mirrors/qemu
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
957a0bb751
qemu/target/arm/sme_helper.c
Richard Henderson 23a5e3859f target/arm: Implement SME integer outer product 
This is SMOPA, SUMOPA, USMOPA_s, UMOPA, for both Int8 and Int16.

Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20220708151540.18136-28-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
2022-07-11 13:43:51 +01:00

1202 lines
43 KiB
C
Raw Blame History

/*
* ARM SME Operations
*
* Copyright (c) 2022 Linaro, Ltd.
*
* This library is free software; you can redistribute it and/or
* modify it under the terms of the GNU Lesser General Public
* License as published by the Free Software Foundation; either
* version 2.1 of the License, or (at your option) any later version.
*
* This library is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
* Lesser General Public License for more details.
*
* You should have received a copy of the GNU Lesser General Public
* License along with this library; if not, see <http://www.gnu.org/licenses/>.
*/
#include "qemu/osdep.h"
#include "cpu.h"
#include "internals.h"
#include "tcg/tcg-gvec-desc.h"
#include "exec/helper-proto.h"
#include "exec/cpu_ldst.h"
#include "exec/exec-all.h"
#include "qemu/int128.h"
#include "fpu/softfloat.h"
#include "vec_internal.h"
#include "sve_ldst_internal.h"
/* ResetSVEState */
void arm_reset_sve_state(CPUARMState *env)
{
memset(env->vfp.zregs, 0, sizeof(env->vfp.zregs));
/* Recall that FFR is stored as pregs[16]. */
memset(env->vfp.pregs, 0, sizeof(env->vfp.pregs));
vfp_set_fpcr(env, 0x0800009f);
}
void helper_set_pstate_sm(CPUARMState *env, uint32_t i)
{
if (i == FIELD_EX64(env->svcr, SVCR, SM)) {
return;
}
env->svcr ^= R_SVCR_SM_MASK;
arm_reset_sve_state(env);
}
void helper_set_pstate_za(CPUARMState *env, uint32_t i)
{
if (i == FIELD_EX64(env->svcr, SVCR, ZA)) {
return;
}
env->svcr ^= R_SVCR_ZA_MASK;
/*
* ResetSMEState.
*
* SetPSTATE_ZA zeros on enable and disable. We can zero this only
* on enable: while disabled, the storage is inaccessible and the
* value does not matter. We're not saving the storage in vmstate
* when disabled either.
*/
if (i) {
memset(env->zarray, 0, sizeof(env->zarray));
}
}
void helper_sme_zero(CPUARMState *env, uint32_t imm, uint32_t svl)
{
uint32_t i;
/*
* Special case clearing the entire ZA space.
* This falls into the CONSTRAINED UNPREDICTABLE zeroing of any
* parts of the ZA storage outside of SVL.
*/
if (imm == 0xff) {
memset(env->zarray, 0, sizeof(env->zarray));
return;
}
/*
* Recall that ZAnH.D[m] is spread across ZA[n+8*m],
* so each row is discontiguous within ZA[].
*/
for (i = 0; i < svl; i++) {
if (imm & (1 << (i % 8))) {
memset(&env->zarray[i], 0, svl);
}
}
}
/*
* When considering the ZA storage as an array of elements of
* type T, the index within that array of the Nth element of
* a vertical slice of a tile can be calculated like this,
* regardless of the size of type T. This is because the tiles
* are interleaved, so if type T is size N bytes then row 1 of
* the tile is N rows away from row 0. The division by N to
* convert a byte offset into an array index and the multiplication
* by N to convert from vslice-index-within-the-tile to
* the index within the ZA storage cancel out.
*/
#define tile_vslice_index(i) ((i) * sizeof(ARMVectorReg))
/*
* When doing byte arithmetic on the ZA storage, the element
* byteoff bytes away in a tile vertical slice is always this
* many bytes away in the ZA storage, regardless of the
* size of the tile element, assuming that byteoff is a multiple
* of the element size. Again this is because of the interleaving
* of the tiles. For instance if we have 1 byte per element then
* each row of the ZA storage has one byte of the vslice data,
* and (counting from 0) byte 8 goes in row 8 of the storage
* at offset (8 * row-size-in-bytes).
* If we have 8 bytes per element then each row of the ZA storage
* has 8 bytes of the data, but there are 8 interleaved tiles and
* so byte 8 of the data goes into row 1 of the tile,
* which is again row 8 of the storage, so the offset is still
* (8 * row-size-in-bytes). Similarly for other element sizes.
*/
#define tile_vslice_offset(byteoff) ((byteoff) * sizeof(ARMVectorReg))
/*
* Move Zreg vector to ZArray column.
*/
#define DO_MOVA_C(NAME, TYPE, H) \
void HELPER(NAME)(void *za, void *vn, void *vg, uint32_t desc) \
{ \
int i, oprsz = simd_oprsz(desc); \
for (i = 0; i < oprsz; ) { \
uint16_t pg = *(uint16_t *)(vg + H1_2(i >> 3)); \
do { \
if (pg & 1) { \
*(TYPE *)(za + tile_vslice_offset(i)) = *(TYPE *)(vn + H(i)); \
} \
i += sizeof(TYPE); \
pg >>= sizeof(TYPE); \
} while (i & 15); \
} \
}
DO_MOVA_C(sme_mova_cz_b, uint8_t, H1)
DO_MOVA_C(sme_mova_cz_h, uint16_t, H1_2)
DO_MOVA_C(sme_mova_cz_s, uint32_t, H1_4)
void HELPER(sme_mova_cz_d)(void *za, void *vn, void *vg, uint32_t desc)
{
int i, oprsz = simd_oprsz(desc) / 8;
uint8_t *pg = vg;
uint64_t *n = vn;
uint64_t *a = za;
for (i = 0; i < oprsz; i++) {
if (pg[H1(i)] & 1) {
a[tile_vslice_index(i)] = n[i];
}
}
}
void HELPER(sme_mova_cz_q)(void *za, void *vn, void *vg, uint32_t desc)
{
int i, oprsz = simd_oprsz(desc) / 16;
uint16_t *pg = vg;
Int128 *n = vn;
Int128 *a = za;
/*
* Int128 is used here simply to copy 16 bytes, and to simplify
* the address arithmetic.
*/
for (i = 0; i < oprsz; i++) {
if (pg[H2(i)] & 1) {
a[tile_vslice_index(i)] = n[i];
}
}
}
#undef DO_MOVA_C
/*
* Move ZArray column to Zreg vector.
*/
#define DO_MOVA_Z(NAME, TYPE, H) \
void HELPER(NAME)(void *vd, void *za, void *vg, uint32_t desc) \
{ \
int i, oprsz = simd_oprsz(desc); \
for (i = 0; i < oprsz; ) { \
uint16_t pg = *(uint16_t *)(vg + H1_2(i >> 3)); \
do { \
if (pg & 1) { \
*(TYPE *)(vd + H(i)) = *(TYPE *)(za + tile_vslice_offset(i)); \
} \
i += sizeof(TYPE); \
pg >>= sizeof(TYPE); \
} while (i & 15); \
} \
}
DO_MOVA_Z(sme_mova_zc_b, uint8_t, H1)
DO_MOVA_Z(sme_mova_zc_h, uint16_t, H1_2)
DO_MOVA_Z(sme_mova_zc_s, uint32_t, H1_4)
void HELPER(sme_mova_zc_d)(void *vd, void *za, void *vg, uint32_t desc)
{
int i, oprsz = simd_oprsz(desc) / 8;
uint8_t *pg = vg;
uint64_t *d = vd;
uint64_t *a = za;
for (i = 0; i < oprsz; i++) {
if (pg[H1(i)] & 1) {
d[i] = a[tile_vslice_index(i)];
}
}
}
void HELPER(sme_mova_zc_q)(void *vd, void *za, void *vg, uint32_t desc)
{
int i, oprsz = simd_oprsz(desc) / 16;
uint16_t *pg = vg;
Int128 *d = vd;
Int128 *a = za;
/*
* Int128 is used here simply to copy 16 bytes, and to simplify
* the address arithmetic.
*/
for (i = 0; i < oprsz; i++, za += sizeof(ARMVectorReg)) {
if (pg[H2(i)] & 1) {
d[i] = a[tile_vslice_index(i)];
}
}
}
#undef DO_MOVA_Z
/*
* Clear elements in a tile slice comprising len bytes.
*/
typedef void ClearFn(void *ptr, size_t off, size_t len);
static void clear_horizontal(void *ptr, size_t off, size_t len)
{
memset(ptr + off, 0, len);
}
static void clear_vertical_b(void *vptr, size_t off, size_t len)
{
for (size_t i = 0; i < len; ++i) {
*(uint8_t *)(vptr + tile_vslice_offset(i + off)) = 0;
}
}
static void clear_vertical_h(void *vptr, size_t off, size_t len)
{
for (size_t i = 0; i < len; i += 2) {
*(uint16_t *)(vptr + tile_vslice_offset(i + off)) = 0;
}
}
static void clear_vertical_s(void *vptr, size_t off, size_t len)
{
for (size_t i = 0; i < len; i += 4) {
*(uint32_t *)(vptr + tile_vslice_offset(i + off)) = 0;
}
}
static void clear_vertical_d(void *vptr, size_t off, size_t len)
{
for (size_t i = 0; i < len; i += 8) {
*(uint64_t *)(vptr + tile_vslice_offset(i + off)) = 0;
}
}
static void clear_vertical_q(void *vptr, size_t off, size_t len)
{
for (size_t i = 0; i < len; i += 16) {
memset(vptr + tile_vslice_offset(i + off), 0, 16);
}
}
/*
* Copy elements from an array into a tile slice comprising len bytes.
*/
typedef void CopyFn(void *dst, const void *src, size_t len);
static void copy_horizontal(void *dst, const void *src, size_t len)
{
memcpy(dst, src, len);
}
static void copy_vertical_b(void *vdst, const void *vsrc, size_t len)
{
const uint8_t *src = vsrc;
uint8_t *dst = vdst;
size_t i;
for (i = 0; i < len; ++i) {
dst[tile_vslice_index(i)] = src[i];
}
}
static void copy_vertical_h(void *vdst, const void *vsrc, size_t len)
{
const uint16_t *src = vsrc;
uint16_t *dst = vdst;
size_t i;
for (i = 0; i < len / 2; ++i) {
dst[tile_vslice_index(i)] = src[i];
}
}
static void copy_vertical_s(void *vdst, const void *vsrc, size_t len)
{
const uint32_t *src = vsrc;
uint32_t *dst = vdst;
size_t i;
for (i = 0; i < len / 4; ++i) {
dst[tile_vslice_index(i)] = src[i];
}
}
static void copy_vertical_d(void *vdst, const void *vsrc, size_t len)
{
const uint64_t *src = vsrc;
uint64_t *dst = vdst;
size_t i;
for (i = 0; i < len / 8; ++i) {
dst[tile_vslice_index(i)] = src[i];
}
}
static void copy_vertical_q(void *vdst, const void *vsrc, size_t len)
{
for (size_t i = 0; i < len; i += 16) {
memcpy(vdst + tile_vslice_offset(i), vsrc + i, 16);
}
}
/*
* Host and TLB primitives for vertical tile slice addressing.
*/
#define DO_LD(NAME, TYPE, HOST, TLB) \
static inline void sme_##NAME##_v_host(void *za, intptr_t off, void *host) \
{ \
TYPE val = HOST(host); \
*(TYPE *)(za + tile_vslice_offset(off)) = val; \
} \
static inline void sme_##NAME##_v_tlb(CPUARMState *env, void *za, \
intptr_t off, target_ulong addr, uintptr_t ra) \
{ \
TYPE val = TLB(env, useronly_clean_ptr(addr), ra); \
*(TYPE *)(za + tile_vslice_offset(off)) = val; \
}
#define DO_ST(NAME, TYPE, HOST, TLB) \
static inline void sme_##NAME##_v_host(void *za, intptr_t off, void *host) \
{ \
TYPE val = *(TYPE *)(za + tile_vslice_offset(off)); \
HOST(host, val); \
} \
static inline void sme_##NAME##_v_tlb(CPUARMState *env, void *za, \
intptr_t off, target_ulong addr, uintptr_t ra) \
{ \
TYPE val = *(TYPE *)(za + tile_vslice_offset(off)); \
TLB(env, useronly_clean_ptr(addr), val, ra); \
}
/*
* The ARMVectorReg elements are stored in host-endian 64-bit units.
* For 128-bit quantities, the sequence defined by the Elem[] pseudocode
* corresponds to storing the two 64-bit pieces in little-endian order.
*/
#define DO_LDQ(HNAME, VNAME, BE, HOST, TLB) \
static inline void HNAME##_host(void *za, intptr_t off, void *host) \
{ \
uint64_t val0 = HOST(host), val1 = HOST(host + 8); \
uint64_t *ptr = za + off; \
ptr[0] = BE ? val1 : val0, ptr[1] = BE ? val0 : val1; \
} \
static inline void VNAME##_v_host(void *za, intptr_t off, void *host) \
{ \
HNAME##_host(za, tile_vslice_offset(off), host); \
} \
static inline void HNAME##_tlb(CPUARMState *env, void *za, intptr_t off, \
target_ulong addr, uintptr_t ra) \
{ \
uint64_t val0 = TLB(env, useronly_clean_ptr(addr), ra); \
uint64_t val1 = TLB(env, useronly_clean_ptr(addr + 8), ra); \
uint64_t *ptr = za + off; \
ptr[0] = BE ? val1 : val0, ptr[1] = BE ? val0 : val1; \
} \
static inline void VNAME##_v_tlb(CPUARMState *env, void *za, intptr_t off, \
target_ulong addr, uintptr_t ra) \
{ \
HNAME##_tlb(env, za, tile_vslice_offset(off), addr, ra); \
}
#define DO_STQ(HNAME, VNAME, BE, HOST, TLB) \
static inline void HNAME##_host(void *za, intptr_t off, void *host) \
{ \
uint64_t *ptr = za + off; \
HOST(host, ptr[BE]); \
HOST(host + 1, ptr[!BE]); \
} \
static inline void VNAME##_v_host(void *za, intptr_t off, void *host) \
{ \
HNAME##_host(za, tile_vslice_offset(off), host); \
} \
static inline void HNAME##_tlb(CPUARMState *env, void *za, intptr_t off, \
target_ulong addr, uintptr_t ra) \
{ \
uint64_t *ptr = za + off; \
TLB(env, useronly_clean_ptr(addr), ptr[BE], ra); \
TLB(env, useronly_clean_ptr(addr + 8), ptr[!BE], ra); \
} \
static inline void VNAME##_v_tlb(CPUARMState *env, void *za, intptr_t off, \
target_ulong addr, uintptr_t ra) \
{ \
HNAME##_tlb(env, za, tile_vslice_offset(off), addr, ra); \
}
DO_LD(ld1b, uint8_t, ldub_p, cpu_ldub_data_ra)
DO_LD(ld1h_be, uint16_t, lduw_be_p, cpu_lduw_be_data_ra)
DO_LD(ld1h_le, uint16_t, lduw_le_p, cpu_lduw_le_data_ra)
DO_LD(ld1s_be, uint32_t, ldl_be_p, cpu_ldl_be_data_ra)
DO_LD(ld1s_le, uint32_t, ldl_le_p, cpu_ldl_le_data_ra)
DO_LD(ld1d_be, uint64_t, ldq_be_p, cpu_ldq_be_data_ra)
DO_LD(ld1d_le, uint64_t, ldq_le_p, cpu_ldq_le_data_ra)
DO_LDQ(sve_ld1qq_be, sme_ld1q_be, 1, ldq_be_p, cpu_ldq_be_data_ra)
DO_LDQ(sve_ld1qq_le, sme_ld1q_le, 0, ldq_le_p, cpu_ldq_le_data_ra)
DO_ST(st1b, uint8_t, stb_p, cpu_stb_data_ra)
DO_ST(st1h_be, uint16_t, stw_be_p, cpu_stw_be_data_ra)
DO_ST(st1h_le, uint16_t, stw_le_p, cpu_stw_le_data_ra)
DO_ST(st1s_be, uint32_t, stl_be_p, cpu_stl_be_data_ra)
DO_ST(st1s_le, uint32_t, stl_le_p, cpu_stl_le_data_ra)
DO_ST(st1d_be, uint64_t, stq_be_p, cpu_stq_be_data_ra)
DO_ST(st1d_le, uint64_t, stq_le_p, cpu_stq_le_data_ra)
DO_STQ(sve_st1qq_be, sme_st1q_be, 1, stq_be_p, cpu_stq_be_data_ra)
DO_STQ(sve_st1qq_le, sme_st1q_le, 0, stq_le_p, cpu_stq_le_data_ra)
#undef DO_LD
#undef DO_ST
#undef DO_LDQ
#undef DO_STQ
/*
* Common helper for all contiguous predicated loads.
*/
static inline QEMU_ALWAYS_INLINE
void sme_ld1(CPUARMState *env, void *za, uint64_t *vg,
const target_ulong addr, uint32_t desc, const uintptr_t ra,
const int esz, uint32_t mtedesc, bool vertical,
sve_ldst1_host_fn *host_fn,
sve_ldst1_tlb_fn *tlb_fn,
ClearFn *clr_fn,
CopyFn *cpy_fn)
{
const intptr_t reg_max = simd_oprsz(desc);
const intptr_t esize = 1 << esz;
intptr_t reg_off, reg_last;
SVEContLdSt info;
void *host;
int flags;
/* Find the active elements. */
if (!sve_cont_ldst_elements(&info, addr, vg, reg_max, esz, esize)) {
/* The entire predicate was false; no load occurs. */
clr_fn(za, 0, reg_max);
return;
}
/* Probe the page(s). Exit with exception for any invalid page. */
sve_cont_ldst_pages(&info, FAULT_ALL, env, addr, MMU_DATA_LOAD, ra);
/* Handle watchpoints for all active elements. */
sve_cont_ldst_watchpoints(&info, env, vg, addr, esize, esize,
BP_MEM_READ, ra);
/*
* Handle mte checks for all active elements.
* Since TBI must be set for MTE, !mtedesc => !mte_active.
*/
if (mtedesc) {
sve_cont_ldst_mte_check(&info, env, vg, addr, esize, esize,
mtedesc, ra);
}
flags = info.page[0].flags | info.page[1].flags;
if (unlikely(flags != 0)) {
#ifdef CONFIG_USER_ONLY
g_assert_not_reached();
#else
/*
* At least one page includes MMIO.
* Any bus operation can fail with cpu_transaction_failed,
* which for ARM will raise SyncExternal. Perform the load
* into scratch memory to preserve register state until the end.
*/
ARMVectorReg scratch = { };
reg_off = info.reg_off_first[0];
reg_last = info.reg_off_last[1];
if (reg_last < 0) {
reg_last = info.reg_off_split;
if (reg_last < 0) {
reg_last = info.reg_off_last[0];
}
}
do {
uint64_t pg = vg[reg_off >> 6];
do {
if ((pg >> (reg_off & 63)) & 1) {
tlb_fn(env, &scratch, reg_off, addr + reg_off, ra);
}
reg_off += esize;
} while (reg_off & 63);
} while (reg_off <= reg_last);
cpy_fn(za, &scratch, reg_max);
return;
#endif
}
/* The entire operation is in RAM, on valid pages. */
reg_off = info.reg_off_first[0];
reg_last = info.reg_off_last[0];
host = info.page[0].host;
if (!vertical) {
memset(za, 0, reg_max);
} else if (reg_off) {
clr_fn(za, 0, reg_off);
}
while (reg_off <= reg_last) {
uint64_t pg = vg[reg_off >> 6];
do {
if ((pg >> (reg_off & 63)) & 1) {
host_fn(za, reg_off, host + reg_off);
} else if (vertical) {
clr_fn(za, reg_off, esize);
}
reg_off += esize;
} while (reg_off <= reg_last && (reg_off & 63));
}
/*
* Use the slow path to manage the cross-page misalignment.
* But we know this is RAM and cannot trap.
*/
reg_off = info.reg_off_split;
if (unlikely(reg_off >= 0)) {
tlb_fn(env, za, reg_off, addr + reg_off, ra);
}
reg_off = info.reg_off_first[1];
if (unlikely(reg_off >= 0)) {
reg_last = info.reg_off_last[1];
host = info.page[1].host;
do {
uint64_t pg = vg[reg_off >> 6];
do {
if ((pg >> (reg_off & 63)) & 1) {
host_fn(za, reg_off, host + reg_off);
} else if (vertical) {
clr_fn(za, reg_off, esize);
}
reg_off += esize;
} while (reg_off & 63);
} while (reg_off <= reg_last);
}
}
static inline QEMU_ALWAYS_INLINE
void sme_ld1_mte(CPUARMState *env, void *za, uint64_t *vg,
target_ulong addr, uint32_t desc, uintptr_t ra,
const int esz, bool vertical,
sve_ldst1_host_fn *host_fn,
sve_ldst1_tlb_fn *tlb_fn,
ClearFn *clr_fn,
CopyFn *cpy_fn)
{
uint32_t mtedesc = desc >> (SIMD_DATA_SHIFT + SVE_MTEDESC_SHIFT);
int bit55 = extract64(addr, 55, 1);
/* Remove mtedesc from the normal sve descriptor. */
desc = extract32(desc, 0, SIMD_DATA_SHIFT + SVE_MTEDESC_SHIFT);
/* Perform gross MTE suppression early. */
if (!tbi_check(desc, bit55) ||
tcma_check(desc, bit55, allocation_tag_from_addr(addr))) {
mtedesc = 0;
}
sme_ld1(env, za, vg, addr, desc, ra, esz, mtedesc, vertical,
host_fn, tlb_fn, clr_fn, cpy_fn);
}
#define DO_LD(L, END, ESZ) \
void HELPER(sme_ld1##L##END##_h)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_ld1(env, za, vg, addr, desc, GETPC(), ESZ, 0, false, \
sve_ld1##L##L##END##_host, sve_ld1##L##L##END##_tlb, \
clear_horizontal, copy_horizontal); \
} \
void HELPER(sme_ld1##L##END##_v)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_ld1(env, za, vg, addr, desc, GETPC(), ESZ, 0, true, \
sme_ld1##L##END##_v_host, sme_ld1##L##END##_v_tlb, \
clear_vertical_##L, copy_vertical_##L); \
} \
void HELPER(sme_ld1##L##END##_h_mte)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_ld1_mte(env, za, vg, addr, desc, GETPC(), ESZ, false, \
sve_ld1##L##L##END##_host, sve_ld1##L##L##END##_tlb, \
clear_horizontal, copy_horizontal); \
} \
void HELPER(sme_ld1##L##END##_v_mte)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_ld1_mte(env, za, vg, addr, desc, GETPC(), ESZ, true, \
sme_ld1##L##END##_v_host, sme_ld1##L##END##_v_tlb, \
clear_vertical_##L, copy_vertical_##L); \
}
DO_LD(b, , MO_8)
DO_LD(h, _be, MO_16)
DO_LD(h, _le, MO_16)
DO_LD(s, _be, MO_32)
DO_LD(s, _le, MO_32)
DO_LD(d, _be, MO_64)
DO_LD(d, _le, MO_64)
DO_LD(q, _be, MO_128)
DO_LD(q, _le, MO_128)
#undef DO_LD
/*
* Common helper for all contiguous predicated stores.
*/
static inline QEMU_ALWAYS_INLINE
void sme_st1(CPUARMState *env, void *za, uint64_t *vg,
const target_ulong addr, uint32_t desc, const uintptr_t ra,
const int esz, uint32_t mtedesc, bool vertical,
sve_ldst1_host_fn *host_fn,
sve_ldst1_tlb_fn *tlb_fn)
{
const intptr_t reg_max = simd_oprsz(desc);
const intptr_t esize = 1 << esz;
intptr_t reg_off, reg_last;
SVEContLdSt info;
void *host;
int flags;
/* Find the active elements. */
if (!sve_cont_ldst_elements(&info, addr, vg, reg_max, esz, esize)) {
/* The entire predicate was false; no store occurs. */
return;
}
/* Probe the page(s). Exit with exception for any invalid page. */
sve_cont_ldst_pages(&info, FAULT_ALL, env, addr, MMU_DATA_STORE, ra);
/* Handle watchpoints for all active elements. */
sve_cont_ldst_watchpoints(&info, env, vg, addr, esize, esize,
BP_MEM_WRITE, ra);
/*
* Handle mte checks for all active elements.
* Since TBI must be set for MTE, !mtedesc => !mte_active.
*/
if (mtedesc) {
sve_cont_ldst_mte_check(&info, env, vg, addr, esize, esize,
mtedesc, ra);
}
flags = info.page[0].flags | info.page[1].flags;
if (unlikely(flags != 0)) {
#ifdef CONFIG_USER_ONLY
g_assert_not_reached();
#else
/*
* At least one page includes MMIO.
* Any bus operation can fail with cpu_transaction_failed,
* which for ARM will raise SyncExternal. We cannot avoid
* this fault and will leave with the store incomplete.
*/
reg_off = info.reg_off_first[0];
reg_last = info.reg_off_last[1];
if (reg_last < 0) {
reg_last = info.reg_off_split;
if (reg_last < 0) {
reg_last = info.reg_off_last[0];
}
}
do {
uint64_t pg = vg[reg_off >> 6];
do {
if ((pg >> (reg_off & 63)) & 1) {
tlb_fn(env, za, reg_off, addr + reg_off, ra);
}
reg_off += esize;
} while (reg_off & 63);
} while (reg_off <= reg_last);
return;
#endif
}
reg_off = info.reg_off_first[0];
reg_last = info.reg_off_last[0];
host = info.page[0].host;
while (reg_off <= reg_last) {
uint64_t pg = vg[reg_off >> 6];
do {
if ((pg >> (reg_off & 63)) & 1) {
host_fn(za, reg_off, host + reg_off);
}
reg_off += 1 << esz;
} while (reg_off <= reg_last && (reg_off & 63));
}
/*
* Use the slow path to manage the cross-page misalignment.
* But we know this is RAM and cannot trap.
*/
reg_off = info.reg_off_split;
if (unlikely(reg_off >= 0)) {
tlb_fn(env, za, reg_off, addr + reg_off, ra);
}
reg_off = info.reg_off_first[1];
if (unlikely(reg_off >= 0)) {
reg_last = info.reg_off_last[1];
host = info.page[1].host;
do {
uint64_t pg = vg[reg_off >> 6];
do {
if ((pg >> (reg_off & 63)) & 1) {
host_fn(za, reg_off, host + reg_off);
}
reg_off += 1 << esz;
} while (reg_off & 63);
} while (reg_off <= reg_last);
}
}
static inline QEMU_ALWAYS_INLINE
void sme_st1_mte(CPUARMState *env, void *za, uint64_t *vg, target_ulong addr,
uint32_t desc, uintptr_t ra, int esz, bool vertical,
sve_ldst1_host_fn *host_fn,
sve_ldst1_tlb_fn *tlb_fn)
{
uint32_t mtedesc = desc >> (SIMD_DATA_SHIFT + SVE_MTEDESC_SHIFT);
int bit55 = extract64(addr, 55, 1);
/* Remove mtedesc from the normal sve descriptor. */
desc = extract32(desc, 0, SIMD_DATA_SHIFT + SVE_MTEDESC_SHIFT);
/* Perform gross MTE suppression early. */
if (!tbi_check(desc, bit55) ||
tcma_check(desc, bit55, allocation_tag_from_addr(addr))) {
mtedesc = 0;
}
sme_st1(env, za, vg, addr, desc, ra, esz, mtedesc,
vertical, host_fn, tlb_fn);
}
#define DO_ST(L, END, ESZ) \
void HELPER(sme_st1##L##END##_h)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_st1(env, za, vg, addr, desc, GETPC(), ESZ, 0, false, \
sve_st1##L##L##END##_host, sve_st1##L##L##END##_tlb); \
} \
void HELPER(sme_st1##L##END##_v)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_st1(env, za, vg, addr, desc, GETPC(), ESZ, 0, true, \
sme_st1##L##END##_v_host, sme_st1##L##END##_v_tlb); \
} \
void HELPER(sme_st1##L##END##_h_mte)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_st1_mte(env, za, vg, addr, desc, GETPC(), ESZ, false, \
sve_st1##L##L##END##_host, sve_st1##L##L##END##_tlb); \
} \
void HELPER(sme_st1##L##END##_v_mte)(CPUARMState *env, void *za, void *vg, \
target_ulong addr, uint32_t desc) \
{ \
sme_st1_mte(env, za, vg, addr, desc, GETPC(), ESZ, true, \
sme_st1##L##END##_v_host, sme_st1##L##END##_v_tlb); \
}
DO_ST(b, , MO_8)
DO_ST(h, _be, MO_16)
DO_ST(h, _le, MO_16)
DO_ST(s, _be, MO_32)
DO_ST(s, _le, MO_32)
DO_ST(d, _be, MO_64)
DO_ST(d, _le, MO_64)
DO_ST(q, _be, MO_128)
DO_ST(q, _le, MO_128)
#undef DO_ST
void HELPER(sme_addha_s)(void *vzda, void *vzn, void *vpn,
void *vpm, uint32_t desc)
{
intptr_t row, col, oprsz = simd_oprsz(desc) / 4;
uint64_t *pn = vpn, *pm = vpm;
uint32_t *zda = vzda, *zn = vzn;
for (row = 0; row < oprsz; ) {
uint64_t pa = pn[row >> 4];
do {
if (pa & 1) {
for (col = 0; col < oprsz; ) {
uint64_t pb = pm[col >> 4];
do {
if (pb & 1) {
zda[tile_vslice_index(row) + H4(col)] += zn[H4(col)];
}
pb >>= 4;
} while (++col & 15);
}
}
pa >>= 4;
} while (++row & 15);
}
}
void HELPER(sme_addha_d)(void *vzda, void *vzn, void *vpn,
void *vpm, uint32_t desc)
{
intptr_t row, col, oprsz = simd_oprsz(desc) / 8;
uint8_t *pn = vpn, *pm = vpm;
uint64_t *zda = vzda, *zn = vzn;
for (row = 0; row < oprsz; ++row) {
if (pn[H1(row)] & 1) {
for (col = 0; col < oprsz; ++col) {
if (pm[H1(col)] & 1) {
zda[tile_vslice_index(row) + col] += zn[col];
}
}
}
}
}
void HELPER(sme_addva_s)(void *vzda, void *vzn, void *vpn,
void *vpm, uint32_t desc)
{
intptr_t row, col, oprsz = simd_oprsz(desc) / 4;
uint64_t *pn = vpn, *pm = vpm;
uint32_t *zda = vzda, *zn = vzn;
for (row = 0; row < oprsz; ) {
uint64_t pa = pn[row >> 4];
do {
if (pa & 1) {
uint32_t zn_row = zn[H4(row)];
for (col = 0; col < oprsz; ) {
uint64_t pb = pm[col >> 4];
do {
if (pb & 1) {
zda[tile_vslice_index(row) + H4(col)] += zn_row;
}
pb >>= 4;
} while (++col & 15);
}
}
pa >>= 4;
} while (++row & 15);
}
}
void HELPER(sme_addva_d)(void *vzda, void *vzn, void *vpn,
void *vpm, uint32_t desc)
{
intptr_t row, col, oprsz = simd_oprsz(desc) / 8;
uint8_t *pn = vpn, *pm = vpm;
uint64_t *zda = vzda, *zn = vzn;
for (row = 0; row < oprsz; ++row) {
if (pn[H1(row)] & 1) {
uint64_t zn_row = zn[row];
for (col = 0; col < oprsz; ++col) {
if (pm[H1(col)] & 1) {
zda[tile_vslice_index(row) + col] += zn_row;
}
}
}
}
}
void HELPER(sme_fmopa_s)(void *vza, void *vzn, void *vzm, void *vpn,
void *vpm, void *vst, uint32_t desc)
{
intptr_t row, col, oprsz = simd_maxsz(desc);
uint32_t neg = simd_data(desc) << 31;
uint16_t *pn = vpn, *pm = vpm;
float_status fpst;
/*
* Make a copy of float_status because this operation does not
* update the cumulative fp exception status. It also produces
* default nans.
*/
fpst = *(float_status *)vst;
set_default_nan_mode(true, &fpst);
for (row = 0; row < oprsz; ) {
uint16_t pa = pn[H2(row >> 4)];
do {
if (pa & 1) {
void *vza_row = vza + tile_vslice_offset(row);
uint32_t n = *(uint32_t *)(vzn + H1_4(row)) ^ neg;
for (col = 0; col < oprsz; ) {
uint16_t pb = pm[H2(col >> 4)];
do {
if (pb & 1) {
uint32_t *a = vza_row + H1_4(col);
uint32_t *m = vzm + H1_4(col);
*a = float32_muladd(n, *m, *a, 0, vst);
}
col += 4;
pb >>= 4;
} while (col & 15);
}
}
row += 4;
pa >>= 4;
} while (row & 15);
}
}
void HELPER(sme_fmopa_d)(void *vza, void *vzn, void *vzm, void *vpn,
void *vpm, void *vst, uint32_t desc)
{
intptr_t row, col, oprsz = simd_oprsz(desc) / 8;
uint64_t neg = (uint64_t)simd_data(desc) << 63;
uint64_t *za = vza, *zn = vzn, *zm = vzm;
uint8_t *pn = vpn, *pm = vpm;
float_status fpst = *(float_status *)vst;
set_default_nan_mode(true, &fpst);
for (row = 0; row < oprsz; ++row) {
if (pn[H1(row)] & 1) {
uint64_t *za_row = &za[tile_vslice_index(row)];
uint64_t n = zn[row] ^ neg;
for (col = 0; col < oprsz; ++col) {
if (pm[H1(col)] & 1) {
uint64_t *a = &za_row[col];
*a = float64_muladd(n, zm[col], *a, 0, &fpst);
}
}
}
}
}
/*
* Alter PAIR as needed for controlling predicates being false,
* and for NEG on an enabled row element.
*/
static inline uint32_t f16mop_adj_pair(uint32_t pair, uint32_t pg, uint32_t neg)
{
/*
* The pseudocode uses a conditional negate after the conditional zero.
* It is simpler here to unconditionally negate before conditional zero.
*/
pair ^= neg;
if (!(pg & 1)) {
pair &= 0xffff0000u;
}
if (!(pg & 4)) {
pair &= 0x0000ffffu;
}
return pair;
}
static float32 f16_dotadd(float32 sum, uint32_t e1, uint32_t e2,
float_status *s_std, float_status *s_odd)
{
float64 e1r = float16_to_float64(e1 & 0xffff, true, s_std);
float64 e1c = float16_to_float64(e1 >> 16, true, s_std);
float64 e2r = float16_to_float64(e2 & 0xffff, true, s_std);
float64 e2c = float16_to_float64(e2 >> 16, true, s_std);
float64 t64;
float32 t32;
/*
* The ARM pseudocode function FPDot performs both multiplies
* and the add with a single rounding operation. Emulate this
* by performing the first multiply in round-to-odd, then doing
* the second multiply as fused multiply-add, and rounding to
* float32 all in one step.
*/
t64 = float64_mul(e1r, e2r, s_odd);
t64 = float64r32_muladd(e1c, e2c, t64, 0, s_std);
/* This conversion is exact, because we've already rounded. */
t32 = float64_to_float32(t64, s_std);
/* The final accumulation step is not fused. */
return float32_add(sum, t32, s_std);
}
void HELPER(sme_fmopa_h)(void *vza, void *vzn, void *vzm, void *vpn,
void *vpm, void *vst, uint32_t desc)
{
intptr_t row, col, oprsz = simd_maxsz(desc);
uint32_t neg = simd_data(desc) * 0x80008000u;
uint16_t *pn = vpn, *pm = vpm;
float_status fpst_odd, fpst_std;
/*
* Make a copy of float_status because this operation does not
* update the cumulative fp exception status. It also produces
* default nans. Make a second copy with round-to-odd -- see above.
*/
fpst_std = *(float_status *)vst;
set_default_nan_mode(true, &fpst_std);
fpst_odd = fpst_std;
set_float_rounding_mode(float_round_to_odd, &fpst_odd);
for (row = 0; row < oprsz; ) {
uint16_t prow = pn[H2(row >> 4)];
do {
void *vza_row = vza + tile_vslice_offset(row);
uint32_t n = *(uint32_t *)(vzn + H1_4(row));
n = f16mop_adj_pair(n, prow, neg);
for (col = 0; col < oprsz; ) {
uint16_t pcol = pm[H2(col >> 4)];
do {
if (prow & pcol & 0b0101) {
uint32_t *a = vza_row + H1_4(col);
uint32_t m = *(uint32_t *)(vzm + H1_4(col));
m = f16mop_adj_pair(m, pcol, 0);
*a = f16_dotadd(*a, n, m, &fpst_std, &fpst_odd);
col += 4;
pcol >>= 4;
}
} while (col & 15);
}
row += 4;
prow >>= 4;
} while (row & 15);
}
}
void HELPER(sme_bfmopa)(void *vza, void *vzn, void *vzm, void *vpn,
void *vpm, uint32_t desc)
{
intptr_t row, col, oprsz = simd_maxsz(desc);
uint32_t neg = simd_data(desc) * 0x80008000u;
uint16_t *pn = vpn, *pm = vpm;
for (row = 0; row < oprsz; ) {
uint16_t prow = pn[H2(row >> 4)];
do {
void *vza_row = vza + tile_vslice_offset(row);
uint32_t n = *(uint32_t *)(vzn + H1_4(row));
n = f16mop_adj_pair(n, prow, neg);
for (col = 0; col < oprsz; ) {
uint16_t pcol = pm[H2(col >> 4)];
do {
if (prow & pcol & 0b0101) {
uint32_t *a = vza_row + H1_4(col);
uint32_t m = *(uint32_t *)(vzm + H1_4(col));
m = f16mop_adj_pair(m, pcol, 0);
*a = bfdotadd(*a, n, m);
col += 4;
pcol >>= 4;
}
} while (col & 15);
}
row += 4;
prow >>= 4;
} while (row & 15);
}
}
typedef uint64_t IMOPFn(uint64_t, uint64_t, uint64_t, uint8_t, bool);
static inline void do_imopa(uint64_t *za, uint64_t *zn, uint64_t *zm,
uint8_t *pn, uint8_t *pm,
uint32_t desc, IMOPFn *fn)
{
intptr_t row, col, oprsz = simd_oprsz(desc) / 8;
bool neg = simd_data(desc);
for (row = 0; row < oprsz; ++row) {
uint8_t pa = pn[H1(row)];
uint64_t *za_row = &za[tile_vslice_index(row)];
uint64_t n = zn[row];
for (col = 0; col < oprsz; ++col) {
uint8_t pb = pm[H1(col)];
uint64_t *a = &za_row[col];
*a = fn(n, zm[col], *a, pa & pb, neg);
}
}
}
#define DEF_IMOP_32(NAME, NTYPE, MTYPE) \
static uint64_t NAME(uint64_t n, uint64_t m, uint64_t a, uint8_t p, bool neg) \
{ \
uint32_t sum0 = 0, sum1 = 0; \
/* Apply P to N as a mask, making the inactive elements 0. */ \
n &= expand_pred_b(p); \
sum0 += (NTYPE)(n >> 0) * (MTYPE)(m >> 0); \
sum0 += (NTYPE)(n >> 8) * (MTYPE)(m >> 8); \
sum0 += (NTYPE)(n >> 16) * (MTYPE)(m >> 16); \
sum0 += (NTYPE)(n >> 24) * (MTYPE)(m >> 24); \
sum1 += (NTYPE)(n >> 32) * (MTYPE)(m >> 32); \
sum1 += (NTYPE)(n >> 40) * (MTYPE)(m >> 40); \
sum1 += (NTYPE)(n >> 48) * (MTYPE)(m >> 48); \
sum1 += (NTYPE)(n >> 56) * (MTYPE)(m >> 56); \
if (neg) { \
sum0 = (uint32_t)a - sum0, sum1 = (uint32_t)(a >> 32) - sum1; \
} else { \
sum0 = (uint32_t)a + sum0, sum1 = (uint32_t)(a >> 32) + sum1; \
} \
return ((uint64_t)sum1 << 32) | sum0; \
}
#define DEF_IMOP_64(NAME, NTYPE, MTYPE) \
static uint64_t NAME(uint64_t n, uint64_t m, uint64_t a, uint8_t p, bool neg) \
{ \
uint64_t sum = 0; \
/* Apply P to N as a mask, making the inactive elements 0. */ \
n &= expand_pred_h(p); \
sum += (NTYPE)(n >> 0) * (MTYPE)(m >> 0); \
sum += (NTYPE)(n >> 16) * (MTYPE)(m >> 16); \
sum += (NTYPE)(n >> 32) * (MTYPE)(m >> 32); \
sum += (NTYPE)(n >> 48) * (MTYPE)(m >> 48); \
return neg ? a - sum : a + sum; \
}
DEF_IMOP_32(smopa_s, int8_t, int8_t)
DEF_IMOP_32(umopa_s, uint8_t, uint8_t)
DEF_IMOP_32(sumopa_s, int8_t, uint8_t)
DEF_IMOP_32(usmopa_s, uint8_t, int8_t)
DEF_IMOP_64(smopa_d, int16_t, int16_t)
DEF_IMOP_64(umopa_d, uint16_t, uint16_t)
DEF_IMOP_64(sumopa_d, int16_t, uint16_t)
DEF_IMOP_64(usmopa_d, uint16_t, int16_t)
#define DEF_IMOPH(NAME) \
void HELPER(sme_##NAME)(void *vza, void *vzn, void *vzm, void *vpn, \
void *vpm, uint32_t desc) \
{ do_imopa(vza, vzn, vzm, vpn, vpm, desc, NAME); }
DEF_IMOPH(smopa_s)
DEF_IMOPH(umopa_s)
DEF_IMOPH(sumopa_s)
DEF_IMOPH(usmopa_s)
DEF_IMOPH(smopa_d)
DEF_IMOPH(umopa_d)
DEF_IMOPH(sumopa_d)
DEF_IMOPH(usmopa_d)
Reference in New Issue View Git Blame Copy Permalink