From: Paolo Bonzini <pbonzini@redhat.com>
To: qemu-devel@nongnu.org
Cc: richard.henderson@linaro.org, paul@nowt.org
Subject: [PATCH v2 16/37] target/i386: provide 3-operand versions of unary scalar helpers
Date: Tue, 20 Sep 2022 19:24:46 +0200 [thread overview]
Message-ID: <20220920172507.95568-17-pbonzini@redhat.com> (raw)
In-Reply-To: <20220920172507.95568-1-pbonzini@redhat.com>
Compared to Paul's implementation, the new decoder will use a different approach
to implement AVX's merging of dst with src1 on scalar operations. Adjust the
old SSE decoder to be compatible with new-style helpers.
The affected instructions are CVTSx2Sx, ROUNDSx, RSQRTSx, SQRTSx, RCPSx.
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Signed-off-by: Paolo Bonzini <pbonzini@redhat.com>
---
target/i386/ops_sse.h | 48 ++++++++++++++++++++++++++++++------
| 16 ++++++------
target/i386/tcg/translate.c | 22 ++++++++++-------
3 files changed, 61 insertions(+), 25 deletions(-)
diff --git a/target/i386/ops_sse.h b/target/i386/ops_sse.h
index ddedc46f36..8bb7293975 100644
--- a/target/i386/ops_sse.h
+++ b/target/i386/ops_sse.h
@@ -617,14 +617,22 @@ void glue(helper_sqrtpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
}
#if SHIFT == 1
-void helper_sqrtss(CPUX86State *env, Reg *d, Reg *s)
+void helper_sqrtss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
{
+ int i;
d->ZMM_S(0) = float32_sqrt(s->ZMM_S(0), &env->sse_status);
+ for (i = 1; i < 2 << SHIFT; i++) {
+ d->ZMM_L(i) = v->ZMM_L(i);
+ }
}
-void helper_sqrtsd(CPUX86State *env, Reg *d, Reg *s)
+void helper_sqrtsd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
{
+ int i;
d->ZMM_D(0) = float64_sqrt(s->ZMM_D(0), &env->sse_status);
+ for (i = 1; i < 1 << SHIFT; i++) {
+ d->ZMM_Q(i) = v->ZMM_Q(i);
+ }
}
#endif
@@ -649,14 +657,22 @@ void glue(helper_cvtpd2ps, SUFFIX)(CPUX86State *env, Reg *d, Reg *s)
}
#if SHIFT == 1
-void helper_cvtss2sd(CPUX86State *env, Reg *d, Reg *s)
+void helper_cvtss2sd(CPUX86State *env, Reg *d, Reg *v, Reg *s)
{
+ int i;
d->ZMM_D(0) = float32_to_float64(s->ZMM_S(0), &env->sse_status);
+ for (i = 1; i < 1 << SHIFT; i++) {
+ d->ZMM_Q(i) = v->ZMM_Q(i);
+ }
}
-void helper_cvtsd2ss(CPUX86State *env, Reg *d, Reg *s)
+void helper_cvtsd2ss(CPUX86State *env, Reg *d, Reg *v, Reg *s)
{
+ int i;
d->ZMM_S(0) = float64_to_float32(s->ZMM_D(0), &env->sse_status);
+ for (i = 1; i < 2 << SHIFT; i++) {
+ d->ZMM_L(i) = v->ZMM_L(i);
+ }
}
#endif
@@ -876,13 +892,17 @@ void glue(helper_rsqrtps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
}
#if SHIFT == 1
-void helper_rsqrtss(CPUX86State *env, ZMMReg *d, ZMMReg *s)
+void helper_rsqrtss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
{
uint8_t old_flags = get_float_exception_flags(&env->sse_status);
+ int i;
d->ZMM_S(0) = float32_div(float32_one,
float32_sqrt(s->ZMM_S(0), &env->sse_status),
&env->sse_status);
set_float_exception_flags(old_flags, &env->sse_status);
+ for (i = 1; i < 2 << SHIFT; i++) {
+ d->ZMM_L(i) = v->ZMM_L(i);
+ }
}
#endif
@@ -897,10 +917,14 @@ void glue(helper_rcpps, SUFFIX)(CPUX86State *env, ZMMReg *d, ZMMReg *s)
}
#if SHIFT == 1
-void helper_rcpss(CPUX86State *env, ZMMReg *d, ZMMReg *s)
+void helper_rcpss(CPUX86State *env, ZMMReg *d, ZMMReg *v, ZMMReg *s)
{
uint8_t old_flags = get_float_exception_flags(&env->sse_status);
+ int i;
d->ZMM_S(0) = float32_div(float32_one, s->ZMM_S(0), &env->sse_status);
+ for (i = 1; i < 2 << SHIFT; i++) {
+ d->ZMM_L(i) = v->ZMM_L(i);
+ }
set_float_exception_flags(old_flags, &env->sse_status);
}
#endif
@@ -1798,11 +1822,12 @@ void glue(helper_roundpd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
}
#if SHIFT == 1
-void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
+void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
uint32_t mode)
{
uint8_t old_flags = get_float_exception_flags(&env->sse_status);
signed char prev_rounding_mode;
+ int i;
prev_rounding_mode = env->sse_status.float_rounding_mode;
if (!(mode & (1 << 2))) {
@@ -1823,6 +1848,9 @@ void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
}
d->ZMM_S(0) = float32_round_to_int(s->ZMM_S(0), &env->sse_status);
+ for (i = 1; i < 2 << SHIFT; i++) {
+ d->ZMM_L(i) = v->ZMM_L(i);
+ }
if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
@@ -1832,11 +1860,12 @@ void glue(helper_roundss, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
env->sse_status.float_rounding_mode = prev_rounding_mode;
}
-void glue(helper_roundsd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
+void glue(helper_roundsd, SUFFIX)(CPUX86State *env, Reg *d, Reg *v, Reg *s,
uint32_t mode)
{
uint8_t old_flags = get_float_exception_flags(&env->sse_status);
signed char prev_rounding_mode;
+ int i;
prev_rounding_mode = env->sse_status.float_rounding_mode;
if (!(mode & (1 << 2))) {
@@ -1857,6 +1886,9 @@ void glue(helper_roundsd, SUFFIX)(CPUX86State *env, Reg *d, Reg *s,
}
d->ZMM_D(0) = float64_round_to_int(s->ZMM_D(0), &env->sse_status);
+ for (i = 1; i < 1 << SHIFT; i++) {
+ d->ZMM_Q(i) = v->ZMM_Q(i);
+ }
if (mode & (1 << 3) && !(old_flags & float_flag_inexact)) {
set_float_exception_flags(get_float_exception_flags(&env->sse_status) &
--git a/target/i386/ops_sse_header.h b/target/i386/ops_sse_header.h
index b60fe2f0d4..32ffa8445b 100644
--- a/target/i386/ops_sse_header.h
+++ b/target/i386/ops_sse_header.h
@@ -148,8 +148,8 @@ DEF_HELPER_3(glue(pshufhw, SUFFIX), void, Reg, Reg, int)
DEF_HELPER_4(name ## sd, void, env, Reg, Reg, Reg)
#define SSE_HELPER_S3(name) \
SSE_HELPER_P3(name) \
- DEF_HELPER_3(name ## ss, void, env, Reg, Reg) \
- DEF_HELPER_3(name ## sd, void, env, Reg, Reg)
+ DEF_HELPER_4(name ## ss, void, env, Reg, Reg, Reg) \
+ DEF_HELPER_4(name ## sd, void, env, Reg, Reg, Reg)
#else
#define SSE_HELPER_S4(name, ...) SSE_HELPER_P4(name)
#define SSE_HELPER_S3(name, ...) SSE_HELPER_P3(name)
@@ -179,8 +179,8 @@ DEF_HELPER_3(glue(cvttps2dq, SUFFIX), void, env, ZMMReg, ZMMReg)
DEF_HELPER_3(glue(cvttpd2dq, SUFFIX), void, env, ZMMReg, ZMMReg)
#if SHIFT == 1
-DEF_HELPER_3(cvtss2sd, void, env, Reg, Reg)
-DEF_HELPER_3(cvtsd2ss, void, env, Reg, Reg)
+DEF_HELPER_4(cvtss2sd, void, env, Reg, Reg, Reg)
+DEF_HELPER_4(cvtsd2ss, void, env, Reg, Reg, Reg)
DEF_HELPER_3(cvtpi2ps, void, env, ZMMReg, MMXReg)
DEF_HELPER_3(cvtpi2pd, void, env, ZMMReg, MMXReg)
DEF_HELPER_3(cvtsi2ss, void, env, ZMMReg, i32)
@@ -214,8 +214,8 @@ DEF_HELPER_3(glue(rsqrtps, SUFFIX), void, env, ZMMReg, ZMMReg)
DEF_HELPER_3(glue(rcpps, SUFFIX), void, env, ZMMReg, ZMMReg)
#if SHIFT == 1
-DEF_HELPER_3(rsqrtss, void, env, ZMMReg, ZMMReg)
-DEF_HELPER_3(rcpss, void, env, ZMMReg, ZMMReg)
+DEF_HELPER_4(rsqrtss, void, env, ZMMReg, ZMMReg, ZMMReg)
+DEF_HELPER_4(rcpss, void, env, ZMMReg, ZMMReg, ZMMReg)
DEF_HELPER_3(extrq_r, void, env, ZMMReg, ZMMReg)
DEF_HELPER_4(extrq_i, void, env, ZMMReg, int, int)
DEF_HELPER_3(insertq_r, void, env, ZMMReg, ZMMReg)
@@ -342,8 +342,8 @@ DEF_HELPER_3(glue(phminposuw, SUFFIX), void, env, Reg, Reg)
DEF_HELPER_4(glue(roundps, SUFFIX), void, env, Reg, Reg, i32)
DEF_HELPER_4(glue(roundpd, SUFFIX), void, env, Reg, Reg, i32)
#if SHIFT == 1
-DEF_HELPER_4(roundss_xmm, void, env, Reg, Reg, i32)
-DEF_HELPER_4(roundsd_xmm, void, env, Reg, Reg, i32)
+DEF_HELPER_5(roundss_xmm, void, env, Reg, Reg, Reg, i32)
+DEF_HELPER_5(roundsd_xmm, void, env, Reg, Reg, Reg, i32)
#endif
DEF_HELPER_5(glue(blendps, SUFFIX), void, env, Reg, Reg, Reg, i32)
DEF_HELPER_5(glue(blendpd, SUFFIX), void, env, Reg, Reg, Reg, i32)
diff --git a/target/i386/tcg/translate.c b/target/i386/tcg/translate.c
index d20835f96d..a974fb28f9 100644
--- a/target/i386/tcg/translate.c
+++ b/target/i386/tcg/translate.c
@@ -2930,6 +2930,9 @@ static bool first = true; static unsigned long limit;
#define SSE_OP(sname, dname, op, flags) OP(op, flags, \
gen_helper_##sname##_xmm, gen_helper_##dname##_xmm, NULL, NULL)
+#define SSE_OP_UNARY(a, b, c, d) \
+ {SSE_OPF_SCALAR | SSE_OPF_V0, {{.op1 = a}, {.op1 = b}, {.op2 = c}, {.op2 = d} } }
+
typedef union SSEFuncs {
SSEFunc_0_epp op1;
SSEFunc_0_ppi op1i;
@@ -2972,12 +2975,12 @@ static const struct SSEOpHelper_table1 sse_op_table1[256] = {
[0x2f] = OP(op1, SSE_OPF_CMP | SSE_OPF_SCALAR | SSE_OPF_V0,
gen_helper_comiss, gen_helper_comisd, NULL, NULL),
[0x50] = SSE_SPECIAL, /* movmskps, movmskpd */
- [0x51] = OP(op1, SSE_OPF_SCALAR | SSE_OPF_V0,
+ [0x51] = SSE_OP_UNARY(
gen_helper_sqrtps_xmm, gen_helper_sqrtpd_xmm,
gen_helper_sqrtss, gen_helper_sqrtsd),
- [0x52] = OP(op1, SSE_OPF_SCALAR | SSE_OPF_V0,
+ [0x52] = SSE_OP_UNARY(
gen_helper_rsqrtps_xmm, NULL, gen_helper_rsqrtss, NULL),
- [0x53] = OP(op1, SSE_OPF_SCALAR | SSE_OPF_V0,
+ [0x53] = SSE_OP_UNARY(
gen_helper_rcpps_xmm, NULL, gen_helper_rcpss, NULL),
[0x54] = SSE_OP(pand, pand, op2, 0), /* andps, andpd */
[0x55] = SSE_OP(pandn, pandn, op2, 0), /* andnps, andnpd */
@@ -2985,9 +2988,9 @@ static const struct SSEOpHelper_table1 sse_op_table1[256] = {
[0x57] = SSE_OP(pxor, pxor, op2, 0), /* xorps, xorpd */
[0x58] = SSE_FOP(add),
[0x59] = SSE_FOP(mul),
- [0x5a] = OP(op1, SSE_OPF_SCALAR | SSE_OPF_V0,
- gen_helper_cvtps2pd_xmm, gen_helper_cvtpd2ps_xmm,
- gen_helper_cvtss2sd, gen_helper_cvtsd2ss),
+ [0x5a] = SSE_OP_UNARY(
+ gen_helper_cvtps2pd_xmm, gen_helper_cvtpd2ps_xmm,
+ gen_helper_cvtss2sd, gen_helper_cvtsd2ss),
[0x5b] = OP(op1, SSE_OPF_V0,
gen_helper_cvtdq2ps_xmm, gen_helper_cvtps2dq_xmm,
gen_helper_cvttps2dq_xmm, NULL),
@@ -3283,8 +3286,8 @@ static const struct SSEOpHelper_table6 sse_op_table6[256] = {
static const struct SSEOpHelper_table7 sse_op_table7[256] = {
[0x08] = UNARY_OP(roundps, SSE41, 0),
[0x09] = UNARY_OP(roundpd, SSE41, 0),
- [0x0a] = UNARY_OP(roundss, SSE41, SSE_OPF_SCALAR),
- [0x0b] = UNARY_OP(roundsd, SSE41, SSE_OPF_SCALAR),
+ [0x0a] = BINARY_OP(roundss, SSE41, SSE_OPF_SCALAR),
+ [0x0b] = BINARY_OP(roundsd, SSE41, SSE_OPF_SCALAR),
[0x0c] = BINARY_OP(blendps, SSE41, 0),
[0x0d] = BINARY_OP(blendpd, SSE41, 0),
[0x0e] = BINARY_OP(pblendw, SSE41, SSE_OPF_MMX),
@@ -4560,7 +4563,8 @@ static void gen_sse(CPUX86State *env, DisasContext *s, int b,
tcg_gen_addi_ptr(s->ptr0, cpu_env, op1_offset);
tcg_gen_addi_ptr(s->ptr1, cpu_env, op2_offset);
- if (sse_op_flags & SSE_OPF_V0) {
+ if ((sse_op_flags & SSE_OPF_V0) &&
+ !((sse_op_flags & SSE_OPF_SCALAR) && b1 >= 2)) {
if (sse_op_flags & SSE_OPF_SHUF) {
val = x86_ldub_code(env, s);
sse_op_fn.op1i(s->ptr0, s->ptr1, tcg_const_i32(val));
--
2.37.2
next prev parent reply other threads:[~2022-09-20 22:43 UTC|newest]
Thread overview: 61+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-09-20 17:24 [PATCH v2 00/37] target/i386: new decoder + AVX implementation Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 01/37] target/i386: Define XMMReg and access macros, align ZMM registers Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 02/37] target/i386: make ldo/sto operations consistent with ldq Paolo Bonzini
2022-09-24 15:00 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 03/37] target/i386: REPZ and REPNZ are mutually exclusive Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 04/37] target/i386: introduce insn_get_addr Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 05/37] target/i386: add core of new i386 decoder Paolo Bonzini
2022-09-24 15:09 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 06/37] target/i386: add ALU load/writeback core Paolo Bonzini
2022-09-24 15:13 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 07/37] target/i386: add CPUID[EAX=7, ECX=0].ECX to DisasContext Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 08/37] target/i386: add CPUID feature checks to new decoder Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 09/37] target/i386: add AVX_EN hflag Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 10/37] target/i386: validate VEX prefixes via the instructions' exception classes Paolo Bonzini
2022-09-24 20:19 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 11/37] target/i386: validate SSE prefixes directly in the decoding table Paolo Bonzini
2022-09-24 20:23 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 12/37] target/i386: move scalar 0F 38 and 0F 3A instruction to new decoder Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 13/37] target/i386: Prepare ops_sse_header.h for 256 bit AVX Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 14/37] target/i386: extend helpers to support VEX.V 3- and 4- operand encodings Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 15/37] target/i386: support operand merging in binary scalar helpers Paolo Bonzini
2022-09-20 17:24 ` Paolo Bonzini [this message]
2022-09-20 17:24 ` [PATCH v2 17/37] target/i386: implement additional AVX comparison operators Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 18/37] target/i386: Introduce 256-bit vector helpers Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 19/37] target/i386: reimplement 0x0f 0x60-0x6f, add AVX Paolo Bonzini
2022-09-24 20:31 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 20/37] target/i386: reimplement 0x0f 0xd8-0xdf, 0xe8-0xef, 0xf8-0xff, " Paolo Bonzini
2022-09-24 20:32 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 21/37] target/i386: reimplement 0x0f 0x50-0x5f, " Paolo Bonzini
2022-09-24 20:37 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 22/37] target/i386: reimplement 0x0f 0x78-0x7f, " Paolo Bonzini
2022-09-24 20:43 ` Richard Henderson
2022-09-26 7:24 ` Paolo Bonzini
2022-09-26 7:46 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 23/37] target/i386: reimplement 0x0f 0x70-0x77, " Paolo Bonzini
2022-09-24 20:53 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 24/37] target/i386: reimplement 0x0f 0xd0-0xd7, 0xe0-0xe7, 0xf0-0xf7, " Paolo Bonzini
2022-09-24 20:56 ` Richard Henderson
2022-09-26 7:56 ` Paolo Bonzini
2022-09-20 17:24 ` [PATCH v2 25/37] target/i386: clarify (un)signedness of immediates from 0F3Ah opcodes Paolo Bonzini
2022-09-24 20:57 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 26/37] target/i386: reimplement 0x0f 0x3a, add AVX Paolo Bonzini
2022-09-24 21:02 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 27/37] target/i386: Use tcg gvec ops for pmovmskb Paolo Bonzini
2022-09-24 21:08 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 28/37] target/i386: reimplement 0x0f 0x38, add AVX Paolo Bonzini
2022-09-24 21:14 ` Richard Henderson
2022-09-20 17:24 ` [PATCH v2 29/37] target/i386: reimplement 0x0f 0xc2, 0xc4-0xc6, " Paolo Bonzini
2022-09-20 17:25 ` [PATCH v2 30/37] target/i386: reimplement 0x0f 0x10-0x17, " Paolo Bonzini
2022-09-24 21:16 ` Richard Henderson
2022-09-20 17:25 ` [PATCH v2 31/37] target/i386: reimplement 0x0f 0x28-0x2f, " Paolo Bonzini
2022-09-24 21:18 ` Richard Henderson
2022-09-20 17:25 ` [PATCH v2 32/37] target/i386: implement XSAVE and XRSTOR of AVX registers Paolo Bonzini
2022-09-24 21:19 ` Richard Henderson
2022-09-20 17:25 ` [PATCH v2 33/37] target/i386: implement VLDMXCSR/VSTMXCSR Paolo Bonzini
2022-09-24 21:20 ` Richard Henderson
2022-09-20 17:25 ` [PATCH v2 34/37] target/i386: Enable AVX cpuid bits when using TCG Paolo Bonzini
2022-09-20 17:25 ` [PATCH v2 35/37] tests/tcg: extend SSE tests to AVX Paolo Bonzini
2022-09-20 17:25 ` [PATCH v2 36/37] target/i386: move 3DNow to the new decoder Paolo Bonzini
2022-09-24 21:24 ` Richard Henderson
2022-09-20 17:25 ` [PATCH v2 37/37] target/i386: remove old SSE decoder Paolo Bonzini
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20220920172507.95568-17-pbonzini@redhat.com \
--to=pbonzini@redhat.com \
--cc=paul@nowt.org \
--cc=qemu-devel@nongnu.org \
--cc=richard.henderson@linaro.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).