aa < bb ? aa : bb;
*(int8_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smin16)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int16_t)) {
int16_t aa = *(int16_t *)(a + i);
int16_t bb = *(int16_t *)(b + i);
int16_t dd = aa < bb ? aa : bb;
*(int16_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smin32)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int32_t)) {
int32_t aa = *(int32_t *)(a + i);
int32_t bb = *(int32_t *)(b + i);
int32_t dd = aa < bb ? aa : bb;
*(int32_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smin64)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int64_t)) {
int64_t aa = *(int64_t *)(a + i);
int64_t bb = *(int64_t *)(b + i);
int64_t dd = aa < bb ? aa : bb;
*(int64_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smax8)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int8_t)) {
int8_t aa = *(int8_t *)(a + i);
int8_t bb = *(int8_t *)(b + i);
int8_t dd = aa > bb ? aa : bb;
*(int8_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smax16)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int16_t)) {
int16_t aa = *(int16_t *)(a + i);
int16_t bb = *(int16_t *)(b + i);
int16_t dd = aa > bb ? aa : bb;
*(int16_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smax32)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int32_t)) {
int32_t aa = *(int32_t *)(a + i);
int32_t bb = *(int32_t *)(b + i);
int32_t dd = aa > bb ? aa : bb;
*(int32_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_smax64)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(int64_t)) {
int64_t aa = *(int64_t *)(a + i);
int64_t bb = *(int64_t *)(b + i);
int64_t dd = aa > bb ? aa : bb;
*(int64_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umin8)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
uint8_t aa = *(uint8_t *)(a + i);
uint8_t bb = *(uint8_t *)(b + i);
uint8_t dd = aa < bb ? aa : bb;
*(uint8_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umin16)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint16_t)) {
uint16_t aa = *(uint16_t *)(a + i);
uint16_t bb = *(uint16_t *)(b + i);
uint16_t dd = aa < bb ? aa : bb;
*(uint16_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umin32)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
uint32_t aa = *(uint32_t *)(a + i);
uint32_t bb = *(uint32_t *)(b + i);
uint32_t dd = aa < bb ? aa : bb;
*(uint32_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umin64)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint64_t)) {
uint64_t aa = *(uint64_t *)(a + i);
uint64_t bb = *(uint64_t *)(b + i);
uint64_t dd = aa < bb ? aa : bb;
*(uint64_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umax8)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint8_t)) {
uint8_t aa = *(uint8_t *)(a + i);
uint8_t bb = *(uint8_t *)(b + i);
uint8_t dd = aa > bb ? aa : bb;
*(uint8_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umax16)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint16_t)) {
uint16_t aa = *(uint16_t *)(a + i);
uint16_t bb = *(uint16_t *)(b + i);
uint16_t dd = aa > bb ? aa : bb;
*(uint16_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umax32)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint32_t)) {
uint32_t aa = *(uint32_t *)(a + i);
uint32_t bb = *(uint32_t *)(b + i);
uint32_t dd = aa > bb ? aa : bb;
*(uint32_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}
void HELPER(gvec_umax64)(void *d, void *a, void *b, uint32_t desc)
{
intptr_t oprsz = simd_oprsz(desc);
intptr_t i;
for (i = 0; i < oprsz; i += sizeof(uint64_t)) {
uint64_t aa = *(uint64_t *)(a + i);
uint64_t bb = *(uint64_t *)(b + i);
uint64_t dd = aa > bb ? aa : bb;
*(uint64_t *)(d + i) = dd;
}
clear_high(d, oprsz, desc);
}

23
accel/tcg/tcg-runtime.h
View File

@ -200,6 +200,26 @@ DEF_HELPER_FLAGS_4(gvec_ussub16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_ussub32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_ussub64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smin8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smin16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smin32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smin64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smax8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smax16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smax32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_smax64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umin8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umin16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umin32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umin64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umax8, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umax16, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umax32, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_umax64, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_neg8, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_neg16, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
DEF_HELPER_FLAGS_3(gvec_neg32, TCG_CALL_NO_RWG, void, ptr, ptr, i32)
@ -211,6 +231,9 @@ DEF_HELPER_FLAGS_4(gvec_or, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_xor, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_andc, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_orc, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_nand, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_nor, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_eqv, TCG_CALL_NO_RWG, void, ptr, ptr, ptr, i32)
DEF_HELPER_FLAGS_4(gvec_ands, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)
DEF_HELPER_FLAGS_4(gvec_xors, TCG_CALL_NO_RWG, void, ptr, ptr, i64, i32)

71
include/exec/cpu-defs.h
View File

@ -67,37 +67,23 @@ typedef uint64_t target_ulong;
#define CPU_TLB_ENTRY_BITS 5
#endif
/* TCG_TARGET_TLB_DISPLACEMENT_BITS is used in CPU_TLB_BITS to ensure that
* the TLB is not unnecessarily small, but still small enough for the
* TLB lookup instruction sequence used by the TCG target.
*
* TCG will have to generate an operand as large as the distance between
* env and the tlb_table[NB_MMU_MODES - 1][0].addend. For simplicity,
* the TCG targets just round everything up to the next power of two, and
* count bits. This works because: 1) the size of each TLB is a largish
* power of two, 2) and because the limit of the displacement is really close
* to a power of two, 3) the offset of tlb_table[0][0] inside env is smaller
* than the size of a TLB.
*
* For example, the maximum displacement 0xFFF0 on PPC and MIPS, but TCG
* just says "the displacement is 16 bits". TCG_TARGET_TLB_DISPLACEMENT_BITS
* then ensures that tlb_table at least 0x8000 bytes large ("not unnecessarily
* small": 2^15). The operand then will come up smaller than 0xFFF0 without
* any particular care, because the TLB for a single MMU mode is larger than
* 0x10000-0xFFF0=16 bytes. In the end, the maximum value of the operand
* could be something like 0xC000 (the offset of the last TLB table) plus
* 0x18 (the offset of the addend field in each TLB entry) plus the offset
* of tlb_table inside env (which is non-trivial but not huge).
*/
#define CPU_TLB_BITS \
MIN(8, \
TCG_TARGET_TLB_DISPLACEMENT_BITS - CPU_TLB_ENTRY_BITS - \
(NB_MMU_MODES <= 1 ? 0 : \
NB_MMU_MODES <= 2 ? 1 : \
NB_MMU_MODES <= 4 ? 2 : \
NB_MMU_MODES <= 8 ? 3 : 4))
#define CPU_TLB_DYN_MIN_BITS 6
#define CPU_TLB_DYN_DEFAULT_BITS 8
#define CPU_TLB_SIZE (1 << CPU_TLB_BITS)
# if HOST_LONG_BITS == 32
/* Make sure we do not require a double-word shift for the TLB load */
# define CPU_TLB_DYN_MAX_BITS (32 - TARGET_PAGE_BITS)
# else /* HOST_LONG_BITS == 64 */
/*
* Assuming TARGET_PAGE_BITS==12, with 2**22 entries we can cover 2**(22+12) ==
* 2**34 == 16G of address space. This is roughly what one would expect a
* TLB to cover in a modern (as of 2018) x86_64 CPU. For instance, Intel
* Skylake's Level-2 STLB has 16 1G entries.
* Also, make sure we do not size the TLB past the guest's address space.
*/
# define CPU_TLB_DYN_MAX_BITS \
MIN(22, TARGET_VIRT_ADDR_SPACE_BITS - TARGET_PAGE_BITS)
# endif
typedef struct CPUTLBEntry {
/* bit TARGET_LONG_BITS to TARGET_PAGE_BITS : virtual address
@ -141,6 +127,18 @@ typedef struct CPUIOTLBEntry {
MemTxAttrs attrs;
} CPUIOTLBEntry;
/**
* struct CPUTLBWindow
* @begin_ns: host time (in ns) at the beginning of the time window
* @max_entries: maximum number of entries observed in the window
*
* See also: tlb_mmu_resize_locked()
*/
typedef struct CPUTLBWindow {
int64_t begin_ns;
size_t max_entries;
} CPUTLBWindow;
typedef struct CPUTLBDesc {
/*
* Describe a region covering all of the large pages allocated
@ -152,6 +150,8 @@ typedef struct CPUTLBDesc {
target_ulong large_page_mask;
/* The next index to use in the tlb victim table. */
size_t vindex;
CPUTLBWindow window;
size_t n_used_entries;
} CPUTLBDesc;
/*
@ -176,6 +176,13 @@ typedef struct CPUTLBCommon {
size_t elide_flush_count;
} CPUTLBCommon;
# define CPU_TLB \
/* tlb_mask[i] contains (n_entries - 1) << CPU_TLB_ENTRY_BITS */ \
uintptr_t tlb_mask[NB_MMU_MODES]; \
CPUTLBEntry *tlb_table[NB_MMU_MODES];
# define CPU_IOTLB \
CPUIOTLBEntry *iotlb[NB_MMU_MODES];
/*
* The meaning of each of the MMU modes is defined in the target code.
* Note that NB_MMU_MODES is not yet defined; we can only reference it
@ -184,9 +191,9 @@ typedef struct CPUTLBCommon {
#define CPU_COMMON_TLB \
CPUTLBCommon tlb_c; \
CPUTLBDesc tlb_d[NB_MMU_MODES]; \
CPUTLBEntry tlb_table[NB_MMU_MODES][CPU_TLB_SIZE]; \
CPU_TLB \
CPUTLBEntry tlb_v_table[NB_MMU_MODES][CPU_VTLB_SIZE]; \
CPUIOTLBEntry iotlb[NB_MMU_MODES][CPU_TLB_SIZE]; \
CPU_IOTLB \
CPUIOTLBEntry iotlb_v[NB_MMU_MODES][CPU_VTLB_SIZE];
#else

9
include/exec/cpu_ldst.h
View File

@ -139,7 +139,14 @@ static inline target_ulong tlb_addr_write(const CPUTLBEntry *entry)
static inline uintptr_t tlb_index(CPUArchState *env, uintptr_t mmu_idx,
target_ulong addr)
{
return (addr >> TARGET_PAGE_BITS) & (CPU_TLB_SIZE - 1);
uintptr_t size_mask = env->tlb_mask[mmu_idx] >> CPU_TLB_ENTRY_BITS;
return (addr >> TARGET_PAGE_BITS) & size_mask;
}
static inline size_t tlb_n_entries(CPUArchState *env, uintptr_t mmu_idx)
{
return (env->tlb_mask[mmu_idx] >> CPU_TLB_ENTRY_BITS) + 1;
}
/* Find the TLB entry corresponding to the mmu_idx + address pair. */

19
tcg/README
View File

@ -554,6 +554,25 @@ E.g. VECL=1 -> 64 << 1 -> v128, and VECE=2 -> 1 << 2 -> i32.
Similarly, v0 = -v1.
* smin_vec:
* umin_vec:
Similarly, v0 = MIN(v1, v2), for signed and unsigned element types.
* smax_vec:
* umax_vec:
Similarly, v0 = MAX(v1, v2), for signed and unsigned element types.
* ssadd_vec:
* sssub_vec:
* usadd_vec:
* ussub_vec:
Signed and unsigned saturating addition and subtraction. If the true
result is not representable within the element type, the element is
set to the minimum or maximum value for the type.
* and_vec v0, v1, v2
* or_vec v0, v1, v2
* xor_vec v0, v1, v2

2
tcg/aarch64/tcg-target.h
View File

@ -135,6 +135,8 @@ typedef enum {
#define TCG_TARGET_HAS_shv_vec 0
#define TCG_TARGET_HAS_cmp_vec 1
#define TCG_TARGET_HAS_mul_vec 1
#define TCG_TARGET_HAS_sat_vec 1
#define TCG_TARGET_HAS_minmax_vec 1
#define TCG_TARGET_DEFAULT_MO (0)
#define TCG_TARGET_HAS_MEMORY_BSWAP 1

148
tcg/aarch64/tcg-target.inc.c
View File

@ -498,6 +498,9 @@ typedef enum {
I3510_EON = 0x4a200000,
I3510_ANDS = 0x6a000000,
/* Logical shifted register instructions (with a shift). */
I3502S_AND_LSR = I3510_AND | (1 << 22),
/* AdvSIMD copy */
I3605_DUP = 0x0e000400,
I3605_INS = 0x4e001c00,
@ -528,6 +531,14 @@ typedef enum {
I3616_CMHI = 0x2e203400,
I3616_CMHS = 0x2e203c00,
I3616_CMEQ = 0x2e208c00,
I3616_SMAX = 0x0e206400,
I3616_SMIN = 0x0e206c00,
I3616_SQADD = 0x0e200c00,
I3616_SQSUB = 0x0e202c00,
I3616_UMAX = 0x2e206400,
I3616_UMIN = 0x2e206c00,
I3616_UQADD = 0x2e200c00,
I3616_UQSUB = 0x2e202c00,
/* AdvSIMD two-reg misc. */
I3617_CMGT0 = 0x0e208800,
@ -1440,6 +1451,14 @@ static void add_qemu_ldst_label(TCGContext *s, bool is_ld, TCGMemOpIdx oi,
label->label_ptr[0] = label_ptr;
}
/* We expect tlb_mask to be before tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
offsetof(CPUArchState, tlb_mask));
/* We expect to use a 24-bit unsigned offset from ENV. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table[NB_MMU_MODES - 1])
> 0xffffff);
/* Load and compare a TLB entry, emitting the conditional jump to the
slow path for the failure case, which will be patched later when finalizing
the slow path. Generated code returns the host addend in X1,
@ -1448,15 +1467,55 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, TCGMemOp opc,
tcg_insn_unit **label_ptr, int mem_index,
bool is_read)
{
int tlb_offset = is_read ?
offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
: offsetof(CPUArchState, tlb_table[mem_index][0].addr_write);
int mask_ofs = offsetof(CPUArchState, tlb_mask[mem_index]);
int table_ofs = offsetof(CPUArchState, tlb_table[mem_index]);
unsigned a_bits = get_alignment_bits(opc);
unsigned s_bits = opc & MO_SIZE;
unsigned a_mask = (1u << a_bits) - 1;
unsigned s_mask = (1u << s_bits) - 1;
TCGReg base = TCG_AREG0, x3;
uint64_t tlb_mask;
TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0, x3;
TCGType mask_type;
uint64_t compare_mask;
if (table_ofs > 0xfff) {
int table_hi = table_ofs & ~0xfff;
int mask_hi = mask_ofs & ~0xfff;
table_base = TCG_REG_X1;
if (mask_hi == table_hi) {
mask_base = table_base;
} else if (mask_hi) {
mask_base = TCG_REG_X0;
tcg_out_insn(s, 3401, ADDI, TCG_TYPE_I64,
mask_base, TCG_AREG0, mask_hi);
}
tcg_out_insn(s, 3401, ADDI, TCG_TYPE_I64,
table_base, TCG_AREG0, table_hi);
mask_ofs -= mask_hi;
table_ofs -= table_hi;
}
mask_type = (TARGET_PAGE_BITS + CPU_TLB_DYN_MAX_BITS > 32
? TCG_TYPE_I64 : TCG_TYPE_I32);
/* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
tcg_out_ld(s, mask_type, TCG_REG_X0, mask_base, mask_ofs);
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_X1, table_base, table_ofs);
/* Extract the TLB index from the address into X0. */
tcg_out_insn(s, 3502S, AND_LSR, mask_type == TCG_TYPE_I64,
TCG_REG_X0, TCG_REG_X0, addr_reg,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
/* Add the tlb_table pointer, creating the CPUTLBEntry address into X1. */
tcg_out_insn(s, 3502, ADD, 1, TCG_REG_X1, TCG_REG_X1, TCG_REG_X0);
/* Load the tlb comparator into X0, and the fast path addend into X1. */
tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_X0, TCG_REG_X1, is_read
? offsetof(CPUTLBEntry, addr_read)
: offsetof(CPUTLBEntry, addr_write));
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_X1, TCG_REG_X1,
offsetof(CPUTLBEntry, addend));
/* For aligned accesses, we check the first byte and include the alignment
bits within the address. For unaligned access, we check that we don't
@ -1468,47 +1527,14 @@ static void tcg_out_tlb_read(TCGContext *s, TCGReg addr_reg, TCGMemOp opc,
TCG_REG_X3, addr_reg, s_mask - a_mask);
x3 = TCG_REG_X3;
}
tlb_mask = (uint64_t)TARGET_PAGE_MASK | a_mask;
/* Extract the TLB index from the address into X0.
X0<CPU_TLB_BITS:0> =
addr_reg<TARGET_PAGE_BITS+CPU_TLB_BITS:TARGET_PAGE_BITS> */
tcg_out_ubfm(s, TARGET_LONG_BITS == 64, TCG_REG_X0, addr_reg,
TARGET_PAGE_BITS, TARGET_PAGE_BITS + CPU_TLB_BITS);
compare_mask = (uint64_t)TARGET_PAGE_MASK | a_mask;
/* Store the page mask part of the address into X3. */
tcg_out_logicali(s, I3404_ANDI, TARGET_LONG_BITS == 64,
TCG_REG_X3, x3, tlb_mask);
/* Add any "high bits" from the tlb offset to the env address into X2,
to take advantage of the LSL12 form of the ADDI instruction.
X2 = env + (tlb_offset & 0xfff000) */
if (tlb_offset & 0xfff000) {
tcg_out_insn(s, 3401, ADDI, TCG_TYPE_I64, TCG_REG_X2, base,
tlb_offset & 0xfff000);
base = TCG_REG_X2;
}
/* Merge the tlb index contribution into X2.
X2 = X2 + (X0 << CPU_TLB_ENTRY_BITS) */
tcg_out_insn(s, 3502S, ADD_LSL, TCG_TYPE_I64, TCG_REG_X2, base,
TCG_REG_X0, CPU_TLB_ENTRY_BITS);
/* Merge "low bits" from tlb offset, load the tlb comparator into X0.
X0 = load [X2 + (tlb_offset & 0x000fff)] */
tcg_out_ldst(s, TARGET_LONG_BITS == 32 ? I3312_LDRW : I3312_LDRX,
TCG_REG_X0, TCG_REG_X2, tlb_offset & 0xfff,
TARGET_LONG_BITS == 32 ? 2 : 3);
/* Load the tlb addend. Do that early to avoid stalling.
X1 = load [X2 + (tlb_offset & 0xfff) + offsetof(addend)] */
tcg_out_ldst(s, I3312_LDRX, TCG_REG_X1, TCG_REG_X2,
(tlb_offset & 0xfff) + (offsetof(CPUTLBEntry, addend)) -
(is_read ? offsetof(CPUTLBEntry, addr_read)
: offsetof(CPUTLBEntry, addr_write)), 3);
TCG_REG_X3, x3, compare_mask);
/* Perform the address comparison. */
tcg_out_cmp(s, (TARGET_LONG_BITS == 64), TCG_REG_X0, TCG_REG_X3, 0);
tcg_out_cmp(s, TARGET_LONG_BITS == 64, TCG_REG_X0, TCG_REG_X3, 0);
/* If not equal, we jump to the slow path. */
*label_ptr = s->code_ptr;
@ -2137,6 +2163,30 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
case INDEX_op_orc_vec:
tcg_out_insn(s, 3616, ORN, is_q, 0, a0, a1, a2);
break;
case INDEX_op_ssadd_vec:
tcg_out_insn(s, 3616, SQADD, is_q, vece, a0, a1, a2);
break;
case INDEX_op_sssub_vec:
tcg_out_insn(s, 3616, SQSUB, is_q, vece, a0, a1, a2);
break;
case INDEX_op_usadd_vec:
tcg_out_insn(s, 3616, UQADD, is_q, vece, a0, a1, a2);
break;
case INDEX_op_ussub_vec:
tcg_out_insn(s, 3616, UQSUB, is_q, vece, a0, a1, a2);
break;
case INDEX_op_smax_vec:
tcg_out_insn(s, 3616, SMAX, is_q, vece, a0, a1, a2);
break;
case INDEX_op_smin_vec:
tcg_out_insn(s, 3616, SMIN, is_q, vece, a0, a1, a2);
break;
case INDEX_op_umax_vec:
tcg_out_insn(s, 3616, UMAX, is_q, vece, a0, a1, a2);
break;
case INDEX_op_umin_vec:
tcg_out_insn(s, 3616, UMIN, is_q, vece, a0, a1, a2);
break;
case INDEX_op_not_vec:
tcg_out_insn(s, 3617, NOT, is_q, 0, a0, a1);
break;
@ -2207,6 +2257,14 @@ int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
case INDEX_op_shli_vec:
case INDEX_op_shri_vec:
case INDEX_op_sari_vec:
case INDEX_op_ssadd_vec:
case INDEX_op_sssub_vec:
case INDEX_op_usadd_vec:
case INDEX_op_ussub_vec:
case INDEX_op_smax_vec:
case INDEX_op_smin_vec:
case INDEX_op_umax_vec:
case INDEX_op_umin_vec:
return 1;
case INDEX_op_mul_vec:
return vece < MO_64;
@ -2386,6 +2444,14 @@ static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
case INDEX_op_xor_vec:
case INDEX_op_andc_vec:
case INDEX_op_orc_vec:
case INDEX_op_ssadd_vec:
case INDEX_op_sssub_vec:
case INDEX_op_usadd_vec:
case INDEX_op_ussub_vec:
case INDEX_op_smax_vec:
case INDEX_op_smin_vec:
case INDEX_op_umax_vec:
case INDEX_op_umin_vec:
return &w_w_w;
case INDEX_op_not_vec:
case INDEX_op_neg_vec:

143
tcg/arm/tcg-target.inc.c
View File

@ -500,6 +500,12 @@ static inline void tcg_out_ldrd_r(TCGContext *s, int cond, TCGReg rt,
tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 0);
}
static inline void tcg_out_ldrd_rwb(TCGContext *s, int cond, TCGReg rt,
TCGReg rn, TCGReg rm)
{
tcg_out_memop_r(s, cond, INSN_LDRD_REG, rt, rn, rm, 1, 1, 1);
}
static inline void tcg_out_strd_8(TCGContext *s, int cond, TCGReg rt,
TCGReg rn, int imm8)
{
@ -1229,8 +1235,13 @@ static TCGReg tcg_out_arg_reg64(TCGContext *s, TCGReg argreg,
#define TLB_SHIFT (CPU_TLB_ENTRY_BITS + CPU_TLB_BITS)
/* We're expecting to use an 8-bit immediate and to mask. */
QEMU_BUILD_BUG_ON(CPU_TLB_BITS > 8);
/* We expect tlb_mask to be before tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
offsetof(CPUArchState, tlb_mask));
/* We expect to use a 20-bit unsigned offset from ENV. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table[NB_MMU_MODES - 1])
> 0xfffff);
/* Load and compare a TLB entry, leaving the flags set. Returns the register
containing the addend of the tlb entry. Clobbers R0, R1, R2, TMP. */
@ -1238,84 +1249,72 @@ QEMU_BUILD_BUG_ON(CPU_TLB_BITS > 8);
static TCGReg tcg_out_tlb_read(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
TCGMemOp opc, int mem_index, bool is_load)
{
TCGReg base = TCG_AREG0;
int cmp_off =
(is_load
? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
: offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
int mask_off;
int cmp_off = (is_load ? offsetof(CPUTLBEntry, addr_read)
: offsetof(CPUTLBEntry, addr_write));
int mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
int table_off = offsetof(CPUArchState, tlb_table[mem_index]);
TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0;
unsigned s_bits = opc & MO_SIZE;
unsigned a_bits = get_alignment_bits(opc);
/* V7 generates the following:
* ubfx r0, addrlo, #TARGET_PAGE_BITS, #CPU_TLB_BITS
* add r2, env, #high
* add r2, r2, r0, lsl #CPU_TLB_ENTRY_BITS
* ldr r0, [r2, #cmp]
* ldr r2, [r2, #add]
* movw tmp, #page_align_mask
* bic tmp, addrlo, tmp
* cmp r0, tmp
*
* Otherwise we generate:
* shr tmp, addrlo, #TARGET_PAGE_BITS
* add r2, env, #high
* and r0, tmp, #(CPU_TLB_SIZE - 1)
* add r2, r2, r0, lsl #CPU_TLB_ENTRY_BITS
* ldr r0, [r2, #cmp]
* ldr r2, [r2, #add]
* tst addrlo, #s_mask
* cmpeq r0, tmp, lsl #TARGET_PAGE_BITS
*/
if (use_armv7_instructions) {
tcg_out_extract(s, COND_AL, TCG_REG_R0, addrlo,
TARGET_PAGE_BITS, CPU_TLB_BITS);
} else {
tcg_out_dat_reg(s, COND_AL, ARITH_MOV, TCG_REG_TMP,
0, addrlo, SHIFT_IMM_LSR(TARGET_PAGE_BITS));
}
if (table_off > 0xfff) {
int mask_hi = mask_off & ~0xfff;
int table_hi = table_off & ~0xfff;
int rot;
/* Add portions of the offset until the memory access is in range.
* If we plan on using ldrd, reduce to an 8-bit offset; otherwise
* we can use a 12-bit offset. */
if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
mask_off = 0xff;
} else {
mask_off = 0xfff;
}
while (cmp_off > mask_off) {
int shift = ctz32(cmp_off & ~mask_off) & ~1;
int rot = ((32 - shift) << 7) & 0xf00;
int addend = cmp_off & (0xff << shift);
tcg_out_dat_imm(s, COND_AL, ARITH_ADD, TCG_REG_R2, base,
rot | ((cmp_off >> shift) & 0xff));
base = TCG_REG_R2;
add_off -= addend;
cmp_off -= addend;
}
if (!use_armv7_instructions) {
tcg_out_dat_imm(s, COND_AL, ARITH_AND,
TCG_REG_R0, TCG_REG_TMP, CPU_TLB_SIZE - 1);
}
tcg_out_dat_reg(s, COND_AL, ARITH_ADD, TCG_REG_R2, base,
TCG_REG_R0, SHIFT_IMM_LSL(CPU_TLB_ENTRY_BITS));
/* Load the tlb comparator. Use ldrd if needed and available,
but due to how the pointer needs setting up, ldm isn't useful.
Base arm5 doesn't have ldrd, but armv5te does. */
if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_REG_R2, cmp_off);
} else {
tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_R2, cmp_off);
if (TARGET_LONG_BITS == 64) {
tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R2, cmp_off + 4);
table_base = TCG_REG_R2;
if (mask_hi == table_hi) {
mask_base = table_base;
} else if (mask_hi) {
mask_base = TCG_REG_TMP;
rot = encode_imm(mask_hi);
assert(rot >= 0);
tcg_out_dat_imm(s, COND_AL, ARITH_ADD, mask_base, TCG_AREG0,
rotl(mask_hi, rot) | (rot << 7));
}
rot = encode_imm(table_hi);
assert(rot >= 0);
tcg_out_dat_imm(s, COND_AL, ARITH_ADD, table_base, TCG_AREG0,
rotl(table_hi, rot) | (rot << 7));
mask_off -= mask_hi;
table_off -= table_hi;
}
/* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP, mask_base, mask_off);
tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R2, table_base, table_off);
/* Extract the tlb index from the address into TMP. */
tcg_out_dat_reg(s, COND_AL, ARITH_AND, TCG_REG_TMP, TCG_REG_TMP, addrlo,
SHIFT_IMM_LSR(TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS));
/*
* Add the tlb_table pointer, creating the CPUTLBEntry address in R2.
* Load the tlb comparator into R0/R1 and the fast path addend into R2.
*/
if (cmp_off == 0) {
if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
tcg_out_ldrd_rwb(s, COND_AL, TCG_REG_R0, TCG_REG_R2, TCG_REG_TMP);
} else {
tcg_out_ld32_rwb(s, COND_AL, TCG_REG_R0, TCG_REG_R2, TCG_REG_TMP);
}
} else {
tcg_out_dat_reg(s, COND_AL, ARITH_ADD,
TCG_REG_R2, TCG_REG_R2, TCG_REG_TMP, 0);
if (use_armv6_instructions && TARGET_LONG_BITS == 64) {
tcg_out_ldrd_8(s, COND_AL, TCG_REG_R0, TCG_REG_R2, cmp_off);
} else {
tcg_out_ld32_12(s, COND_AL, TCG_REG_R0, TCG_REG_R2, cmp_off);
}
}
if (!use_armv6_instructions && TARGET_LONG_BITS == 64) {
tcg_out_ld32_12(s, COND_AL, TCG_REG_R1, TCG_REG_R2, cmp_off + 4);
}
/* Load the tlb addend. */
tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R2, add_off);
tcg_out_ld32_12(s, COND_AL, TCG_REG_R2, TCG_REG_R2,
offsetof(CPUTLBEntry, addend));
/* Check alignment. We don't support inline unaligned acceses,
but we can easily support overalignment checks. */

2
tcg/i386/tcg-target.h
View File

@ -185,6 +185,8 @@ extern bool have_avx2;
#define TCG_TARGET_HAS_shv_vec 0
#define TCG_TARGET_HAS_cmp_vec 1
#define TCG_TARGET_HAS_mul_vec 1
#define TCG_TARGET_HAS_sat_vec 1
#define TCG_TARGET_HAS_minmax_vec 1
#define TCG_TARGET_deposit_i32_valid(ofs, len) \
(((ofs) == 0 && (len) == 8) || ((ofs) == 8 && (len) == 8) || \

608
tcg/i386/tcg-target.inc.c
View File

@ -329,6 +329,7 @@ static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
#define OPC_ARITH_GvEv (0x03) /* ... plus (ARITH_FOO << 3) */
#define OPC_ANDN (0xf2 | P_EXT38)
#define OPC_ADD_GvEv (OPC_ARITH_GvEv | (ARITH_ADD << 3))
#define OPC_AND_GvEv (OPC_ARITH_GvEv | (ARITH_AND << 3))
#define OPC_BLENDPS (0x0c | P_EXT3A | P_DATA16)
#define OPC_BSF (0xbc | P_EXT)
#define OPC_BSR (0xbd | P_EXT)
@ -377,6 +378,10 @@ static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
#define OPC_PADDW (0xfd | P_EXT | P_DATA16)
#define OPC_PADDD (0xfe | P_EXT | P_DATA16)
#define OPC_PADDQ (0xd4 | P_EXT | P_DATA16)
#define OPC_PADDSB (0xec | P_EXT | P_DATA16)
#define OPC_PADDSW (0xed | P_EXT | P_DATA16)
#define OPC_PADDUB (0xdc | P_EXT | P_DATA16)
#define OPC_PADDUW (0xdd | P_EXT | P_DATA16)
#define OPC_PAND (0xdb | P_EXT | P_DATA16)
#define OPC_PANDN (0xdf | P_EXT | P_DATA16)
#define OPC_PBLENDW (0x0e | P_EXT3A | P_DATA16)
@ -388,6 +393,18 @@ static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
#define OPC_PCMPGTW (0x65 | P_EXT | P_DATA16)
#define OPC_PCMPGTD (0x66 | P_EXT | P_DATA16)
#define OPC_PCMPGTQ (0x37 | P_EXT38 | P_DATA16)
#define OPC_PMAXSB (0x3c | P_EXT38 | P_DATA16)
#define OPC_PMAXSW (0xee | P_EXT | P_DATA16)
#define OPC_PMAXSD (0x3d | P_EXT38 | P_DATA16)
#define OPC_PMAXUB (0xde | P_EXT | P_DATA16)
#define OPC_PMAXUW (0x3e | P_EXT38 | P_DATA16)
#define OPC_PMAXUD (0x3f | P_EXT38 | P_DATA16)
#define OPC_PMINSB (0x38 | P_EXT38 | P_DATA16)
#define OPC_PMINSW (0xea | P_EXT | P_DATA16)
#define OPC_PMINSD (0x39 | P_EXT38 | P_DATA16)
#define OPC_PMINUB (0xda | P_EXT | P_DATA16)
#define OPC_PMINUW (0x3a | P_EXT38 | P_DATA16)
#define OPC_PMINUD (0x3b | P_EXT38 | P_DATA16)
#define OPC_PMOVSXBW (0x20 | P_EXT38 | P_DATA16)
#define OPC_PMOVSXWD (0x23 | P_EXT38 | P_DATA16)
#define OPC_PMOVSXDQ (0x25 | P_EXT38 | P_DATA16)
@ -408,6 +425,10 @@ static inline int tcg_target_const_match(tcg_target_long val, TCGType type,
#define OPC_PSUBW (0xf9 | P_EXT | P_DATA16)
#define OPC_PSUBD (0xfa | P_EXT | P_DATA16)
#define OPC_PSUBQ (0xfb | P_EXT | P_DATA16)
#define OPC_PSUBSB (0xe8 | P_EXT | P_DATA16)
#define OPC_PSUBSW (0xe9 | P_EXT | P_DATA16)
#define OPC_PSUBUB (0xd8 | P_EXT | P_DATA16)
#define OPC_PSUBUW (0xd9 | P_EXT | P_DATA16)
#define OPC_PUNPCKLBW (0x60 | P_EXT | P_DATA16)
#define OPC_PUNPCKLWD (0x61 | P_EXT | P_DATA16)
#define OPC_PUNPCKLDQ (0x62 | P_EXT | P_DATA16)
@ -1621,7 +1642,7 @@ static inline void tcg_out_tlb_load(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
}
if (TCG_TYPE_PTR == TCG_TYPE_I64) {
hrexw = P_REXW;
if (TARGET_PAGE_BITS + CPU_TLB_BITS > 32) {
if (TARGET_PAGE_BITS + CPU_TLB_DYN_MAX_BITS > 32) {
tlbtype = TCG_TYPE_I64;
tlbrexw = P_REXW;
}
@ -1629,6 +1650,15 @@ static inline void tcg_out_tlb_load(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
}
tcg_out_mov(s, tlbtype, r0, addrlo);
tcg_out_shifti(s, SHIFT_SHR + tlbrexw, r0,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tcg_out_modrm_offset(s, OPC_AND_GvEv + trexw, r0, TCG_AREG0,
offsetof(CPUArchState, tlb_mask[mem_index]));
tcg_out_modrm_offset(s, OPC_ADD_GvEv + hrexw, r0, TCG_AREG0,
offsetof(CPUArchState, tlb_table[mem_index]));
/* If the required alignment is at least as large as the access, simply
copy the address and mask. For lesser alignments, check that we don't
cross pages for the complete access. */
@ -1638,20 +1668,10 @@ static inline void tcg_out_tlb_load(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
tcg_out_modrm_offset(s, OPC_LEA + trexw, r1, addrlo, s_mask - a_mask);
}
tlb_mask = (target_ulong)TARGET_PAGE_MASK | a_mask;
tcg_out_shifti(s, SHIFT_SHR + tlbrexw, r0,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tgen_arithi(s, ARITH_AND + trexw, r1, tlb_mask, 0);
tgen_arithi(s, ARITH_AND + tlbrexw, r0,
(CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS, 0);
tcg_out_modrm_sib_offset(s, OPC_LEA + hrexw, r0, TCG_AREG0, r0, 0,
offsetof(CPUArchState, tlb_table[mem_index][0])
+ which);
/* cmp 0(r0), r1 */
tcg_out_modrm_offset(s, OPC_CMP_GvEv + trexw, r1, r0, 0);
tcg_out_modrm_offset(s, OPC_CMP_GvEv + trexw, r1, r0, which);
/* Prepare for both the fast path add of the tlb addend, and the slow
path function argument setup. */
@ -1664,7 +1684,7 @@ static inline void tcg_out_tlb_load(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
if (TARGET_LONG_BITS > TCG_TARGET_REG_BITS) {
/* cmp 4(r0), addrhi */
tcg_out_modrm_offset(s, OPC_CMP_GvEv, addrhi, r0, 4);
tcg_out_modrm_offset(s, OPC_CMP_GvEv, addrhi, r0, which + 4);
/* jne slow_path */
tcg_out_opc(s, OPC_JCC_long + JCC_JNE, 0, 0, 0);
@ -1676,7 +1696,7 @@ static inline void tcg_out_tlb_load(TCGContext *s, TCGReg addrlo, TCGReg addrhi,
/* add addend(r0), r1 */
tcg_out_modrm_offset(s, OPC_ADD_GvEv + hrexw, r1, r0,
offsetof(CPUTLBEntry, addend) - which);
offsetof(CPUTLBEntry, addend));
}
/*
@ -2591,9 +2611,21 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
static int const add_insn[4] = {
OPC_PADDB, OPC_PADDW, OPC_PADDD, OPC_PADDQ
};
static int const ssadd_insn[4] = {
OPC_PADDSB, OPC_PADDSW, OPC_UD2, OPC_UD2
};
static int const usadd_insn[4] = {
OPC_PADDSB, OPC_PADDSW, OPC_UD2, OPC_UD2
};
static int const sub_insn[4] = {
OPC_PSUBB, OPC_PSUBW, OPC_PSUBD, OPC_PSUBQ
};
static int const sssub_insn[4] = {
OPC_PSUBSB, OPC_PSUBSW, OPC_UD2, OPC_UD2
};
static int const ussub_insn[4] = {
OPC_PSUBSB, OPC_PSUBSW, OPC_UD2, OPC_UD2
};
static int const mul_insn[4] = {
OPC_UD2, OPC_PMULLW, OPC_PMULLD, OPC_UD2
};
@ -2618,6 +2650,18 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
static int const packus_insn[4] = {
OPC_PACKUSWB, OPC_PACKUSDW, OPC_UD2, OPC_UD2
};
static int const smin_insn[4] = {
OPC_PMINSB, OPC_PMINSW, OPC_PMINSD, OPC_UD2
};
static int const smax_insn[4] = {
OPC_PMAXSB, OPC_PMAXSW, OPC_PMAXSD, OPC_UD2
};
static int const umin_insn[4] = {
OPC_PMINUB, OPC_PMINUW, OPC_PMINUD, OPC_UD2
};
static int const umax_insn[4] = {
OPC_PMAXUB, OPC_PMAXUW, OPC_PMAXUD, OPC_UD2
};
TCGType type = vecl + TCG_TYPE_V64;
int insn, sub;
@ -2631,9 +2675,21 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
case INDEX_op_add_vec:
insn = add_insn[vece];
goto gen_simd;
case INDEX_op_ssadd_vec:
insn = ssadd_insn[vece];
goto gen_simd;
case INDEX_op_usadd_vec:
insn = usadd_insn[vece];
goto gen_simd;
case INDEX_op_sub_vec:
insn = sub_insn[vece];
goto gen_simd;
case INDEX_op_sssub_vec:
insn = sssub_insn[vece];
goto gen_simd;
case INDEX_op_ussub_vec:
insn = ussub_insn[vece];
goto gen_simd;
case INDEX_op_mul_vec:
insn = mul_insn[vece];
goto gen_simd;
@ -2646,6 +2702,18 @@ static void tcg_out_vec_op(TCGContext *s, TCGOpcode opc,
case INDEX_op_xor_vec:
insn = OPC_PXOR;
goto gen_simd;
case INDEX_op_smin_vec:
insn = smin_insn[vece];
goto gen_simd;
case INDEX_op_umin_vec:
insn = umin_insn[vece];
goto gen_simd;
case INDEX_op_smax_vec:
insn = smax_insn[vece];
goto gen_simd;
case INDEX_op_umax_vec:
insn = umax_insn[vece];
goto gen_simd;
case INDEX_op_x86_punpckl_vec:
insn = punpckl_insn[vece];
goto gen_simd;
@ -3007,6 +3075,14 @@ static const TCGTargetOpDef *tcg_target_op_def(TCGOpcode op)
case INDEX_op_or_vec:
case INDEX_op_xor_vec:
case INDEX_op_andc_vec:
case INDEX_op_ssadd_vec:
case INDEX_op_usadd_vec:
case INDEX_op_sssub_vec:
case INDEX_op_ussub_vec:
case INDEX_op_smin_vec:
case INDEX_op_umin_vec:
case INDEX_op_smax_vec:
case INDEX_op_umax_vec:
case INDEX_op_cmp_vec:
case INDEX_op_x86_shufps_vec:
case INDEX_op_x86_blend_vec:
@ -3074,258 +3150,310 @@ int tcg_can_emit_vec_op(TCGOpcode opc, TCGType type, unsigned vece)
}
return 1;
case INDEX_op_ssadd_vec:
case INDEX_op_usadd_vec:
case INDEX_op_sssub_vec:
case INDEX_op_ussub_vec:
return vece <= MO_16;
case INDEX_op_smin_vec:
case INDEX_op_smax_vec:
case INDEX_op_umin_vec:
case INDEX_op_umax_vec:
return vece <= MO_32 ? 1 : -1;
default:
return 0;
}
}
static void expand_vec_shi(TCGType type, unsigned vece, bool shr,
TCGv_vec v0, TCGv_vec v1, TCGArg imm)
{
TCGv_vec t1, t2;
tcg_debug_assert(vece == MO_8);
t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type);
/* Unpack to W, shift, and repack. Tricky bits:
(1) Use punpck*bw x,x to produce DDCCBBAA,
i.e. duplicate in other half of the 16-bit lane.
(2) For right-shift, add 8 so that the high half of
the lane becomes zero. For left-shift, we must
shift up and down again.
(3) Step 2 leaves high half zero such that PACKUSWB
(pack with unsigned saturation) does not modify
the quantity. */
vec_gen_3(INDEX_op_x86_punpckl_vec, type, MO_8,
tcgv_vec_arg(t1), tcgv_vec_arg(v1), tcgv_vec_arg(v1));
vec_gen_3(INDEX_op_x86_punpckh_vec, type, MO_8,
tcgv_vec_arg(t2), tcgv_vec_arg(v1), tcgv_vec_arg(v1));
if (shr) {
tcg_gen_shri_vec(MO_16, t1, t1, imm + 8);
tcg_gen_shri_vec(MO_16, t2, t2, imm + 8);
} else {
tcg_gen_shli_vec(MO_16, t1, t1, imm + 8);
tcg_gen_shli_vec(MO_16, t2, t2, imm + 8);
tcg_gen_shri_vec(MO_16, t1, t1, 8);
tcg_gen_shri_vec(MO_16, t2, t2, 8);
}
vec_gen_3(INDEX_op_x86_packus_vec, type, MO_8,
tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(t2));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
}
static void expand_vec_sari(TCGType type, unsigned vece,
TCGv_vec v0, TCGv_vec v1, TCGArg imm)
{
TCGv_vec t1, t2;
switch (vece) {
case MO_8:
/* Unpack to W, shift, and repack, as in expand_vec_shi. */
t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type);
vec_gen_3(INDEX_op_x86_punpckl_vec, type, MO_8,
tcgv_vec_arg(t1), tcgv_vec_arg(v1), tcgv_vec_arg(v1));
vec_gen_3(INDEX_op_x86_punpckh_vec, type, MO_8,
tcgv_vec_arg(t2), tcgv_vec_arg(v1), tcgv_vec_arg(v1));
tcg_gen_sari_vec(MO_16, t1, t1, imm + 8);
tcg_gen_sari_vec(MO_16, t2, t2, imm + 8);
vec_gen_3(INDEX_op_x86_packss_vec, type, MO_8,
tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(t2));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
break;
case MO_64:
if (imm <= 32) {
/* We can emulate a small sign extend by performing an arithmetic
* 32-bit shift and overwriting the high half of a 64-bit logical
* shift (note that the ISA says shift of 32 is valid).
*/
t1 = tcg_temp_new_vec(type);
tcg_gen_sari_vec(MO_32, t1, v1, imm);
tcg_gen_shri_vec(MO_64, v0, v1, imm);
vec_gen_4(INDEX_op_x86_blend_vec, type, MO_32,
tcgv_vec_arg(v0), tcgv_vec_arg(v0),
tcgv_vec_arg(t1), 0xaa);
tcg_temp_free_vec(t1);
} else {
/* Otherwise we will need to use a compare vs 0 to produce
* the sign-extend, shift and merge.
*/
t1 = tcg_const_zeros_vec(type);
tcg_gen_cmp_vec(TCG_COND_GT, MO_64, t1, t1, v1);
tcg_gen_shri_vec(MO_64, v0, v1, imm);
tcg_gen_shli_vec(MO_64, t1, t1, 64 - imm);
tcg_gen_or_vec(MO_64, v0, v0, t1);
tcg_temp_free_vec(t1);
}
break;
default:
g_assert_not_reached();
}
}
static void expand_vec_mul(TCGType type, unsigned vece,
TCGv_vec v0, TCGv_vec v1, TCGv_vec v2)
{
TCGv_vec t1, t2, t3, t4;
tcg_debug_assert(vece == MO_8);
/*
* Unpack v1 bytes to words, 0 | x.
* Unpack v2 bytes to words, y | 0.
* This leaves the 8-bit result, x * y, with 8 bits of right padding.
* Shift logical right by 8 bits to clear the high 8 bytes before
* using an unsigned saturated pack.
*
* The difference between the V64, V128 and V256 cases is merely how
* we distribute the expansion between temporaries.
*/
switch (type) {
case TCG_TYPE_V64:
t1 = tcg_temp_new_vec(TCG_TYPE_V128);
t2 = tcg_temp_new_vec(TCG_TYPE_V128);
tcg_gen_dup16i_vec(t2, 0);
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t1), tcgv_vec_arg(v1), tcgv_vec_arg(t2));
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t2), tcgv_vec_arg(t2), tcgv_vec_arg(v2));
tcg_gen_mul_vec(MO_16, t1, t1, t2);
tcg_gen_shri_vec(MO_16, t1, t1, 8);
vec_gen_3(INDEX_op_x86_packus_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(t1));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
break;
case TCG_TYPE_V128:
case TCG_TYPE_V256:
t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type);
t3 = tcg_temp_new_vec(type);
t4 = tcg_temp_new_vec(type);
tcg_gen_dup16i_vec(t4, 0);
vec_gen_3(INDEX_op_x86_punpckl_vec, type, MO_8,
tcgv_vec_arg(t1), tcgv_vec_arg(v1), tcgv_vec_arg(t4));
vec_gen_3(INDEX_op_x86_punpckl_vec, type, MO_8,
tcgv_vec_arg(t2), tcgv_vec_arg(t4), tcgv_vec_arg(v2));
vec_gen_3(INDEX_op_x86_punpckh_vec, type, MO_8,
tcgv_vec_arg(t3), tcgv_vec_arg(v1), tcgv_vec_arg(t4));
vec_gen_3(INDEX_op_x86_punpckh_vec, type, MO_8,
tcgv_vec_arg(t4), tcgv_vec_arg(t4), tcgv_vec_arg(v2));
tcg_gen_mul_vec(MO_16, t1, t1, t2);
tcg_gen_mul_vec(MO_16, t3, t3, t4);
tcg_gen_shri_vec(MO_16, t1, t1, 8);
tcg_gen_shri_vec(MO_16, t3, t3, 8);
vec_gen_3(INDEX_op_x86_packus_vec, type, MO_8,
tcgv_vec_arg(v0), tcgv_vec_arg(t1), tcgv_vec_arg(t3));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
tcg_temp_free_vec(t3);
tcg_temp_free_vec(t4);
break;
default:
g_assert_not_reached();
}
}
static void expand_vec_cmp(TCGType type, unsigned vece, TCGv_vec v0,
TCGv_vec v1, TCGv_vec v2, TCGCond cond)
{
enum {
NEED_SWAP = 1,
NEED_INV = 2,
NEED_BIAS = 4
};
static const uint8_t fixups[16] = {
[0 ... 15] = -1,
[TCG_COND_EQ] = 0,
[TCG_COND_NE] = NEED_INV,
[TCG_COND_GT] = 0,
[TCG_COND_LT] = NEED_SWAP,
[TCG_COND_LE] = NEED_INV,
[TCG_COND_GE] = NEED_SWAP | NEED_INV,
[TCG_COND_GTU] = NEED_BIAS,
[TCG_COND_LTU] = NEED_BIAS | NEED_SWAP,
[TCG_COND_LEU] = NEED_BIAS | NEED_INV,
[TCG_COND_GEU] = NEED_BIAS | NEED_SWAP | NEED_INV,
};
TCGv_vec t1, t2;
uint8_t fixup;
fixup = fixups[cond & 15];
tcg_debug_assert(fixup != 0xff);
if (fixup & NEED_INV) {
cond = tcg_invert_cond(cond);
}
if (fixup & NEED_SWAP) {
t1 = v1, v1 = v2, v2 = t1;
cond = tcg_swap_cond(cond);
}
t1 = t2 = NULL;
if (fixup & NEED_BIAS) {
t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type);
tcg_gen_dupi_vec(vece, t2, 1ull << ((8 << vece) - 1));
tcg_gen_sub_vec(vece, t1, v1, t2);
tcg_gen_sub_vec(vece, t2, v2, t2);
v1 = t1;
v2 = t2;
cond = tcg_signed_cond(cond);
}
tcg_debug_assert(cond == TCG_COND_EQ || cond == TCG_COND_GT);
/* Expand directly; do not recurse. */
vec_gen_4(INDEX_op_cmp_vec, type, vece,
tcgv_vec_arg(v0), tcgv_vec_arg(v1), tcgv_vec_arg(v2), cond);
if (t1) {
tcg_temp_free_vec(t1);
if (t2) {
tcg_temp_free_vec(t2);
}
}
if (fixup & NEED_INV) {
tcg_gen_not_vec(vece, v0, v0);
}
}
static void expand_vec_minmax(TCGType type, unsigned vece,
TCGCond cond, bool min,
TCGv_vec v0, TCGv_vec v1, TCGv_vec v2)
{
TCGv_vec t1 = tcg_temp_new_vec(type);
tcg_debug_assert(vece == MO_64);
tcg_gen_cmp_vec(cond, vece, t1, v1, v2);
if (min) {
TCGv_vec t2;
t2 = v1, v1 = v2, v2 = t2;
}
vec_gen_4(INDEX_op_x86_vpblendvb_vec, type, vece,
tcgv_vec_arg(v0), tcgv_vec_arg(v1),
tcgv_vec_arg(v2), tcgv_vec_arg(t1));
tcg_temp_free_vec(t1);
}
void tcg_expand_vec_op(TCGOpcode opc, TCGType type, unsigned vece,
TCGArg a0, ...)
{
va_list va;
TCGArg a1, a2;
TCGv_vec v0, t1, t2, t3, t4;
TCGArg a2;
TCGv_vec v0, v1, v2;
va_start(va, a0);
v0 = temp_tcgv_vec(arg_temp(a0));
v1 = temp_tcgv_vec(arg_temp(va_arg(va, TCGArg)));
a2 = va_arg(va, TCGArg);
switch (opc) {
case INDEX_op_shli_vec:
case INDEX_op_shri_vec:
tcg_debug_assert(vece == MO_8);
a1 = va_arg(va, TCGArg);
a2 = va_arg(va, TCGArg);
/* Unpack to W, shift, and repack. Tricky bits:
(1) Use punpck*bw x,x to produce DDCCBBAA,
i.e. duplicate in other half of the 16-bit lane.
(2) For right-shift, add 8 so that the high half of
the lane becomes zero. For left-shift, we must
shift up and down again.
(3) Step 2 leaves high half zero such that PACKUSWB
(pack with unsigned saturation) does not modify
the quantity. */
t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type);
vec_gen_3(INDEX_op_x86_punpckl_vec, type, MO_8,
tcgv_vec_arg(t1), a1, a1);
vec_gen_3(INDEX_op_x86_punpckh_vec, type, MO_8,
tcgv_vec_arg(t2), a1, a1);
if (opc == INDEX_op_shri_vec) {
vec_gen_3(INDEX_op_shri_vec, type, MO_16,
tcgv_vec_arg(t1), tcgv_vec_arg(t1), a2 + 8);
vec_gen_3(INDEX_op_shri_vec, type, MO_16,
tcgv_vec_arg(t2), tcgv_vec_arg(t2), a2 + 8);
} else {
vec_gen_3(INDEX_op_shli_vec, type, MO_16,
tcgv_vec_arg(t1), tcgv_vec_arg(t1), a2 + 8);
vec_gen_3(INDEX_op_shli_vec, type, MO_16,
tcgv_vec_arg(t2), tcgv_vec_arg(t2), a2 + 8);
vec_gen_3(INDEX_op_shri_vec, type, MO_16,
tcgv_vec_arg(t1), tcgv_vec_arg(t1), 8);
vec_gen_3(INDEX_op_shri_vec, type, MO_16,
tcgv_vec_arg(t2), tcgv_vec_arg(t2), 8);
}
vec_gen_3(INDEX_op_x86_packus_vec, type, MO_8,
a0, tcgv_vec_arg(t1), tcgv_vec_arg(t2));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
expand_vec_shi(type, vece, opc == INDEX_op_shri_vec, v0, v1, a2);
break;
case INDEX_op_sari_vec:
a1 = va_arg(va, TCGArg);
a2 = va_arg(va, TCGArg);
if (vece == MO_8) {
/* Unpack to W, shift, and repack, as above. */
t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type);
vec_gen_3(INDEX_op_x86_punpckl_vec, type, MO_8,
tcgv_vec_arg(t1), a1, a1);
vec_gen_3(INDEX_op_x86_punpckh_vec, type, MO_8,
tcgv_vec_arg(t2), a1, a1);
vec_gen_3(INDEX_op_sari_vec, type, MO_16,
tcgv_vec_arg(t1), tcgv_vec_arg(t1), a2 + 8);
vec_gen_3(INDEX_op_sari_vec, type, MO_16,
tcgv_vec_arg(t2), tcgv_vec_arg(t2), a2 + 8);
vec_gen_3(INDEX_op_x86_packss_vec, type, MO_8,
a0, tcgv_vec_arg(t1), tcgv_vec_arg(t2));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
break;
}
tcg_debug_assert(vece == MO_64);
/* MO_64: If the shift is <= 32, we can emulate the sign extend by
performing an arithmetic 32-bit shift and overwriting the high
half of the result (note that the ISA says shift of 32 is valid). */
if (a2 <= 32) {
t1 = tcg_temp_new_vec(type);
vec_gen_3(INDEX_op_sari_vec, type, MO_32, tcgv_vec_arg(t1), a1, a2);
vec_gen_3(INDEX_op_shri_vec, type, MO_64, a0, a1, a2);
vec_gen_4(INDEX_op_x86_blend_vec, type, MO_32,
a0, a0, tcgv_vec_arg(t1), 0xaa);
tcg_temp_free_vec(t1);
break;
}
/* Otherwise we will need to use a compare vs 0 to produce the
sign-extend, shift and merge. */
t1 = tcg_temp_new_vec(type);
t2 = tcg_const_zeros_vec(type);
vec_gen_4(INDEX_op_cmp_vec, type, MO_64,
tcgv_vec_arg(t1), tcgv_vec_arg(t2), a1, TCG_COND_GT);
tcg_temp_free_vec(t2);
vec_gen_3(INDEX_op_shri_vec, type, MO_64, a0, a1, a2);
vec_gen_3(INDEX_op_shli_vec, type, MO_64,
tcgv_vec_arg(t1), tcgv_vec_arg(t1), 64 - a2);
vec_gen_3(INDEX_op_or_vec, type, MO_64, a0, a0, tcgv_vec_arg(t1));
tcg_temp_free_vec(t1);
expand_vec_sari(type, vece, v0, v1, a2);
break;
case INDEX_op_mul_vec:
tcg_debug_assert(vece == MO_8);
a1 = va_arg(va, TCGArg);
a2 = va_arg(va, TCGArg);
switch (type) {
case TCG_TYPE_V64:
t1 = tcg_temp_new_vec(TCG_TYPE_V128);
t2 = tcg_temp_new_vec(TCG_TYPE_V128);
tcg_gen_dup16i_vec(t2, 0);
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t1), a1, tcgv_vec_arg(t2));
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t2), tcgv_vec_arg(t2), a2);
tcg_gen_mul_vec(MO_16, t1, t1, t2);
tcg_gen_shri_vec(MO_16, t1, t1, 8);
vec_gen_3(INDEX_op_x86_packus_vec, TCG_TYPE_V128, MO_8,
a0, tcgv_vec_arg(t1), tcgv_vec_arg(t1));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
break;
case TCG_TYPE_V128:
t1 = tcg_temp_new_vec(TCG_TYPE_V128);
t2 = tcg_temp_new_vec(TCG_TYPE_V128);
t3 = tcg_temp_new_vec(TCG_TYPE_V128);
t4 = tcg_temp_new_vec(TCG_TYPE_V128);
tcg_gen_dup16i_vec(t4, 0);
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t1), a1, tcgv_vec_arg(t4));
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t2), tcgv_vec_arg(t4), a2);
vec_gen_3(INDEX_op_x86_punpckh_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t3), a1, tcgv_vec_arg(t4));
vec_gen_3(INDEX_op_x86_punpckh_vec, TCG_TYPE_V128, MO_8,
tcgv_vec_arg(t4), tcgv_vec_arg(t4), a2);
tcg_gen_mul_vec(MO_16, t1, t1, t2);
tcg_gen_mul_vec(MO_16, t3, t3, t4);
tcg_gen_shri_vec(MO_16, t1, t1, 8);
tcg_gen_shri_vec(MO_16, t3, t3, 8);
vec_gen_3(INDEX_op_x86_packus_vec, TCG_TYPE_V128, MO_8,
a0, tcgv_vec_arg(t1), tcgv_vec_arg(t3));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
tcg_temp_free_vec(t3);
tcg_temp_free_vec(t4);
break;
case TCG_TYPE_V256:
t1 = tcg_temp_new_vec(TCG_TYPE_V256);
t2 = tcg_temp_new_vec(TCG_TYPE_V256);
t3 = tcg_temp_new_vec(TCG_TYPE_V256);
t4 = tcg_temp_new_vec(TCG_TYPE_V256);
tcg_gen_dup16i_vec(t4, 0);
/* a1: A[0-7] ... D[0-7]; a2: W[0-7] ... Z[0-7]
t1: extends of B[0-7], D[0-7]
t2: extends of X[0-7], Z[0-7]
t3: extends of A[0-7], C[0-7]
t4: extends of W[0-7], Y[0-7]. */
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V256, MO_8,
tcgv_vec_arg(t1), a1, tcgv_vec_arg(t4));
vec_gen_3(INDEX_op_x86_punpckl_vec, TCG_TYPE_V256, MO_8,
tcgv_vec_arg(t2), tcgv_vec_arg(t4), a2);
vec_gen_3(INDEX_op_x86_punpckh_vec, TCG_TYPE_V256, MO_8,
tcgv_vec_arg(t3), a1, tcgv_vec_arg(t4));
vec_gen_3(INDEX_op_x86_punpckh_vec, TCG_TYPE_V256, MO_8,
tcgv_vec_arg(t4), tcgv_vec_arg(t4), a2);
/* t1: BX DZ; t2: AW CY. */
tcg_gen_mul_vec(MO_16, t1, t1, t2);
tcg_gen_mul_vec(MO_16, t3, t3, t4);
tcg_gen_shri_vec(MO_16, t1, t1, 8);
tcg_gen_shri_vec(MO_16, t3, t3, 8);
/* a0: AW BX CY DZ. */
vec_gen_3(INDEX_op_x86_packus_vec, TCG_TYPE_V256, MO_8,
a0, tcgv_vec_arg(t1), tcgv_vec_arg(t3));
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
tcg_temp_free_vec(t3);
tcg_temp_free_vec(t4);
break;
default:
g_assert_not_reached();
}
v2 = temp_tcgv_vec(arg_temp(a2));
expand_vec_mul(type, vece, v0, v1, v2);
break;
case INDEX_op_cmp_vec:
{
enum {
NEED_SWAP = 1,
NEED_INV = 2,
NEED_BIAS = 4
};
static const uint8_t fixups[16] = {
[0 ... 15] = -1,
[TCG_COND_EQ] = 0,
[TCG_COND_NE] = NEED_INV,
[TCG_COND_GT] = 0,
[TCG_COND_LT] = NEED_SWAP,
[TCG_COND_LE] = NEED_INV,
[TCG_COND_GE] = NEED_SWAP | NEED_INV,
[TCG_COND_GTU] = NEED_BIAS,
[TCG_COND_LTU] = NEED_BIAS | NEED_SWAP,
[TCG_COND_LEU] = NEED_BIAS | NEED_INV,
[TCG_COND_GEU] = NEED_BIAS | NEED_SWAP | NEED_INV,
};
v2 = temp_tcgv_vec(arg_temp(a2));
expand_vec_cmp(type, vece, v0, v1, v2, va_arg(va, TCGArg));
break;
TCGCond cond;
uint8_t fixup;
a1 = va_arg(va, TCGArg);
a2 = va_arg(va, TCGArg);
cond = va_arg(va, TCGArg);
fixup = fixups[cond & 15];
tcg_debug_assert(fixup != 0xff);
if (fixup & NEED_INV) {
cond = tcg_invert_cond(cond);
}
if (fixup & NEED_SWAP) {
TCGArg t;
t = a1, a1 = a2, a2 = t;
cond = tcg_swap_cond(cond);
}
t1 = t2 = NULL;
if (fixup & NEED_BIAS) {
t1 = tcg_temp_new_vec(type);
t2 = tcg_temp_new_vec(type);
tcg_gen_dupi_vec(vece, t2, 1ull << ((8 << vece) - 1));
tcg_gen_sub_vec(vece, t1, temp_tcgv_vec(arg_temp(a1)), t2);
tcg_gen_sub_vec(vece, t2, temp_tcgv_vec(arg_temp(a2)), t2);
a1 = tcgv_vec_arg(t1);
a2 = tcgv_vec_arg(t2);
cond = tcg_signed_cond(cond);
}
tcg_debug_assert(cond == TCG_COND_EQ || cond == TCG_COND_GT);
vec_gen_4(INDEX_op_cmp_vec, type, vece, a0, a1, a2, cond);
if (fixup & NEED_BIAS) {
tcg_temp_free_vec(t1);
tcg_temp_free_vec(t2);
}
if (fixup & NEED_INV) {
tcg_gen_not_vec(vece, v0, v0);
}
}
case INDEX_op_smin_vec:
v2 = temp_tcgv_vec(arg_temp(a2));
expand_vec_minmax(type, vece, TCG_COND_GT, true, v0, v1, v2);
break;
case INDEX_op_smax_vec:
v2 = temp_tcgv_vec(arg_temp(a2));
expand_vec_minmax(type, vece, TCG_COND_GT, false, v0, v1, v2);
break;
case INDEX_op_umin_vec:
v2 = temp_tcgv_vec(arg_temp(a2));
expand_vec_minmax(type, vece, TCG_COND_GTU, true, v0, v1, v2);
break;
case INDEX_op_umax_vec:
v2 = temp_tcgv_vec(arg_temp(a2));
expand_vec_minmax(type, vece, TCG_COND_GTU, false, v0, v1, v2);
break;
default:

97
tcg/mips/tcg-target.inc.c
View File

@ -1201,8 +1201,19 @@ static int tcg_out_call_iarg_reg2(TCGContext *s, int i, TCGReg al, TCGReg ah)
return i;
}
/* Perform the tlb comparison operation. The complete host address is
placed in BASE. Clobbers TMP0, TMP1, TMP2, A0. */
/* We expect tlb_mask to be before tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
offsetof(CPUArchState, tlb_mask));
/* We expect tlb_mask to be "near" tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) -
offsetof(CPUArchState, tlb_mask) >= 0x8000);
/*
* Perform the tlb comparison operation.
* The complete host address is placed in BASE.
* Clobbers TMP0, TMP1, TMP2, TMP3.
*/
static void tcg_out_tlb_load(TCGContext *s, TCGReg base, TCGReg addrl,
TCGReg addrh, TCGMemOpIdx oi,
tcg_insn_unit *label_ptr[2], bool is_load)
@ -1210,52 +1221,73 @@ static void tcg_out_tlb_load(TCGContext *s, TCGReg base, TCGReg addrl,
TCGMemOp opc = get_memop(oi);
unsigned s_bits = opc & MO_SIZE;
unsigned a_bits = get_alignment_bits(opc);
target_ulong mask;
int mem_index = get_mmuidx(oi);
int cmp_off
= (is_load
? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
: offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
int mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
int table_off = offsetof(CPUArchState, tlb_table[mem_index]);
int add_off = offsetof(CPUTLBEntry, addend);
int cmp_off = (is_load ? offsetof(CPUTLBEntry, addr_read)
: offsetof(CPUTLBEntry, addr_write));
TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0;
target_ulong mask;
tcg_out_opc_sa(s, ALIAS_TSRL, TCG_REG_A0, addrl,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_A0, TCG_REG_A0,
(CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS);
tcg_out_opc_reg(s, ALIAS_PADD, TCG_REG_A0, TCG_REG_A0, TCG_AREG0);
if (table_off > 0x7fff) {
int mask_hi = mask_off - (int16_t)mask_off;
int table_hi = table_off - (int16_t)table_off;
/* Compensate for very large offsets. */
while (add_off >= 0x8000) {
/* Most target env are smaller than 32k, but a few are larger than 64k,
* so handle an arbitrarily large offset.
*/
tcg_out_opc_imm(s, ALIAS_PADDI, TCG_REG_A0, TCG_REG_A0, 0x7ff0);
cmp_off -= 0x7ff0;
add_off -= 0x7ff0;
table_base = TCG_TMP1;
if (likely(mask_hi == table_hi)) {
mask_base = table_base;
tcg_out_opc_imm(s, OPC_LUI, mask_base, TCG_REG_ZERO, mask_hi >> 16);
tcg_out_opc_reg(s, ALIAS_PADD, mask_base, mask_base, TCG_AREG0);
mask_off -= mask_hi;
table_off -= mask_hi;
} else {
if (mask_hi != 0) {
mask_base = TCG_TMP0;
tcg_out_opc_imm(s, OPC_LUI,
mask_base, TCG_REG_ZERO, mask_hi >> 16);
tcg_out_opc_reg(s, ALIAS_PADD,
mask_base, mask_base, TCG_AREG0);
}
table_off -= mask_off;
mask_off -= mask_hi;
tcg_out_opc_imm(s, ALIAS_PADDI, table_base, mask_base, mask_off);
}
}
/* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP0, mask_base, mask_off);
tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP1, table_base, table_off);
/* Extract the TLB index from the address into TMP3. */
tcg_out_opc_sa(s, ALIAS_TSRL, TCG_TMP3, addrl,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tcg_out_opc_reg(s, OPC_AND, TCG_TMP3, TCG_TMP3, TCG_TMP0);
/* Add the tlb_table pointer, creating the CPUTLBEntry address in TMP3. */
tcg_out_opc_reg(s, ALIAS_PADD, TCG_TMP3, TCG_TMP3, TCG_TMP1);
/* We don't currently support unaligned accesses.
We could do so with mips32r6. */
if (a_bits < s_bits) {
a_bits = s_bits;
}
/* Mask the page bits, keeping the alignment bits to compare against. */
mask = (target_ulong)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
/* Load the (low half) tlb comparator. Mask the page bits, keeping the
alignment bits to compare against. */
/* Load the (low-half) tlb comparator. */
if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
tcg_out_ld(s, TCG_TYPE_I32, TCG_TMP0, TCG_REG_A0, cmp_off + LO_OFF);
tcg_out_ld(s, TCG_TYPE_I32, TCG_TMP0, TCG_TMP3, cmp_off + LO_OFF);
tcg_out_movi(s, TCG_TYPE_I32, TCG_TMP1, mask);
} else {
tcg_out_ldst(s,
(TARGET_LONG_BITS == 64 ? OPC_LD
: TCG_TARGET_REG_BITS == 64 ? OPC_LWU : OPC_LW),
TCG_TMP0, TCG_REG_A0, cmp_off);
tcg_out_ldst(s, (TARGET_LONG_BITS == 64 ? OPC_LD
: TCG_TARGET_REG_BITS == 64 ? OPC_LWU : OPC_LW),
TCG_TMP0, TCG_TMP3, cmp_off);
tcg_out_movi(s, TCG_TYPE_TL, TCG_TMP1, mask);
/* No second compare is required here;
load the tlb addend for the fast path. */
tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP2, TCG_REG_A0, add_off);
tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP2, TCG_TMP3, add_off);
}
tcg_out_opc_reg(s, OPC_AND, TCG_TMP1, TCG_TMP1, addrl);
@ -1271,10 +1303,10 @@ static void tcg_out_tlb_load(TCGContext *s, TCGReg base, TCGReg addrl,
/* Load and test the high half tlb comparator. */
if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
/* delay slot */
tcg_out_ld(s, TCG_TYPE_I32, TCG_TMP0, TCG_REG_A0, cmp_off + HI_OFF);
tcg_out_ld(s, TCG_TYPE_I32, TCG_TMP0, TCG_TMP3, cmp_off + HI_OFF);
/* Load the tlb addend for the fast path. */
tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP2, TCG_REG_A0, add_off);
tcg_out_ld(s, TCG_TYPE_PTR, TCG_TMP2, TCG_TMP3, add_off);
label_ptr[1] = s->code_ptr;
tcg_out_opc_br(s, OPC_BNE, addrh, TCG_TMP0);
@ -1343,8 +1375,9 @@ static void tcg_out_qemu_ld_slow_path(TCGContext *s, TCGLabelQemuLdst *l)
}
}
reloc_pc16(s->code_ptr, l->raddr);
tcg_out_opc_br(s, OPC_BEQ, TCG_REG_ZERO, TCG_REG_ZERO);
reloc_pc16(s->code_ptr - 1, l->raddr);
/* delay slot */
if (TCG_TARGET_REG_BITS == 64 && l->type == TCG_TYPE_I32) {
/* we always sign-extend 32-bit loads */

91
tcg/ppc/tcg-target.inc.c
View File

@ -327,6 +327,7 @@ static int tcg_target_const_match(tcg_target_long val, TCGType type,
#define LHZ OPCD( 40)
#define LHA OPCD( 42)
#define LWZ OPCD( 32)
#define LWZUX XO31( 55)
#define STB OPCD( 38)
#define STH OPCD( 44)
#define STW OPCD( 36)
@ -338,6 +339,7 @@ static int tcg_target_const_match(tcg_target_long val, TCGType type,
#define LD XO58( 0)
#define LDX XO31( 21)
#define LDU XO58( 1)
#define LDUX XO31( 53)
#define LWA XO58( 2)
#define LWAX XO31(341)
@ -1503,6 +1505,10 @@ static void * const qemu_st_helpers[16] = {
[MO_BEQ] = helper_be_stq_mmu,
};
/* We expect tlb_mask to be before tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
offsetof(CPUArchState, tlb_mask));
/* Perform the TLB load and compare. Places the result of the comparison
in CR7, loads the addend of the TLB into R3, and returns the register
containing the guest address (zero-extended into R4). Clobbers R0 and R2. */
@ -1513,61 +1519,63 @@ static TCGReg tcg_out_tlb_read(TCGContext *s, TCGMemOp opc,
{
int cmp_off
= (is_read
? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
: offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
TCGReg base = TCG_AREG0;
? offsetof(CPUTLBEntry, addr_read)
: offsetof(CPUTLBEntry, addr_write));
int mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
int table_off = offsetof(CPUArchState, tlb_table[mem_index]);
TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0;
unsigned s_bits = opc & MO_SIZE;
unsigned a_bits = get_alignment_bits(opc);
/* Extract the page index, shifted into place for tlb index. */
if (TCG_TARGET_REG_BITS == 64) {
if (TARGET_LONG_BITS == 32) {
/* Zero-extend the address into a place helpful for further use. */
tcg_out_ext32u(s, TCG_REG_R4, addrlo);
addrlo = TCG_REG_R4;
} else {
tcg_out_rld(s, RLDICL, TCG_REG_R3, addrlo,
64 - TARGET_PAGE_BITS, 64 - CPU_TLB_BITS);
if (table_off > 0x7fff) {
int mask_hi = mask_off - (int16_t)mask_off;
int table_hi = table_off - (int16_t)table_off;
table_base = TCG_REG_R4;
if (mask_hi == table_hi) {
mask_base = table_base;
} else if (mask_hi) {
mask_base = TCG_REG_R3;
tcg_out32(s, ADDIS | TAI(mask_base, TCG_AREG0, mask_hi >> 16));
}
tcg_out32(s, ADDIS | TAI(table_base, TCG_AREG0, table_hi >> 16));
mask_off -= mask_hi;
table_off -= table_hi;
}
/* Compensate for very large offsets. */
if (add_off >= 0x8000) {
int low = (int16_t)cmp_off;
int high = cmp_off - low;
assert((high & 0xffff) == 0);
assert(cmp_off - high == (int16_t)(cmp_off - high));
assert(add_off - high == (int16_t)(add_off - high));
tcg_out32(s, ADDIS | TAI(TCG_REG_TMP1, base, high >> 16));
base = TCG_REG_TMP1;
cmp_off -= high;
add_off -= high;
}
/* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, mask_base, mask_off);
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R4, table_base, table_off);
/* Extraction and shifting, part 2. */
if (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32) {
tcg_out_rlw(s, RLWINM, TCG_REG_R3, addrlo,
32 - (TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS),
32 - (CPU_TLB_BITS + CPU_TLB_ENTRY_BITS),
31 - CPU_TLB_ENTRY_BITS);
/* Extract the page index, shifted into place for tlb index. */
if (TCG_TARGET_REG_BITS == 32) {
tcg_out_shri32(s, TCG_REG_TMP1, addrlo,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
} else {
tcg_out_shli64(s, TCG_REG_R3, TCG_REG_R3, CPU_TLB_ENTRY_BITS);
tcg_out_shri64(s, TCG_REG_TMP1, addrlo,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
}
tcg_out32(s, AND | SAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_TMP1));
tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, base));
/* Load the tlb comparator. */
if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
/* Load the TLB comparator. */
if (cmp_off == 0 && TCG_TARGET_REG_BITS >= TARGET_LONG_BITS) {
uint32_t lxu = (TCG_TARGET_REG_BITS == 32 || TARGET_LONG_BITS == 32
? LWZUX : LDUX);
tcg_out32(s, lxu | TAB(TCG_REG_TMP1, TCG_REG_R3, TCG_REG_R4));
} else {
tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
tcg_out32(s, ADD | TAB(TCG_REG_R3, TCG_REG_R3, TCG_REG_R4));
if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_TMP1, TCG_REG_R3, cmp_off + 4);
tcg_out_ld(s, TCG_TYPE_I32, TCG_REG_R4, TCG_REG_R3, cmp_off);
} else {
tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP1, TCG_REG_R3, cmp_off);
}
}
/* Load the TLB addend for use on the fast path. Do this asap
to minimize any load use delay. */
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3, add_off);
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_R3, TCG_REG_R3,
offsetof(CPUTLBEntry, addend));
/* Clear the non-page, non-alignment bits from the address */
if (TCG_TARGET_REG_BITS == 32) {
@ -1600,6 +1608,9 @@ static TCGReg tcg_out_tlb_read(TCGContext *s, TCGMemOp opc,
if (TARGET_LONG_BITS == 32) {
tcg_out_rlw(s, RLWINM, TCG_REG_R0, t, 0,
(32 - a_bits) & 31, 31 - TARGET_PAGE_BITS);
/* Zero-extend the address for use in the final address. */
tcg_out_ext32u(s, TCG_REG_R4, addrlo);
addrlo = TCG_REG_R4;
} else if (a_bits == 0) {
tcg_out_rld(s, RLDICR, TCG_REG_R0, t, 0, 63 - TARGET_PAGE_BITS);
} else {

126
tcg/riscv/tcg-target.inc.c
View File

@ -958,6 +958,17 @@ static void * const qemu_st_helpers[16] = {
[MO_BEQ] = helper_be_stq_mmu,
};
/* We don't support oversize guests */
QEMU_BUILD_BUG_ON(TCG_TARGET_REG_BITS < TARGET_LONG_BITS);
/* We expect tlb_mask to be before tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
offsetof(CPUArchState, tlb_mask));
/* We expect tlb_mask to be "near" tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) -
offsetof(CPUArchState, tlb_mask) >= 0x800);
static void tcg_out_tlb_load(TCGContext *s, TCGReg addrl,
TCGReg addrh, TCGMemOpIdx oi,
tcg_insn_unit **label_ptr, bool is_load)
@ -965,94 +976,67 @@ static void tcg_out_tlb_load(TCGContext *s, TCGReg addrl,
TCGMemOp opc = get_memop(oi);
unsigned s_bits = opc & MO_SIZE;
unsigned a_bits = get_alignment_bits(opc);
target_ulong mask;
tcg_target_long compare_mask;
int mem_index = get_mmuidx(oi);
int cmp_off
= (is_load
? offsetof(CPUArchState, tlb_table[mem_index][0].addr_read)
: offsetof(CPUArchState, tlb_table[mem_index][0].addr_write));
int add_off = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
RISCVInsn load_cmp_op = (TARGET_LONG_BITS == 64 ? OPC_LD :
TCG_TARGET_REG_BITS == 64 ? OPC_LWU : OPC_LW);
RISCVInsn load_add_op = TCG_TARGET_REG_BITS == 64 ? OPC_LD : OPC_LW;
TCGReg base = TCG_AREG0;
int mask_off, table_off;
TCGReg mask_base = TCG_AREG0, table_base = TCG_AREG0;
/* We don't support oversize guests */
if (TCG_TARGET_REG_BITS < TARGET_LONG_BITS) {
g_assert_not_reached();
mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
table_off = offsetof(CPUArchState, tlb_table[mem_index]);
if (table_off > 0x7ff) {
int mask_hi = mask_off - sextreg(mask_off, 0, 12);
int table_hi = table_off - sextreg(table_off, 0, 12);
if (likely(mask_hi == table_hi)) {
mask_base = table_base = TCG_REG_TMP1;
tcg_out_opc_upper(s, OPC_LUI, mask_base, mask_hi);
tcg_out_opc_reg(s, OPC_ADD, mask_base, mask_base, TCG_AREG0);
mask_off -= mask_hi;
table_off -= mask_hi;
} else {
mask_base = TCG_REG_TMP0;
table_base = TCG_REG_TMP1;
tcg_out_opc_upper(s, OPC_LUI, mask_base, mask_hi);
tcg_out_opc_reg(s, OPC_ADD, mask_base, mask_base, TCG_AREG0);
table_off -= mask_off;
mask_off -= mask_hi;
tcg_out_opc_imm(s, OPC_ADDI, table_base, mask_base, mask_off);
}
}
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP0, mask_base, mask_off);
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP1, table_base, table_off);
tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP2, addrl,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP0);
tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, TCG_REG_TMP2, TCG_REG_TMP1);
/* Load the tlb comparator and the addend. */
tcg_out_ld(s, TCG_TYPE_TL, TCG_REG_TMP0, TCG_REG_TMP2,
is_load ? offsetof(CPUTLBEntry, addr_read)
: offsetof(CPUTLBEntry, addr_write));
tcg_out_ld(s, TCG_TYPE_PTR, TCG_REG_TMP2, TCG_REG_TMP2,
offsetof(CPUTLBEntry, addend));
/* We don't support unaligned accesses. */
if (a_bits < s_bits) {
a_bits = s_bits;
}
mask = (target_ulong)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
/* Compensate for very large offsets. */
if (add_off >= 0x1000) {
int adj;
base = TCG_REG_TMP2;
if (cmp_off <= 2 * 0xfff) {
adj = 0xfff;
tcg_out_opc_imm(s, OPC_ADDI, base, TCG_AREG0, adj);
} else {
adj = cmp_off - sextreg(cmp_off, 0, 12);
tcg_debug_assert(add_off - adj >= -0x1000
&& add_off - adj < 0x1000);
tcg_out_opc_upper(s, OPC_LUI, base, adj);
tcg_out_opc_reg(s, OPC_ADD, base, base, TCG_AREG0);
}
add_off -= adj;
cmp_off -= adj;
}
/* Extract the page index. */
if (CPU_TLB_BITS + CPU_TLB_ENTRY_BITS < 12) {
tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP0, addrl,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP0, TCG_REG_TMP0,
MAKE_64BIT_MASK(CPU_TLB_ENTRY_BITS, CPU_TLB_BITS));
} else if (TARGET_PAGE_BITS >= 12) {
tcg_out_opc_upper(s, OPC_LUI, TCG_REG_TMP0,
MAKE_64BIT_MASK(TARGET_PAGE_BITS, CPU_TLB_BITS));
tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP0, TCG_REG_TMP0, addrl);
tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP0, TCG_REG_TMP0,
CPU_TLB_BITS - CPU_TLB_ENTRY_BITS);
} else {
tcg_out_opc_imm(s, OPC_SRLI, TCG_REG_TMP0, addrl, TARGET_PAGE_BITS);
tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP0, TCG_REG_TMP0,
MAKE_64BIT_MASK(0, CPU_TLB_BITS));
tcg_out_opc_imm(s, OPC_SLLI, TCG_REG_TMP0, TCG_REG_TMP0,
CPU_TLB_ENTRY_BITS);
}
/* Add that to the base address to index the tlb. */
tcg_out_opc_reg(s, OPC_ADD, TCG_REG_TMP2, base, TCG_REG_TMP0);
base = TCG_REG_TMP2;
/* Load the tlb comparator and the addend. */
tcg_out_ldst(s, load_cmp_op, TCG_REG_TMP0, base, cmp_off);
tcg_out_ldst(s, load_add_op, TCG_REG_TMP2, base, add_off);
/* Clear the non-page, non-alignment bits from the address. */
if (mask == sextreg(mask, 0, 12)) {
tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addrl, mask);
compare_mask = (tcg_target_long)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
if (compare_mask == sextreg(compare_mask, 0, 12)) {
tcg_out_opc_imm(s, OPC_ANDI, TCG_REG_TMP1, addrl, compare_mask);
} else {
tcg_out_movi(s, TCG_TYPE_REG, TCG_REG_TMP1, mask);
tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_TMP1, compare_mask);
tcg_out_opc_reg(s, OPC_AND, TCG_REG_TMP1, TCG_REG_TMP1, addrl);
}
}
/* Compare masked address with the TLB entry. */
label_ptr[0] = s->code_ptr;
tcg_out_opc_branch(s, OPC_BNE, TCG_REG_TMP0, TCG_REG_TMP1, 0);
/* NOP to allow patching later */
tcg_out_opc_imm(s, OPC_ADDI, TCG_REG_ZERO, TCG_REG_ZERO, 0);
/* TODO: Move this out of line
* see:
* https://lists.nongnu.org/archive/html/qemu-devel/2018-11/msg02234.html
*/
/* TLB Hit - translate address using addend. */
if (TCG_TARGET_REG_BITS > TARGET_LONG_BITS) {

45
tcg/s390/tcg-target.inc.c
View File

@ -1537,10 +1537,10 @@ static void tcg_out_qemu_st_direct(TCGContext *s, TCGMemOp opc, TCGReg data,
#if defined(CONFIG_SOFTMMU)
#include "tcg-ldst.inc.c"
/* We're expecting to use a 20-bit signed offset on the tlb memory ops.
Using the offset of the second entry in the last tlb table ensures
that we can index all of the elements of the first entry. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table[NB_MMU_MODES - 1][1])
/* We're expecting to use a 20-bit signed offset on the tlb memory ops. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_mask[NB_MMU_MODES - 1])
> 0x7ffff);
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table[NB_MMU_MODES - 1])
> 0x7ffff);
/* Load and compare a TLB entry, leaving the flags set. Loads the TLB
@ -1552,48 +1552,41 @@ static TCGReg tcg_out_tlb_read(TCGContext* s, TCGReg addr_reg, TCGMemOp opc,
unsigned a_bits = get_alignment_bits(opc);
unsigned s_mask = (1 << s_bits) - 1;
unsigned a_mask = (1 << a_bits) - 1;
int mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
int table_off = offsetof(CPUArchState, tlb_table[mem_index]);
int ofs, a_off;
uint64_t tlb_mask;
tcg_out_sh64(s, RSY_SRLG, TCG_REG_R2, addr_reg, TCG_REG_NONE,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tcg_out_insn(s, RXY, NG, TCG_REG_R2, TCG_AREG0, TCG_REG_NONE, mask_off);
tcg_out_insn(s, RXY, AG, TCG_REG_R2, TCG_AREG0, TCG_REG_NONE, table_off);
/* For aligned accesses, we check the first byte and include the alignment
bits within the address. For unaligned access, we check that we don't
cross pages using the address of the last byte of the access. */
a_off = (a_bits >= s_bits ? 0 : s_mask - a_mask);
tlb_mask = (uint64_t)TARGET_PAGE_MASK | a_mask;
if (s390_facilities & FACILITY_GEN_INST_EXT) {
tcg_out_risbg(s, TCG_REG_R2, addr_reg,
64 - CPU_TLB_BITS - CPU_TLB_ENTRY_BITS,
63 - CPU_TLB_ENTRY_BITS,
64 + CPU_TLB_ENTRY_BITS - TARGET_PAGE_BITS, 1);
if (a_off) {
tcg_out_insn(s, RX, LA, TCG_REG_R3, addr_reg, TCG_REG_NONE, a_off);
tgen_andi(s, TCG_TYPE_TL, TCG_REG_R3, tlb_mask);
} else {
tgen_andi_risbg(s, TCG_REG_R3, addr_reg, tlb_mask);
}
if ((s390_facilities & FACILITY_GEN_INST_EXT) && a_off == 0) {
tgen_andi_risbg(s, TCG_REG_R3, addr_reg, tlb_mask);
} else {
tcg_out_sh64(s, RSY_SRLG, TCG_REG_R2, addr_reg, TCG_REG_NONE,
TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS);
tcg_out_insn(s, RX, LA, TCG_REG_R3, addr_reg, TCG_REG_NONE, a_off);
tgen_andi(s, TCG_TYPE_I64, TCG_REG_R2,
(CPU_TLB_SIZE - 1) << CPU_TLB_ENTRY_BITS);
tgen_andi(s, TCG_TYPE_TL, TCG_REG_R3, tlb_mask);
}
if (is_ld) {
ofs = offsetof(CPUArchState, tlb_table[mem_index][0].addr_read);
ofs = offsetof(CPUTLBEntry, addr_read);
} else {
ofs = offsetof(CPUArchState, tlb_table[mem_index][0].addr_write);
ofs = offsetof(CPUTLBEntry, addr_write);
}
if (TARGET_LONG_BITS == 32) {
tcg_out_mem(s, RX_C, RXY_CY, TCG_REG_R3, TCG_REG_R2, TCG_AREG0, ofs);
tcg_out_insn(s, RX, C, TCG_REG_R3, TCG_REG_R2, TCG_REG_NONE, ofs);
} else {
tcg_out_mem(s, 0, RXY_CG, TCG_REG_R3, TCG_REG_R2, TCG_AREG0, ofs);
tcg_out_insn(s, RXY, CG, TCG_REG_R3, TCG_REG_R2, TCG_REG_NONE, ofs);
}
ofs = offsetof(CPUArchState, tlb_table[mem_index][0].addend);
tcg_out_mem(s, 0, RXY_LG, TCG_REG_R2, TCG_REG_R2, TCG_AREG0, ofs);
tcg_out_insn(s, RXY, LG, TCG_REG_R2, TCG_REG_R2, TCG_REG_NONE,
offsetof(CPUTLBEntry, addend));
if (TARGET_LONG_BITS == 32) {
tgen_ext32u(s, TCG_REG_R3, addr_reg);

82
tcg/sparc/tcg-target.inc.c
View File

@ -1074,54 +1074,72 @@ static void tcg_out_nop_fill(tcg_insn_unit *p, int count)
The result of the TLB comparison is in %[ix]cc. The sanitized address
is in the returned register, maybe %o0. The TLB addend is in %o1. */
/* We expect tlb_mask to be before tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) <
offsetof(CPUArchState, tlb_mask));
/* We expect tlb_mask to be "near" tlb_table. */
QEMU_BUILD_BUG_ON(offsetof(CPUArchState, tlb_table) -
offsetof(CPUArchState, tlb_mask) >= (1 << 13));
static TCGReg tcg_out_tlb_load(TCGContext *s, TCGReg addr, int mem_index,
TCGMemOp opc, int which)
{
int mask_off = offsetof(CPUArchState, tlb_mask[mem_index]);
int table_off = offsetof(CPUArchState, tlb_table[mem_index]);
TCGReg base = TCG_AREG0;
const TCGReg r0 = TCG_REG_O0;
const TCGReg r1 = TCG_REG_O1;
const TCGReg r2 = TCG_REG_O2;
unsigned s_bits = opc & MO_SIZE;
unsigned a_bits = get_alignment_bits(opc);
int tlb_ofs;
tcg_target_long compare_mask;
/* Shift the page number down. */
tcg_out_arithi(s, r1, addr, TARGET_PAGE_BITS, SHIFT_SRL);
if (!check_fit_i32(table_off, 13)) {
int table_hi;
base = r1;
if (table_off <= 2 * 0xfff) {
table_hi = 0xfff;
tcg_out_arithi(s, base, TCG_AREG0, table_hi, ARITH_ADD);
} else {
table_hi = table_off & ~0x3ff;
tcg_out_sethi(s, base, table_hi);
tcg_out_arith(s, base, TCG_AREG0, base, ARITH_ADD);
}
mask_off -= table_hi;
table_off -= table_hi;
tcg_debug_assert(check_fit_i32(mask_off, 13));
}
/* Load tlb_mask[mmu_idx] and tlb_table[mmu_idx]. */
tcg_out_ld(s, TCG_TYPE_PTR, r0, base, mask_off);
tcg_out_ld(s, TCG_TYPE_PTR, r1, base, table_off);
/* Extract the page index, shifted into place for tlb index. */
tcg_out_arithi(s, r2, addr, TARGET_PAGE_BITS - CPU_TLB_ENTRY_BITS,
SHIFT_SRL);
tcg_out_arith(s, r2, r2, r0, ARITH_AND);
/* Add the tlb_table pointer, creating the CPUTLBEntry address into R2. */
tcg_out_arith(s, r2, r2, r1, ARITH_ADD);
/* Load the tlb comparator and the addend. */
tcg_out_ld(s, TCG_TYPE_TL, r0, r2, which);
tcg_out_ld(s, TCG_TYPE_PTR, r1, r2, offsetof(CPUTLBEntry, addend));
/* Mask out the page offset, except for the required alignment.
We don't support unaligned accesses. */
if (a_bits < s_bits) {
a_bits = s_bits;
}
tcg_out_movi(s, TCG_TYPE_TL, TCG_REG_T1,
TARGET_PAGE_MASK | ((1 << a_bits) - 1));
/* Mask the tlb index. */
tcg_out_arithi(s, r1, r1, CPU_TLB_SIZE - 1, ARITH_AND);
/* Mask page, part 2. */
tcg_out_arith(s, r0, addr, TCG_REG_T1, ARITH_AND);
/* Shift the tlb index into place. */
tcg_out_arithi(s, r1, r1, CPU_TLB_ENTRY_BITS, SHIFT_SLL);
/* Relative to the current ENV. */
tcg_out_arith(s, r1, TCG_AREG0, r1, ARITH_ADD);
/* Find a base address that can load both tlb comparator and addend. */
tlb_ofs = offsetof(CPUArchState, tlb_table[mem_index][0]);
if (!check_fit_ptr(tlb_ofs + sizeof(CPUTLBEntry), 13)) {
if (tlb_ofs & ~0x3ff) {
tcg_out_movi(s, TCG_TYPE_PTR, TCG_REG_T1, tlb_ofs & ~0x3ff);
tcg_out_arith(s, r1, r1, TCG_REG_T1, ARITH_ADD);
}
tlb_ofs &= 0x3ff;
compare_mask = (tcg_target_ulong)TARGET_PAGE_MASK | ((1 << a_bits) - 1);
if (check_fit_tl(compare_mask, 13)) {
tcg_out_arithi(s, r2, addr, compare_mask, ARITH_AND);
} else {
tcg_out_movi(s, TCG_TYPE_TL, r2, compare_mask);
tcg_out_arith(s, r2, addr, r2, ARITH_AND);
}
/* Load the tlb comparator and the addend. */
tcg_out_ld(s, TCG_TYPE_TL, r2, r1, tlb_ofs + which);
tcg_out_ld(s, TCG_TYPE_PTR, r1, r1, tlb_ofs+offsetof(CPUTLBEntry, addend));
/* subcc arg0, arg2, %g0 */
tcg_out_cmp(s, r0, r2, 0);
/* If the guest address must be zero-extended, do so now. */

305
tcg/tcg-op-gvec.c
View File

@ -665,7 +665,7 @@ static void expand_3_i32(uint32_t dofs, uint32_t aofs,
/* Expand OPSZ bytes worth of three-operand operations using i32 elements. */
static void expand_4_i32(uint32_t dofs, uint32_t aofs, uint32_t bofs,
uint32_t cofs, uint32_t oprsz,
uint32_t cofs, uint32_t oprsz, bool write_aofs,
void (*fni)(TCGv_i32, TCGv_i32, TCGv_i32, TCGv_i32))
{
TCGv_i32 t0 = tcg_temp_new_i32();
@ -680,6 +680,9 @@ static void expand_4_i32(uint32_t dofs, uint32_t aofs, uint32_t bofs,
tcg_gen_ld_i32(t3, cpu_env, cofs + i);
fni(t0, t1, t2, t3);
tcg_gen_st_i32(t0, cpu_env, dofs + i);
if (write_aofs) {
tcg_gen_st_i32(t1, cpu_env, aofs + i);
}
}
tcg_temp_free_i32(t3);
tcg_temp_free_i32(t2);
@ -769,7 +772,7 @@ static void expand_3_i64(uint32_t dofs, uint32_t aofs,
/* Expand OPSZ bytes worth of three-operand operations using i64 elements. */
static void expand_4_i64(uint32_t dofs, uint32_t aofs, uint32_t bofs,
uint32_t cofs, uint32_t oprsz,
uint32_t cofs, uint32_t oprsz, bool write_aofs,
void (*fni)(TCGv_i64, TCGv_i64, TCGv_i64, TCGv_i64))
{
TCGv_i64 t0 = tcg_temp_new_i64();
@ -784,6 +787,9 @@ static void expand_4_i64(uint32_t dofs, uint32_t aofs, uint32_t bofs,
tcg_gen_ld_i64(t3, cpu_env, cofs + i);
fni(t0, t1, t2, t3);
tcg_gen_st_i64(t0, cpu_env, dofs + i);
if (write_aofs) {
tcg_gen_st_i64(t1, cpu_env, aofs + i);
}
}
tcg_temp_free_i64(t3);
tcg_temp_free_i64(t2);
@ -880,7 +886,7 @@ static void expand_3_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
/* Expand OPSZ bytes worth of four-operand operations using host vectors. */
static void expand_4_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t cofs, uint32_t oprsz,
uint32_t tysz, TCGType type,
uint32_t tysz, TCGType type, bool write_aofs,
void (*fni)(unsigned, TCGv_vec, TCGv_vec,
TCGv_vec, TCGv_vec))
{
@ -896,6 +902,9 @@ static void expand_4_vec(unsigned vece, uint32_t dofs, uint32_t aofs,
tcg_gen_ld_vec(t3, cpu_env, cofs + i);
fni(vece, t0, t1, t2, t3);
tcg_gen_st_vec(t0, cpu_env, dofs + i);
if (write_aofs) {
tcg_gen_st_vec(t1, cpu_env, aofs + i);
}
}
tcg_temp_free_vec(t3);
tcg_temp_free_vec(t2);
@ -1187,7 +1196,7 @@ void tcg_gen_gvec_4(uint32_t dofs, uint32_t aofs, uint32_t bofs, uint32_t cofs,
*/
some = QEMU_ALIGN_DOWN(oprsz, 32);
expand_4_vec(g->vece, dofs, aofs, bofs, cofs, some,
32, TCG_TYPE_V256, g->fniv);
32, TCG_TYPE_V256, g->write_aofs, g->fniv);
if (some == oprsz) {
break;
}
@ -1200,18 +1209,20 @@ void tcg_gen_gvec_4(uint32_t dofs, uint32_t aofs, uint32_t bofs, uint32_t cofs,
/* fallthru */
case TCG_TYPE_V128:
expand_4_vec(g->vece, dofs, aofs, bofs, cofs, oprsz,
16, TCG_TYPE_V128, g->fniv);
16, TCG_TYPE_V128, g->write_aofs, g->fniv);
break;
case TCG_TYPE_V64:
expand_4_vec(g->vece, dofs, aofs, bofs, cofs, oprsz,
8, TCG_TYPE_V64, g->fniv);
8, TCG_TYPE_V64, g->write_aofs, g->fniv);
break;
case 0:
if (g->fni8 && check_size_impl(oprsz, 8)) {
expand_4_i64(dofs, aofs, bofs, cofs, oprsz, g->fni8);
expand_4_i64(dofs, aofs, bofs, cofs, oprsz,
g->write_aofs, g->fni8);
} else if (g->fni4 && check_size_impl(oprsz, 4)) {
expand_4_i32(dofs, aofs, bofs, cofs, oprsz, g->fni4);
expand_4_i32(dofs, aofs, bofs, cofs, oprsz,
g->write_aofs, g->fni4);
} else {
assert(g->fno != NULL);
tcg_gen_gvec_4_ool(dofs, aofs, bofs, cofs,
@ -1667,10 +1678,22 @@ void tcg_gen_gvec_ssadd(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fno = gen_helper_gvec_ssadd8, .vece = MO_8 },
{ .fno = gen_helper_gvec_ssadd16, .vece = MO_16 },
{ .fno = gen_helper_gvec_ssadd32, .vece = MO_32 },
{ .fno = gen_helper_gvec_ssadd64, .vece = MO_64 }
{ .fniv = tcg_gen_ssadd_vec,
.fno = gen_helper_gvec_ssadd8,
.opc = INDEX_op_ssadd_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_ssadd_vec,
.fno = gen_helper_gvec_ssadd16,
.opc = INDEX_op_ssadd_vec,
.vece = MO_16 },
{ .fniv = tcg_gen_ssadd_vec,
.fno = gen_helper_gvec_ssadd32,
.opc = INDEX_op_ssadd_vec,
.vece = MO_32 },
{ .fniv = tcg_gen_ssadd_vec,
.fno = gen_helper_gvec_ssadd64,
.opc = INDEX_op_ssadd_vec,
.vece = MO_64 },
};
tcg_debug_assert(vece <= MO_64);
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
@ -1680,16 +1703,28 @@ void tcg_gen_gvec_sssub(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fno = gen_helper_gvec_sssub8, .vece = MO_8 },
{ .fno = gen_helper_gvec_sssub16, .vece = MO_16 },
{ .fno = gen_helper_gvec_sssub32, .vece = MO_32 },
{ .fno = gen_helper_gvec_sssub64, .vece = MO_64 }
{ .fniv = tcg_gen_sssub_vec,
.fno = gen_helper_gvec_sssub8,
.opc = INDEX_op_sssub_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_sssub_vec,
.fno = gen_helper_gvec_sssub16,
.opc = INDEX_op_sssub_vec,
.vece = MO_16 },
{ .fniv = tcg_gen_sssub_vec,
.fno = gen_helper_gvec_sssub32,
.opc = INDEX_op_sssub_vec,
.vece = MO_32 },
{ .fniv = tcg_gen_sssub_vec,
.fno = gen_helper_gvec_sssub64,
.opc = INDEX_op_sssub_vec,
.vece = MO_64 },
};
tcg_debug_assert(vece <= MO_64);
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
}
static void tcg_gen_vec_usadd32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
static void tcg_gen_usadd_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
{
TCGv_i32 max = tcg_const_i32(-1);
tcg_gen_add_i32(d, a, b);
@ -1697,7 +1732,7 @@ static void tcg_gen_vec_usadd32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
tcg_temp_free_i32(max);
}
static void tcg_gen_vec_usadd32_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
static void tcg_gen_usadd_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
{
TCGv_i64 max = tcg_const_i64(-1);
tcg_gen_add_i64(d, a, b);
@ -1709,20 +1744,30 @@ void tcg_gen_gvec_usadd(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fno = gen_helper_gvec_usadd8, .vece = MO_8 },
{ .fno = gen_helper_gvec_usadd16, .vece = MO_16 },
{ .fni4 = tcg_gen_vec_usadd32_i32,
{ .fniv = tcg_gen_usadd_vec,
.fno = gen_helper_gvec_usadd8,
.opc = INDEX_op_usadd_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_usadd_vec,
.fno = gen_helper_gvec_usadd16,
.opc = INDEX_op_usadd_vec,
.vece = MO_16 },
{ .fni4 = tcg_gen_usadd_i32,
.fniv = tcg_gen_usadd_vec,
.fno = gen_helper_gvec_usadd32,
.opc = INDEX_op_usadd_vec,
.vece = MO_32 },
{ .fni8 = tcg_gen_vec_usadd32_i64,
{ .fni8 = tcg_gen_usadd_i64,
.fniv = tcg_gen_usadd_vec,
.fno = gen_helper_gvec_usadd64,
.opc = INDEX_op_usadd_vec,
.vece = MO_64 }
};
tcg_debug_assert(vece <= MO_64);
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
}
static void tcg_gen_vec_ussub32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
static void tcg_gen_ussub_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
{
TCGv_i32 min = tcg_const_i32(0);
tcg_gen_sub_i32(d, a, b);
@ -1730,7 +1775,7 @@ static void tcg_gen_vec_ussub32_i32(TCGv_i32 d, TCGv_i32 a, TCGv_i32 b)
tcg_temp_free_i32(min);
}
static void tcg_gen_vec_ussub32_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
static void tcg_gen_ussub_i64(TCGv_i64 d, TCGv_i64 a, TCGv_i64 b)
{
TCGv_i64 min = tcg_const_i64(0);
tcg_gen_sub_i64(d, a, b);
@ -1742,13 +1787,131 @@ void tcg_gen_gvec_ussub(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fno = gen_helper_gvec_ussub8, .vece = MO_8 },
{ .fno = gen_helper_gvec_ussub16, .vece = MO_16 },
{ .fni4 = tcg_gen_vec_ussub32_i32,
{ .fniv = tcg_gen_ussub_vec,
.fno = gen_helper_gvec_ussub8,
.opc = INDEX_op_ussub_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_ussub_vec,
.fno = gen_helper_gvec_ussub16,
.opc = INDEX_op_ussub_vec,
.vece = MO_16 },
{ .fni4 = tcg_gen_ussub_i32,
.fniv = tcg_gen_ussub_vec,
.fno = gen_helper_gvec_ussub32,
.opc = INDEX_op_ussub_vec,
.vece = MO_32 },
{ .fni8 = tcg_gen_vec_ussub32_i64,
{ .fni8 = tcg_gen_ussub_i64,
.fniv = tcg_gen_ussub_vec,
.fno = gen_helper_gvec_ussub64,
.opc = INDEX_op_ussub_vec,
.vece = MO_64 }
};
tcg_debug_assert(vece <= MO_64);
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
}
void tcg_gen_gvec_smin(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fniv = tcg_gen_smin_vec,
.fno = gen_helper_gvec_smin8,
.opc = INDEX_op_smin_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_smin_vec,
.fno = gen_helper_gvec_smin16,
.opc = INDEX_op_smin_vec,
.vece = MO_16 },
{ .fni4 = tcg_gen_smin_i32,
.fniv = tcg_gen_smin_vec,
.fno = gen_helper_gvec_smin32,
.opc = INDEX_op_smin_vec,
.vece = MO_32 },
{ .fni8 = tcg_gen_smin_i64,
.fniv = tcg_gen_smin_vec,
.fno = gen_helper_gvec_smin64,
.opc = INDEX_op_smin_vec,
.vece = MO_64 }
};
tcg_debug_assert(vece <= MO_64);
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
}
void tcg_gen_gvec_umin(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fniv = tcg_gen_umin_vec,
.fno = gen_helper_gvec_umin8,
.opc = INDEX_op_umin_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_umin_vec,
.fno = gen_helper_gvec_umin16,
.opc = INDEX_op_umin_vec,
.vece = MO_16 },
{ .fni4 = tcg_gen_umin_i32,
.fniv = tcg_gen_umin_vec,
.fno = gen_helper_gvec_umin32,
.opc = INDEX_op_umin_vec,
.vece = MO_32 },
{ .fni8 = tcg_gen_umin_i64,
.fniv = tcg_gen_umin_vec,
.fno = gen_helper_gvec_umin64,
.opc = INDEX_op_umin_vec,
.vece = MO_64 }
};
tcg_debug_assert(vece <= MO_64);
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
}
void tcg_gen_gvec_smax(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fniv = tcg_gen_smax_vec,
.fno = gen_helper_gvec_smax8,
.opc = INDEX_op_smax_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_smax_vec,
.fno = gen_helper_gvec_smax16,
.opc = INDEX_op_smax_vec,
.vece = MO_16 },
{ .fni4 = tcg_gen_smax_i32,
.fniv = tcg_gen_smax_vec,
.fno = gen_helper_gvec_smax32,
.opc = INDEX_op_smax_vec,
.vece = MO_32 },
{ .fni8 = tcg_gen_smax_i64,
.fniv = tcg_gen_smax_vec,
.fno = gen_helper_gvec_smax64,
.opc = INDEX_op_smax_vec,
.vece = MO_64 }
};
tcg_debug_assert(vece <= MO_64);
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g[vece]);
}
void tcg_gen_gvec_umax(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g[4] = {
{ .fniv = tcg_gen_umax_vec,
.fno = gen_helper_gvec_umax8,
.opc = INDEX_op_umax_vec,
.vece = MO_8 },
{ .fniv = tcg_gen_umax_vec,
.fno = gen_helper_gvec_umax16,
.opc = INDEX_op_umax_vec,
.vece = MO_16 },
{ .fni4 = tcg_gen_umax_i32,
.fniv = tcg_gen_umax_vec,
.fno = gen_helper_gvec_umax32,
.opc = INDEX_op_umax_vec,
.vece = MO_32 },
{ .fni8 = tcg_gen_umax_i64,
.fniv = tcg_gen_umax_vec,
.fno = gen_helper_gvec_umax64,
.opc = INDEX_op_umax_vec,
.vece = MO_64 }
};
tcg_debug_assert(vece <= MO_64);
@ -1840,7 +2003,12 @@ void tcg_gen_gvec_and(unsigned vece, uint32_t dofs, uint32_t aofs,
.opc = INDEX_op_and_vec,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
if (aofs == bofs) {
tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
void tcg_gen_gvec_or(unsigned vece, uint32_t dofs, uint32_t aofs,
@ -1853,7 +2021,12 @@ void tcg_gen_gvec_or(unsigned vece, uint32_t dofs, uint32_t aofs,
.opc = INDEX_op_or_vec,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
if (aofs == bofs) {
tcg_gen_gvec_mov(vece, dofs, aofs, oprsz, maxsz);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
void tcg_gen_gvec_xor(unsigned vece, uint32_t dofs, uint32_t aofs,
@ -1866,7 +2039,12 @@ void tcg_gen_gvec_xor(unsigned vece, uint32_t dofs, uint32_t aofs,
.opc = INDEX_op_xor_vec,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
if (aofs == bofs) {
tcg_gen_gvec_dup8i(dofs, oprsz, maxsz, 0);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
void tcg_gen_gvec_andc(unsigned vece, uint32_t dofs, uint32_t aofs,
@ -1879,7 +2057,12 @@ void tcg_gen_gvec_andc(unsigned vece, uint32_t dofs, uint32_t aofs,
.opc = INDEX_op_andc_vec,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
if (aofs == bofs) {
tcg_gen_gvec_dup8i(dofs, oprsz, maxsz, 0);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
void tcg_gen_gvec_orc(unsigned vece, uint32_t dofs, uint32_t aofs,
@ -1892,7 +2075,63 @@ void tcg_gen_gvec_orc(unsigned vece, uint32_t dofs, uint32_t aofs,
.opc = INDEX_op_orc_vec,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
if (aofs == bofs) {
tcg_gen_gvec_dup8i(dofs, oprsz, maxsz, -1);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
void tcg_gen_gvec_nand(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g = {
.fni8 = tcg_gen_nand_i64,
.fniv = tcg_gen_nand_vec,
.fno = gen_helper_gvec_nand,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
if (aofs == bofs) {
tcg_gen_gvec_not(vece, dofs, aofs, oprsz, maxsz);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
void tcg_gen_gvec_nor(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g = {
.fni8 = tcg_gen_nor_i64,
.fniv = tcg_gen_nor_vec,
.fno = gen_helper_gvec_nor,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
if (aofs == bofs) {
tcg_gen_gvec_not(vece, dofs, aofs, oprsz, maxsz);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
void tcg_gen_gvec_eqv(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz)
{
static const GVecGen3 g = {
.fni8 = tcg_gen_eqv_i64,
.fniv = tcg_gen_eqv_vec,
.fno = gen_helper_gvec_eqv,
.prefer_i64 = TCG_TARGET_REG_BITS == 64,
};
if (aofs == bofs) {
tcg_gen_gvec_dup8i(dofs, oprsz, maxsz, -1);
} else {
tcg_gen_gvec_3(dofs, aofs, bofs, oprsz, maxsz, &g);
}
}
static const GVecGen2s gop_ands = {

18
tcg/tcg-op-gvec.h
View File

@ -181,6 +181,8 @@ typedef struct {
uint8_t vece;
/* Prefer i64 to v64. */
bool prefer_i64;
/* Write aofs as a 2nd dest operand. */
bool write_aofs;
} GVecGen4;
void tcg_gen_gvec_2(uint32_t dofs, uint32_t aofs,
@ -232,6 +234,16 @@ void tcg_gen_gvec_usadd(unsigned vece, uint32_t dofs, uint32_t aofs,
void tcg_gen_gvec_ussub(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
/* Min/max. */
void tcg_gen_gvec_smin(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_umin(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_smax(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_umax(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_and(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_or(unsigned vece, uint32_t dofs, uint32_t aofs,
@ -242,6 +254,12 @@ void tcg_gen_gvec_andc(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_orc(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_nand(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_nor(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_eqv(unsigned vece, uint32_t dofs, uint32_t aofs,
uint32_t bofs, uint32_t oprsz, uint32_t maxsz);
void tcg_gen_gvec_andi(unsigned vece, uint32_t dofs, uint32_t aofs,
int64_t c, uint32_t oprsz, uint32_t maxsz);

75
tcg/tcg-op-vec.c
View File

@ -275,6 +275,27 @@ void tcg_gen_orc_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
}
}
void tcg_gen_nand_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
/* TODO: Add TCG_TARGET_HAS_nand_vec when adding a backend supports it. */
tcg_gen_and_vec(0, r, a, b);
tcg_gen_not_vec(0, r, r);
}
void tcg_gen_nor_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
/* TODO: Add TCG_TARGET_HAS_nor_vec when adding a backend supports it. */
tcg_gen_or_vec(0, r, a, b);
tcg_gen_not_vec(0, r, r);
}
void tcg_gen_eqv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
/* TODO: Add TCG_TARGET_HAS_eqv_vec when adding a backend supports it. */
tcg_gen_xor_vec(0, r, a, b);
tcg_gen_not_vec(0, r, r);
}
void tcg_gen_not_vec(unsigned vece, TCGv_vec r, TCGv_vec a)
{
if (TCG_TARGET_HAS_not_vec) {
@ -365,7 +386,8 @@ void tcg_gen_cmp_vec(TCGCond cond, unsigned vece,
}
}
void tcg_gen_mul_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
static void do_op3(unsigned vece, TCGv_vec r, TCGv_vec a,
TCGv_vec b, TCGOpcode opc)
{
TCGTemp *rt = tcgv_vec_temp(r);
TCGTemp *at = tcgv_vec_temp(a);
@ -378,11 +400,56 @@ void tcg_gen_mul_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
tcg_debug_assert(at->base_type >= type);
tcg_debug_assert(bt->base_type >= type);
can = tcg_can_emit_vec_op(INDEX_op_mul_vec, type, vece);
can = tcg_can_emit_vec_op(opc, type, vece);
if (can > 0) {
vec_gen_3(INDEX_op_mul_vec, type, vece, ri, ai, bi);
vec_gen_3(opc, type, vece, ri, ai, bi);
} else {
tcg_debug_assert(can < 0);
tcg_expand_vec_op(INDEX_op_mul_vec, type, vece, ri, ai, bi);
tcg_expand_vec_op(opc, type, vece, ri, ai, bi);
}
}
void tcg_gen_mul_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_mul_vec);
}
void tcg_gen_ssadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_ssadd_vec);
}
void tcg_gen_usadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_usadd_vec);
}
void tcg_gen_sssub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_sssub_vec);
}
void tcg_gen_ussub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_ussub_vec);
}
void tcg_gen_smin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_smin_vec);
}
void tcg_gen_umin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_umin_vec);
}
void tcg_gen_smax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_smax_vec);
}
void tcg_gen_umax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b)
{
do_op3(vece, r, a, b, INDEX_op_umax_vec);
}

11
tcg/tcg-op.h
View File

@ -962,8 +962,19 @@ void tcg_gen_or_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_xor_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_andc_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_orc_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_nand_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_nor_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_eqv_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_not_vec(unsigned vece, TCGv_vec r, TCGv_vec a);
void tcg_gen_neg_vec(unsigned vece, TCGv_vec r, TCGv_vec a);
void tcg_gen_ssadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_usadd_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_sssub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_ussub_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_smin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_umin_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_smax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_umax_vec(unsigned vece, TCGv_vec r, TCGv_vec a, TCGv_vec b);
void tcg_gen_shli_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i);
void tcg_gen_shri_vec(unsigned vece, TCGv_vec r, TCGv_vec a, int64_t i);

8
tcg/tcg-opc.h
View File

@ -222,6 +222,14 @@ DEF(add_vec, 1, 2, 0, IMPLVEC)
DEF(sub_vec, 1, 2, 0, IMPLVEC)
DEF(mul_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_mul_vec))
DEF(neg_vec, 1, 1, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_neg_vec))
DEF(ssadd_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_sat_vec))
DEF(usadd_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_sat_vec))
DEF(sssub_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_sat_vec))
DEF(ussub_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_sat_vec))
DEF(smin_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_minmax_vec))
DEF(umin_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_minmax_vec))
DEF(smax_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_minmax_vec))
DEF(umax_vec, 1, 2, 0, IMPLVEC | IMPL(TCG_TARGET_HAS_minmax_vec))
DEF(and_vec, 1, 2, 0, IMPLVEC)
DEF(or_vec, 1, 2, 0, IMPLVEC)

10
tcg/tcg.c
View File

@ -1607,6 +1607,16 @@ bool tcg_op_supported(TCGOpcode op)
case INDEX_op_shrv_vec:
case INDEX_op_sarv_vec:
return have_vec && TCG_TARGET_HAS_shv_vec;
case INDEX_op_ssadd_vec:
case INDEX_op_usadd_vec:
case INDEX_op_sssub_vec:
case INDEX_op_ussub_vec:
return have_vec && TCG_TARGET_HAS_sat_vec;
case INDEX_op_smin_vec:
case INDEX_op_umin_vec:
case INDEX_op_smax_vec:
case INDEX_op_umax_vec:
return have_vec && TCG_TARGET_HAS_minmax_vec;
default:
tcg_debug_assert(op > INDEX_op_last_generic && op < NB_OPS);

2
tcg/tcg.h
View File

@ -183,6 +183,8 @@ typedef uint64_t TCGRegSet;
#define TCG_TARGET_HAS_shs_vec 0
#define TCG_TARGET_HAS_shv_vec 0
#define TCG_TARGET_HAS_mul_vec 0
#define TCG_TARGET_HAS_sat_vec 0
#define TCG_TARGET_HAS_minmax_vec 0
#else
#define TCG_TARGET_MAYBE_vec 1
#endif