1
The following changes since commit ad1b4ec39caa5b3f17cbd8160283a03a3dcfe2ae:
1
A few last patches to go in for rc3...
2
2
3
Merge remote-tracking branch 'remotes/kraxel/tags/input-20180515-pull-request' into staging (2018-05-15 12:50:06 +0100)
3
The following changes since commit c1e90def01bdb8fcbdbebd9d1eaa8e4827ece620:
4
5
Merge remote-tracking branch 'remotes/pmaydell/tags/pull-target-arm-20210412' into staging (2021-04-12 12:12:09 +0100)
4
6
5
are available in the Git repository at:
7
are available in the Git repository at:
6
8
7
git://git.linaro.org/people/pmaydell/qemu-arm.git tags/pull-target-arm-20180515
9
https://git.linaro.org/people/pmaydell/qemu-arm.git tags/pull-target-arm-20210413
8
10
9
for you to fetch changes up to ae7651804748c6b479d5ae09aeac4edb9c44f76e:
11
for you to fetch changes up to 2d18b4ca023ca1a3aee18064251d6e6e1084f3eb:
10
12
11
tcg: Optionally log FPU state in TCG -d cpu logging (2018-05-15 14:58:44 +0100)
13
sphinx: qapidoc: Wrap "If" section body in a paragraph node (2021-04-13 10:14:58 +0100)
12
14
13
----------------------------------------------------------------
15
----------------------------------------------------------------
14
target-arm queue:
16
target-arm queue:
15
* Fix coverity nit in int_to_float code
17
* Fix MPC setting for AN524 SRAM block
16
* Don't set Invalid for float-to-int(MAXINT)
18
* sphinx: qapidoc: Wrap "If" section body in a paragraph node
17
* Fix fp_status_f16 tininess before rounding
18
* Add various missing insns from the v8.2-FP16 extension
19
* Fix sqrt_f16 exception raising
20
* sdcard: Correct CRC16 offset in sd_function_switch()
21
* tcg: Optionally log FPU state in TCG -d cpu logging
22
19
23
----------------------------------------------------------------
20
----------------------------------------------------------------
24
Alex Bennée (5):
21
John Snow (1):
25
fpu/softfloat: int_to_float ensure r fully initialised
22
sphinx: qapidoc: Wrap "If" section body in a paragraph node
26
target/arm: Implement FCMP for fp16
27
target/arm: Implement FCSEL for fp16
28
target/arm: Implement FMOV (immediate) for fp16
29
target/arm: Fix sqrt_f16 exception raising
30
23
31
Peter Maydell (3):
24
Peter Maydell (2):
32
fpu/softfloat: Don't set Invalid for float-to-int(MAXINT)
25
hw/arm/mps2-tz: Fix MPC setting for AN524 SRAM block
33
target/arm: Fix fp_status_f16 tininess before rounding
26
hw/arm/mps2-tz: Assert if more than one RAM is attached to an MPC
34
tcg: Optionally log FPU state in TCG -d cpu logging
35
27
36
Philippe Mathieu-Daudé (1):
28
docs/sphinx/qapidoc.py | 4 +++-
37
sdcard: Correct CRC16 offset in sd_function_switch()
29
hw/arm/mps2-tz.c | 10 +++++++---
30
2 files changed, 10 insertions(+), 4 deletions(-)
38
31
39
Richard Henderson (7):
40
target/arm: Implement FMOV (general) for fp16
41
target/arm: Early exit after unallocated_encoding in disas_fp_int_conv
42
target/arm: Implement FCVT (scalar, integer) for fp16
43
target/arm: Implement FCVT (scalar, fixed-point) for fp16
44
target/arm: Introduce and use read_fp_hreg
45
target/arm: Implement FP data-processing (2 source) for fp16
46
target/arm: Implement FP data-processing (3 source) for fp16
47
48
include/qemu/log.h | 1 +
49
target/arm/helper-a64.h | 2 +
50
target/arm/helper.h | 6 +
51
accel/tcg/cpu-exec.c | 9 +-
52
fpu/softfloat.c | 6 +-
53
hw/sd/sd.c | 2 +-
54
target/arm/cpu.c | 2 +
55
target/arm/helper-a64.c | 10 ++
56
target/arm/helper.c | 38 +++-
57
target/arm/translate-a64.c | 421 ++++++++++++++++++++++++++++++++++++++-------
58
util/log.c | 2 +
59
11 files changed, 428 insertions(+), 71 deletions(-)
60
diff view generated by jsdifflib
1
From: Philippe Mathieu-Daudé <f4bug@amsat.org>
1
The AN524 has three MPCs: one for the BRAM, one for the QSPI flash,
2
and one for the DDR. We incorrectly set the .mpc field in the
3
RAMInfo struct for the SRAM block to 1, giving it the same MPC we are
4
using for the QSPI. The effect of this was that the QSPI didn't get
5
mapped into the system address space at all, via an MPC or otherwise,
6
and guest programs which tried to read from the QSPI would get a bus
7
error. Correct the SRAM RAMInfo to indicate that it does not have an
8
associated MPC.
2
9
3
Per the Physical Layer Simplified Spec. "4.3.10.4 Switch Function Status":
10
Fixes: 25ff112a8cc ("hw/arm/mps2-tz: Add new mps3-an524 board")
4
5
The block length is predefined to 512 bits
6
7
and "4.10.2 SD Status":
8
9
The SD Status contains status bits that are related to the SD Memory Card
10
proprietary features and may be used for future application-specific usage.
11
The size of the SD Status is one data block of 512 bit. The content of this
12
register is transmitted to the Host over the DAT bus along with a 16-bit CRC.
13
14
Thus the 16-bit CRC goes at offset 64.
15
16
Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
17
Message-id: 20180509060104.4458-3-f4bug@amsat.org
18
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
19
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
12
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
13
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
14
Message-id: 20210409150527.15053-2-peter.maydell@linaro.org
20
---
15
---
21
hw/sd/sd.c | 2 +-
16
hw/arm/mps2-tz.c | 2 +-
22
1 file changed, 1 insertion(+), 1 deletion(-)
17
1 file changed, 1 insertion(+), 1 deletion(-)
23
18
24
diff --git a/hw/sd/sd.c b/hw/sd/sd.c
19
diff --git a/hw/arm/mps2-tz.c b/hw/arm/mps2-tz.c
25
index XXXXXXX..XXXXXXX 100644
20
index XXXXXXX..XXXXXXX 100644
26
--- a/hw/sd/sd.c
21
--- a/hw/arm/mps2-tz.c
27
+++ b/hw/sd/sd.c
22
+++ b/hw/arm/mps2-tz.c
28
@@ -XXX,XX +XXX,XX @@ static void sd_function_switch(SDState *sd, uint32_t arg)
23
@@ -XXX,XX +XXX,XX @@ static const RAMInfo an524_raminfo[] = { {
29
sd->data[14 + (i >> 1)] = new_func << ((i * 4) & 4);
24
.name = "sram",
30
}
25
.base = 0x20000000,
31
memset(&sd->data[17], 0, 47);
26
.size = 32 * 4 * KiB,
32
- stw_be_p(sd->data + 65, sd_crc16(sd->data, 64));
27
- .mpc = 1,
33
+ stw_be_p(sd->data + 64, sd_crc16(sd->data, 64));
28
+ .mpc = -1,
34
}
29
.mrindex = 1,
35
30
}, {
36
static inline bool sd_wp_addr(SDState *sd, uint64_t addr)
31
/* We don't model QSPI flash yet; for now expose it as simple ROM */
37
--
32
--
38
2.17.0
33
2.20.1
39
34
40
35
diff view generated by jsdifflib
1
Usually the logging of the CPU state produced by -d cpu is sufficient
1
Each board in mps2-tz.c specifies a RAMInfo[] array providing
2
to diagnose problems, but sometimes you want to see the state of
2
information about each RAM in the board. The .mpc field of the
3
the floating point registers as well. We don't want to enable that
3
RAMInfo struct specifies which MPC, if any, the RAM is attached to.
4
by default as it adds a lot of extra data to the log; instead,
4
We already assert if the array doesn't have any entry for an MPC, but
5
allow it to be optionally enabled via -d fpu.
5
we don't diagnose the error of using the same MPC number twice (which
6
is quite easy to do by accident if copy-and-pasting structure
7
entries).
8
9
Enhance find_raminfo_for_mpc() so that it detects multiple entries
10
for the MPC as well as missing entries.
6
11
7
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
12
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
8
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
13
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
9
Message-id: 20180510130024.31678-1-peter.maydell@linaro.org
14
Reviewed-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
15
Message-id: 20210409150527.15053-3-peter.maydell@linaro.org
10
---
16
---
11
include/qemu/log.h | 1 +
17
hw/arm/mps2-tz.c | 8 ++++++--
12
accel/tcg/cpu-exec.c | 9 ++++++---
18
1 file changed, 6 insertions(+), 2 deletions(-)
13
util/log.c | 2 ++
14
3 files changed, 9 insertions(+), 3 deletions(-)
15
19
16
diff --git a/include/qemu/log.h b/include/qemu/log.h
20
diff --git a/hw/arm/mps2-tz.c b/hw/arm/mps2-tz.c
17
index XXXXXXX..XXXXXXX 100644
21
index XXXXXXX..XXXXXXX 100644
18
--- a/include/qemu/log.h
22
--- a/hw/arm/mps2-tz.c
19
+++ b/include/qemu/log.h
23
+++ b/hw/arm/mps2-tz.c
20
@@ -XXX,XX +XXX,XX @@ static inline bool qemu_log_separate(void)
24
@@ -XXX,XX +XXX,XX @@ static const RAMInfo *find_raminfo_for_mpc(MPS2TZMachineState *mms, int mpc)
21
#define CPU_LOG_PAGE (1 << 14)
25
{
22
/* LOG_TRACE (1 << 15) is defined in log-for-trace.h */
26
MPS2TZMachineClass *mmc = MPS2TZ_MACHINE_GET_CLASS(mms);
23
#define CPU_LOG_TB_OP_IND (1 << 16)
27
const RAMInfo *p;
24
+#define CPU_LOG_TB_FPU (1 << 17)
28
+ const RAMInfo *found = NULL;
25
29
26
/* Lock output for a series of related logs. Since this is not needed
30
for (p = mmc->raminfo; p->name; p++) {
27
* for a single qemu_log / qemu_log_mask / qemu_log_mask_and_addr, we
31
if (p->mpc == mpc && !(p->flags & IS_ALIAS)) {
28
diff --git a/accel/tcg/cpu-exec.c b/accel/tcg/cpu-exec.c
32
- return p;
29
index XXXXXXX..XXXXXXX 100644
33
+ /* There should only be one entry in the array for this MPC */
30
--- a/accel/tcg/cpu-exec.c
34
+ g_assert(!found);
31
+++ b/accel/tcg/cpu-exec.c
35
+ found = p;
32
@@ -XXX,XX +XXX,XX @@ static inline tcg_target_ulong cpu_tb_exec(CPUState *cpu, TranslationBlock *itb)
36
}
33
if (qemu_loglevel_mask(CPU_LOG_TB_CPU)
34
&& qemu_log_in_addr_range(itb->pc)) {
35
qemu_log_lock();
36
+ int flags = 0;
37
+ if (qemu_loglevel_mask(CPU_LOG_TB_FPU)) {
38
+ flags |= CPU_DUMP_FPU;
39
+ }
40
#if defined(TARGET_I386)
41
- log_cpu_state(cpu, CPU_DUMP_CCOP);
42
-#else
43
- log_cpu_state(cpu, 0);
44
+ flags |= CPU_DUMP_CCOP;
45
#endif
46
+ log_cpu_state(cpu, flags);
47
qemu_log_unlock();
48
}
37
}
49
#endif /* DEBUG_DISAS */
38
/* if raminfo array doesn't have an entry for each MPC this is a bug */
50
diff --git a/util/log.c b/util/log.c
39
- g_assert_not_reached();
51
index XXXXXXX..XXXXXXX 100644
40
+ assert(found);
52
--- a/util/log.c
41
+ return found;
53
+++ b/util/log.c
42
}
54
@@ -XXX,XX +XXX,XX @@ const QEMULogItem qemu_log_items[] = {
43
55
"show trace before each executed TB (lots of logs)" },
44
static MemoryRegion *mr_for_raminfo(MPS2TZMachineState *mms,
56
{ CPU_LOG_TB_CPU, "cpu",
57
"show CPU registers before entering a TB (lots of logs)" },
58
+ { CPU_LOG_TB_FPU, "fpu",
59
+ "include FPU registers in the 'cpu' logging" },
60
{ CPU_LOG_MMU, "mmu",
61
"log MMU-related activities" },
62
{ CPU_LOG_PCALL, "pcall",
63
--
45
--
64
2.17.0
46
2.20.1
65
47
66
48
diff view generated by jsdifflib
1
From: Alex Bennée <alex.bennee@linaro.org>
1
From: John Snow <jsnow@redhat.com>
2
2
3
Reported by Coverity (CID1390635). We ensure this for uint_to_float
3
These sections need to be wrapped in a block-level element, such as
4
later on so we might as well mirror that.
4
Paragraph in order for them to be rendered into Texinfo correctly.
5
5
6
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
6
Before (e.g.):
7
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
7
8
<section ids="qapidoc-713">
9
<title>If</title>
10
<literal>defined(CONFIG_REPLICATION)</literal>
11
</section>
12
13
became:
14
15
.SS If
16
\fBdefined(CONFIG_REPLICATION)\fP.SS \fBBlockdevOptionsReplication\fP (Object)
17
...
18
19
After:
20
21
<section ids="qapidoc-713">
22
<title>If</title>
23
<paragraph>
24
<literal>defined(CONFIG_REPLICATION)</literal>
25
</paragraph>
26
</section>
27
28
becomes:
29
30
.SS If
31
.sp
32
\fBdefined(CONFIG_REPLICATION)\fP
33
.SS \fBBlockdevOptionsReplication\fP (Object)
34
...
35
36
Reported-by: Markus Armbruster <armbru@redhat.com>
37
Tested-by: Markus Armbruster <armbru@redhat.com>
38
Signed-off-by: John Snow <jsnow@redhat.com>
8
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
39
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
40
Message-id: 20210406141909.1992225-2-jsnow@redhat.com
9
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
41
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
10
---
42
---
11
fpu/softfloat.c | 2 +-
43
docs/sphinx/qapidoc.py | 4 +++-
12
1 file changed, 1 insertion(+), 1 deletion(-)
44
1 file changed, 3 insertions(+), 1 deletion(-)
13
45
14
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
46
diff --git a/docs/sphinx/qapidoc.py b/docs/sphinx/qapidoc.py
15
index XXXXXXX..XXXXXXX 100644
47
index XXXXXXX..XXXXXXX 100644
16
--- a/fpu/softfloat.c
48
--- a/docs/sphinx/qapidoc.py
17
+++ b/fpu/softfloat.c
49
+++ b/docs/sphinx/qapidoc.py
18
@@ -XXX,XX +XXX,XX @@ FLOAT_TO_UINT(64, 64)
50
@@ -XXX,XX +XXX,XX @@ def _nodes_for_if_section(self, ifcond):
19
51
nodelist = []
20
static FloatParts int_to_float(int64_t a, float_status *status)
52
if ifcond:
21
{
53
snode = self._make_section('If')
22
- FloatParts r;
54
- snode += self._nodes_for_ifcond(ifcond, with_if=False)
23
+ FloatParts r = {};
55
+ snode += nodes.paragraph(
24
if (a == 0) {
56
+ '', '', *self._nodes_for_ifcond(ifcond, with_if=False)
25
r.cls = float_class_zero;
57
+ )
26
r.sign = false;
58
nodelist.append(snode)
59
return nodelist
60
27
--
61
--
28
2.17.0
62
2.20.1
29
63
30
64
diff view generated by jsdifflib
Deleted patch
1
In float-to-integer conversion, if the floating point input
2
converts exactly to the largest or smallest integer that
3
fits in to the result type, this is not an overflow.
4
In this situation we were producing the correct result value,
5
but were incorrectly setting the Invalid flag.
6
For example for Arm A64, "FCVTAS w0, d0" on an input of
7
0x41dfffffffc00000 should produce 0x7fffffff and set no flags.
8
1
9
Fix the boundary case to take the right half of the if()
10
statements.
11
12
This fixes a regression from 2.11 introduced by the softfloat
13
refactoring.
14
15
Cc: qemu-stable@nongnu.org
16
Fixes: ab52f973a50
17
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
18
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
19
Message-id: 20180510140141.12120-1-peter.maydell@linaro.org
20
---
21
fpu/softfloat.c | 4 ++--
22
1 file changed, 2 insertions(+), 2 deletions(-)
23
24
diff --git a/fpu/softfloat.c b/fpu/softfloat.c
25
index XXXXXXX..XXXXXXX 100644
26
--- a/fpu/softfloat.c
27
+++ b/fpu/softfloat.c
28
@@ -XXX,XX +XXX,XX @@ static int64_t round_to_int_and_pack(FloatParts in, int rmode,
29
r = UINT64_MAX;
30
}
31
if (p.sign) {
32
- if (r < -(uint64_t) min) {
33
+ if (r <= -(uint64_t) min) {
34
return -r;
35
} else {
36
s->float_exception_flags = orig_flags | float_flag_invalid;
37
return min;
38
}
39
} else {
40
- if (r < max) {
41
+ if (r <= max) {
42
return r;
43
} else {
44
s->float_exception_flags = orig_flags | float_flag_invalid;
45
--
46
2.17.0
47
48
diff view generated by jsdifflib
Deleted patch
1
In commit d81ce0ef2c4f105 we added an extra float_status field
2
fp_status_fp16 for Arm, but forgot to initialize it correctly
3
by setting it to float_tininess_before_rounding. This currently
4
will only cause problems for the new V8_FP16 feature, since the
5
float-to-float conversion code doesn't use it yet. The effect
6
would be that we failed to set the Underflow IEEE exception flag
7
in all the cases where we should.
8
1
9
Add the missing initialization.
10
11
Fixes: d81ce0ef2c4f105
12
Cc: qemu-stable@nongnu.org
13
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
14
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
15
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
16
Message-id: 20180512004311.9299-16-richard.henderson@linaro.org
17
---
18
target/arm/cpu.c | 2 ++
19
1 file changed, 2 insertions(+)
20
21
diff --git a/target/arm/cpu.c b/target/arm/cpu.c
22
index XXXXXXX..XXXXXXX 100644
23
--- a/target/arm/cpu.c
24
+++ b/target/arm/cpu.c
25
@@ -XXX,XX +XXX,XX @@ static void arm_cpu_reset(CPUState *s)
26
&env->vfp.fp_status);
27
set_float_detect_tininess(float_tininess_before_rounding,
28
&env->vfp.standard_fp_status);
29
+ set_float_detect_tininess(float_tininess_before_rounding,
30
+ &env->vfp.fp_status_f16);
31
#ifndef CONFIG_USER_ONLY
32
if (kvm_enabled()) {
33
kvm_arm_reset_vcpu(cpu);
34
--
35
2.17.0
36
37
diff view generated by jsdifflib
Deleted patch
1
From: Richard Henderson <richard.henderson@linaro.org>
2
1
3
Adding the fp16 moves to/from general registers.
4
5
Cc: qemu-stable@nongnu.org
6
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
7
Tested-by: Alex Bennée <alex.bennee@linaro.org>
8
Message-id: 20180512003217.9105-2-richard.henderson@linaro.org
9
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
10
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11
---
12
target/arm/translate-a64.c | 21 +++++++++++++++++++++
13
1 file changed, 21 insertions(+)
14
15
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
16
index XXXXXXX..XXXXXXX 100644
17
--- a/target/arm/translate-a64.c
18
+++ b/target/arm/translate-a64.c
19
@@ -XXX,XX +XXX,XX @@ static void handle_fmov(DisasContext *s, int rd, int rn, int type, bool itof)
20
tcg_gen_st_i64(tcg_rn, cpu_env, fp_reg_hi_offset(s, rd));
21
clear_vec_high(s, true, rd);
22
break;
23
+ case 3:
24
+ /* 16 bit */
25
+ tmp = tcg_temp_new_i64();
26
+ tcg_gen_ext16u_i64(tmp, tcg_rn);
27
+ write_fp_dreg(s, rd, tmp);
28
+ tcg_temp_free_i64(tmp);
29
+ break;
30
+ default:
31
+ g_assert_not_reached();
32
}
33
} else {
34
TCGv_i64 tcg_rd = cpu_reg(s, rd);
35
@@ -XXX,XX +XXX,XX @@ static void handle_fmov(DisasContext *s, int rd, int rn, int type, bool itof)
36
/* 64 bits from top half */
37
tcg_gen_ld_i64(tcg_rd, cpu_env, fp_reg_hi_offset(s, rn));
38
break;
39
+ case 3:
40
+ /* 16 bit */
41
+ tcg_gen_ld16u_i64(tcg_rd, cpu_env, fp_reg_offset(s, rn, MO_16));
42
+ break;
43
+ default:
44
+ g_assert_not_reached();
45
}
46
}
47
}
48
@@ -XXX,XX +XXX,XX @@ static void disas_fp_int_conv(DisasContext *s, uint32_t insn)
49
case 0xa: /* 64 bit */
50
case 0xd: /* 64 bit to top half of quad */
51
break;
52
+ case 0x6: /* 16-bit float, 32-bit int */
53
+ case 0xe: /* 16-bit float, 64-bit int */
54
+ if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
55
+ break;
56
+ }
57
+ /* fallthru */
58
default:
59
/* all other sf/type/rmode combinations are invalid */
60
unallocated_encoding(s);
61
--
62
2.17.0
63
64
diff view generated by jsdifflib
Deleted patch
1
From: Richard Henderson <richard.henderson@linaro.org>
2
1
3
No sense in emitting code after the exception.
4
5
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6
Tested-by: Alex Bennée <alex.bennee@linaro.org>
7
Message-id: 20180512003217.9105-3-richard.henderson@linaro.org
8
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
9
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
10
---
11
target/arm/translate-a64.c | 2 +-
12
1 file changed, 1 insertion(+), 1 deletion(-)
13
14
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
15
index XXXXXXX..XXXXXXX 100644
16
--- a/target/arm/translate-a64.c
17
+++ b/target/arm/translate-a64.c
18
@@ -XXX,XX +XXX,XX @@ static void disas_fp_int_conv(DisasContext *s, uint32_t insn)
19
default:
20
/* all other sf/type/rmode combinations are invalid */
21
unallocated_encoding(s);
22
- break;
23
+ return;
24
}
25
26
if (!fp_access_check(s)) {
27
--
28
2.17.0
29
30
diff view generated by jsdifflib
Deleted patch
1
From: Richard Henderson <richard.henderson@linaro.org>
2
1
3
Cc: qemu-stable@nongnu.org
4
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
5
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6
Tested-by: Alex Bennée <alex.bennee@linaro.org>
7
Message-id: 20180512003217.9105-4-richard.henderson@linaro.org
8
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9
---
10
target/arm/helper.h | 6 +++
11
target/arm/helper.c | 38 ++++++++++++++-
12
target/arm/translate-a64.c | 96 +++++++++++++++++++++++++++++++-------
13
3 files changed, 122 insertions(+), 18 deletions(-)
14
15
diff --git a/target/arm/helper.h b/target/arm/helper.h
16
index XXXXXXX..XXXXXXX 100644
17
--- a/target/arm/helper.h
18
+++ b/target/arm/helper.h
19
@@ -XXX,XX +XXX,XX @@ DEF_HELPER_3(vfp_touhd_round_to_zero, i64, f64, i32, ptr)
20
DEF_HELPER_3(vfp_tould_round_to_zero, i64, f64, i32, ptr)
21
DEF_HELPER_3(vfp_touhh, i32, f16, i32, ptr)
22
DEF_HELPER_3(vfp_toshh, i32, f16, i32, ptr)
23
+DEF_HELPER_3(vfp_toulh, i32, f16, i32, ptr)
24
+DEF_HELPER_3(vfp_toslh, i32, f16, i32, ptr)
25
+DEF_HELPER_3(vfp_touqh, i64, f16, i32, ptr)
26
+DEF_HELPER_3(vfp_tosqh, i64, f16, i32, ptr)
27
DEF_HELPER_3(vfp_toshs, i32, f32, i32, ptr)
28
DEF_HELPER_3(vfp_tosls, i32, f32, i32, ptr)
29
DEF_HELPER_3(vfp_tosqs, i64, f32, i32, ptr)
30
@@ -XXX,XX +XXX,XX @@ DEF_HELPER_3(vfp_ultod, f64, i64, i32, ptr)
31
DEF_HELPER_3(vfp_uqtod, f64, i64, i32, ptr)
32
DEF_HELPER_3(vfp_sltoh, f16, i32, i32, ptr)
33
DEF_HELPER_3(vfp_ultoh, f16, i32, i32, ptr)
34
+DEF_HELPER_3(vfp_sqtoh, f16, i64, i32, ptr)
35
+DEF_HELPER_3(vfp_uqtoh, f16, i64, i32, ptr)
36
37
DEF_HELPER_FLAGS_2(set_rmode, TCG_CALL_NO_RWG, i32, i32, ptr)
38
DEF_HELPER_FLAGS_2(set_neon_rmode, TCG_CALL_NO_RWG, i32, i32, env)
39
diff --git a/target/arm/helper.c b/target/arm/helper.c
40
index XXXXXXX..XXXXXXX 100644
41
--- a/target/arm/helper.c
42
+++ b/target/arm/helper.c
43
@@ -XXX,XX +XXX,XX @@ VFP_CONV_FIX_A64(uq, s, 32, 64, uint64)
44
#undef VFP_CONV_FIX_A64
45
46
/* Conversion to/from f16 can overflow to infinity before/after scaling.
47
- * Therefore we convert to f64 (which does not round), scale,
48
- * and then convert f64 to f16 (which may round).
49
+ * Therefore we convert to f64, scale, and then convert f64 to f16; or
50
+ * vice versa for conversion to integer.
51
+ *
52
+ * For 16- and 32-bit integers, the conversion to f64 never rounds.
53
+ * For 64-bit integers, any integer that would cause rounding will also
54
+ * overflow to f16 infinity, so there is no double rounding problem.
55
*/
56
57
static float16 do_postscale_fp16(float64 f, int shift, float_status *fpst)
58
@@ -XXX,XX +XXX,XX @@ float16 HELPER(vfp_ultoh)(uint32_t x, uint32_t shift, void *fpst)
59
return do_postscale_fp16(uint32_to_float64(x, fpst), shift, fpst);
60
}
61
62
+float16 HELPER(vfp_sqtoh)(uint64_t x, uint32_t shift, void *fpst)
63
+{
64
+ return do_postscale_fp16(int64_to_float64(x, fpst), shift, fpst);
65
+}
66
+
67
+float16 HELPER(vfp_uqtoh)(uint64_t x, uint32_t shift, void *fpst)
68
+{
69
+ return do_postscale_fp16(uint64_to_float64(x, fpst), shift, fpst);
70
+}
71
+
72
static float64 do_prescale_fp16(float16 f, int shift, float_status *fpst)
73
{
74
if (unlikely(float16_is_any_nan(f))) {
75
@@ -XXX,XX +XXX,XX @@ uint32_t HELPER(vfp_touhh)(float16 x, uint32_t shift, void *fpst)
76
return float64_to_uint16(do_prescale_fp16(x, shift, fpst), fpst);
77
}
78
79
+uint32_t HELPER(vfp_toslh)(float16 x, uint32_t shift, void *fpst)
80
+{
81
+ return float64_to_int32(do_prescale_fp16(x, shift, fpst), fpst);
82
+}
83
+
84
+uint32_t HELPER(vfp_toulh)(float16 x, uint32_t shift, void *fpst)
85
+{
86
+ return float64_to_uint32(do_prescale_fp16(x, shift, fpst), fpst);
87
+}
88
+
89
+uint64_t HELPER(vfp_tosqh)(float16 x, uint32_t shift, void *fpst)
90
+{
91
+ return float64_to_int64(do_prescale_fp16(x, shift, fpst), fpst);
92
+}
93
+
94
+uint64_t HELPER(vfp_touqh)(float16 x, uint32_t shift, void *fpst)
95
+{
96
+ return float64_to_uint64(do_prescale_fp16(x, shift, fpst), fpst);
97
+}
98
+
99
/* Set the current fp rounding mode and return the old one.
100
* The argument is a softfloat float_round_ value.
101
*/
102
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
103
index XXXXXXX..XXXXXXX 100644
104
--- a/target/arm/translate-a64.c
105
+++ b/target/arm/translate-a64.c
106
@@ -XXX,XX +XXX,XX @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
107
bool itof, int rmode, int scale, int sf, int type)
108
{
109
bool is_signed = !(opcode & 1);
110
- bool is_double = type;
111
TCGv_ptr tcg_fpstatus;
112
- TCGv_i32 tcg_shift;
113
+ TCGv_i32 tcg_shift, tcg_single;
114
+ TCGv_i64 tcg_double;
115
116
- tcg_fpstatus = get_fpstatus_ptr(false);
117
+ tcg_fpstatus = get_fpstatus_ptr(type == 3);
118
119
tcg_shift = tcg_const_i32(64 - scale);
120
121
@@ -XXX,XX +XXX,XX @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
122
tcg_int = tcg_extend;
123
}
124
125
- if (is_double) {
126
- TCGv_i64 tcg_double = tcg_temp_new_i64();
127
+ switch (type) {
128
+ case 1: /* float64 */
129
+ tcg_double = tcg_temp_new_i64();
130
if (is_signed) {
131
gen_helper_vfp_sqtod(tcg_double, tcg_int,
132
tcg_shift, tcg_fpstatus);
133
@@ -XXX,XX +XXX,XX @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
134
}
135
write_fp_dreg(s, rd, tcg_double);
136
tcg_temp_free_i64(tcg_double);
137
- } else {
138
- TCGv_i32 tcg_single = tcg_temp_new_i32();
139
+ break;
140
+
141
+ case 0: /* float32 */
142
+ tcg_single = tcg_temp_new_i32();
143
if (is_signed) {
144
gen_helper_vfp_sqtos(tcg_single, tcg_int,
145
tcg_shift, tcg_fpstatus);
146
@@ -XXX,XX +XXX,XX @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
147
}
148
write_fp_sreg(s, rd, tcg_single);
149
tcg_temp_free_i32(tcg_single);
150
+ break;
151
+
152
+ case 3: /* float16 */
153
+ tcg_single = tcg_temp_new_i32();
154
+ if (is_signed) {
155
+ gen_helper_vfp_sqtoh(tcg_single, tcg_int,
156
+ tcg_shift, tcg_fpstatus);
157
+ } else {
158
+ gen_helper_vfp_uqtoh(tcg_single, tcg_int,
159
+ tcg_shift, tcg_fpstatus);
160
+ }
161
+ write_fp_sreg(s, rd, tcg_single);
162
+ tcg_temp_free_i32(tcg_single);
163
+ break;
164
+
165
+ default:
166
+ g_assert_not_reached();
167
}
168
} else {
169
TCGv_i64 tcg_int = cpu_reg(s, rd);
170
@@ -XXX,XX +XXX,XX @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
171
172
gen_helper_set_rmode(tcg_rmode, tcg_rmode, tcg_fpstatus);
173
174
- if (is_double) {
175
- TCGv_i64 tcg_double = read_fp_dreg(s, rn);
176
+ switch (type) {
177
+ case 1: /* float64 */
178
+ tcg_double = read_fp_dreg(s, rn);
179
if (is_signed) {
180
if (!sf) {
181
gen_helper_vfp_tosld(tcg_int, tcg_double,
182
@@ -XXX,XX +XXX,XX @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
183
tcg_shift, tcg_fpstatus);
184
}
185
}
186
+ if (!sf) {
187
+ tcg_gen_ext32u_i64(tcg_int, tcg_int);
188
+ }
189
tcg_temp_free_i64(tcg_double);
190
- } else {
191
- TCGv_i32 tcg_single = read_fp_sreg(s, rn);
192
+ break;
193
+
194
+ case 0: /* float32 */
195
+ tcg_single = read_fp_sreg(s, rn);
196
if (sf) {
197
if (is_signed) {
198
gen_helper_vfp_tosqs(tcg_int, tcg_single,
199
@@ -XXX,XX +XXX,XX @@ static void handle_fpfpcvt(DisasContext *s, int rd, int rn, int opcode,
200
tcg_temp_free_i32(tcg_dest);
201
}
202
tcg_temp_free_i32(tcg_single);
203
+ break;
204
+
205
+ case 3: /* float16 */
206
+ tcg_single = read_fp_sreg(s, rn);
207
+ if (sf) {
208
+ if (is_signed) {
209
+ gen_helper_vfp_tosqh(tcg_int, tcg_single,
210
+ tcg_shift, tcg_fpstatus);
211
+ } else {
212
+ gen_helper_vfp_touqh(tcg_int, tcg_single,
213
+ tcg_shift, tcg_fpstatus);
214
+ }
215
+ } else {
216
+ TCGv_i32 tcg_dest = tcg_temp_new_i32();
217
+ if (is_signed) {
218
+ gen_helper_vfp_toslh(tcg_dest, tcg_single,
219
+ tcg_shift, tcg_fpstatus);
220
+ } else {
221
+ gen_helper_vfp_toulh(tcg_dest, tcg_single,
222
+ tcg_shift, tcg_fpstatus);
223
+ }
224
+ tcg_gen_extu_i32_i64(tcg_int, tcg_dest);
225
+ tcg_temp_free_i32(tcg_dest);
226
+ }
227
+ tcg_temp_free_i32(tcg_single);
228
+ break;
229
+
230
+ default:
231
+ g_assert_not_reached();
232
}
233
234
gen_helper_set_rmode(tcg_rmode, tcg_rmode, tcg_fpstatus);
235
tcg_temp_free_i32(tcg_rmode);
236
-
237
- if (!sf) {
238
- tcg_gen_ext32u_i64(tcg_int, tcg_int);
239
- }
240
}
241
242
tcg_temp_free_ptr(tcg_fpstatus);
243
@@ -XXX,XX +XXX,XX @@ static void disas_fp_int_conv(DisasContext *s, uint32_t insn)
244
/* actual FP conversions */
245
bool itof = extract32(opcode, 1, 1);
246
247
- if (type > 1 || (rmode != 0 && opcode > 1)) {
248
+ if (rmode != 0 && opcode > 1) {
249
+ unallocated_encoding(s);
250
+ return;
251
+ }
252
+ switch (type) {
253
+ case 0: /* float32 */
254
+ case 1: /* float64 */
255
+ break;
256
+ case 3: /* float16 */
257
+ if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
258
+ break;
259
+ }
260
+ /* fallthru */
261
+ default:
262
unallocated_encoding(s);
263
return;
264
}
265
--
266
2.17.0
267
268
diff view generated by jsdifflib
Deleted patch
1
From: Richard Henderson <richard.henderson@linaro.org>
2
1
3
Cc: qemu-stable@nongnu.org
4
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
5
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6
Tested-by: Alex Bennée <alex.bennee@linaro.org>
7
Message-id: 20180512003217.9105-5-richard.henderson@linaro.org
8
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9
---
10
target/arm/translate-a64.c | 17 +++++++++++++++--
11
1 file changed, 15 insertions(+), 2 deletions(-)
12
13
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
14
index XXXXXXX..XXXXXXX 100644
15
--- a/target/arm/translate-a64.c
16
+++ b/target/arm/translate-a64.c
17
@@ -XXX,XX +XXX,XX @@ static void disas_fp_fixed_conv(DisasContext *s, uint32_t insn)
18
bool sf = extract32(insn, 31, 1);
19
bool itof;
20
21
- if (sbit || (type > 1)
22
- || (!sf && scale < 32)) {
23
+ if (sbit || (!sf && scale < 32)) {
24
+ unallocated_encoding(s);
25
+ return;
26
+ }
27
+
28
+ switch (type) {
29
+ case 0: /* float32 */
30
+ case 1: /* float64 */
31
+ break;
32
+ case 3: /* float16 */
33
+ if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
34
+ break;
35
+ }
36
+ /* fallthru */
37
+ default:
38
unallocated_encoding(s);
39
return;
40
}
41
--
42
2.17.0
43
44
diff view generated by jsdifflib
Deleted patch
1
From: Richard Henderson <richard.henderson@linaro.org>
2
1
3
Cc: qemu-stable@nongnu.org
4
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
5
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
6
Tested-by: Alex Bennée <alex.bennee@linaro.org>
7
Message-id: 20180512003217.9105-6-richard.henderson@linaro.org
8
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
9
---
10
target/arm/translate-a64.c | 30 ++++++++++++++----------------
11
1 file changed, 14 insertions(+), 16 deletions(-)
12
13
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
14
index XXXXXXX..XXXXXXX 100644
15
--- a/target/arm/translate-a64.c
16
+++ b/target/arm/translate-a64.c
17
@@ -XXX,XX +XXX,XX @@ static TCGv_i32 read_fp_sreg(DisasContext *s, int reg)
18
return v;
19
}
20
21
+static TCGv_i32 read_fp_hreg(DisasContext *s, int reg)
22
+{
23
+ TCGv_i32 v = tcg_temp_new_i32();
24
+
25
+ tcg_gen_ld16u_i32(v, cpu_env, fp_reg_offset(s, reg, MO_16));
26
+ return v;
27
+}
28
+
29
/* Clear the bits above an N-bit vector, for N = (is_q ? 128 : 64).
30
* If SVE is not enabled, then there are only 128 bits in the vector.
31
*/
32
@@ -XXX,XX +XXX,XX @@ static void disas_fp_csel(DisasContext *s, uint32_t insn)
33
static void handle_fp_1src_half(DisasContext *s, int opcode, int rd, int rn)
34
{
35
TCGv_ptr fpst = NULL;
36
- TCGv_i32 tcg_op = tcg_temp_new_i32();
37
+ TCGv_i32 tcg_op = read_fp_hreg(s, rn);
38
TCGv_i32 tcg_res = tcg_temp_new_i32();
39
40
- read_vec_element_i32(s, tcg_op, rn, 0, MO_16);
41
-
42
switch (opcode) {
43
case 0x0: /* FMOV */
44
tcg_gen_mov_i32(tcg_res, tcg_op);
45
@@ -XXX,XX +XXX,XX @@ static void disas_simd_scalar_three_reg_diff(DisasContext *s, uint32_t insn)
46
tcg_temp_free_i64(tcg_op2);
47
tcg_temp_free_i64(tcg_res);
48
} else {
49
- TCGv_i32 tcg_op1 = tcg_temp_new_i32();
50
- TCGv_i32 tcg_op2 = tcg_temp_new_i32();
51
+ TCGv_i32 tcg_op1 = read_fp_hreg(s, rn);
52
+ TCGv_i32 tcg_op2 = read_fp_hreg(s, rm);
53
TCGv_i64 tcg_res = tcg_temp_new_i64();
54
55
- read_vec_element_i32(s, tcg_op1, rn, 0, MO_16);
56
- read_vec_element_i32(s, tcg_op2, rm, 0, MO_16);
57
-
58
gen_helper_neon_mull_s16(tcg_res, tcg_op1, tcg_op2);
59
gen_helper_neon_addl_saturate_s32(tcg_res, cpu_env, tcg_res, tcg_res);
60
61
@@ -XXX,XX +XXX,XX @@ static void disas_simd_scalar_three_reg_same_fp16(DisasContext *s,
62
63
fpst = get_fpstatus_ptr(true);
64
65
- tcg_op1 = tcg_temp_new_i32();
66
- tcg_op2 = tcg_temp_new_i32();
67
+ tcg_op1 = read_fp_hreg(s, rn);
68
+ tcg_op2 = read_fp_hreg(s, rm);
69
tcg_res = tcg_temp_new_i32();
70
71
- read_vec_element_i32(s, tcg_op1, rn, 0, MO_16);
72
- read_vec_element_i32(s, tcg_op2, rm, 0, MO_16);
73
-
74
switch (fpopcode) {
75
case 0x03: /* FMULX */
76
gen_helper_advsimd_mulxh(tcg_res, tcg_op1, tcg_op2, fpst);
77
@@ -XXX,XX +XXX,XX @@ static void disas_simd_two_reg_misc_fp16(DisasContext *s, uint32_t insn)
78
}
79
80
if (is_scalar) {
81
- TCGv_i32 tcg_op = tcg_temp_new_i32();
82
+ TCGv_i32 tcg_op = read_fp_hreg(s, rn);
83
TCGv_i32 tcg_res = tcg_temp_new_i32();
84
85
- read_vec_element_i32(s, tcg_op, rn, 0, MO_16);
86
-
87
switch (fpop) {
88
case 0x1a: /* FCVTNS */
89
case 0x1b: /* FCVTMS */
90
--
91
2.17.0
92
93
diff view generated by jsdifflib
Deleted patch
1
From: Richard Henderson <richard.henderson@linaro.org>
2
1
3
We missed all of the scalar fp16 binary operations.
4
5
Cc: qemu-stable@nongnu.org
6
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
7
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
8
Tested-by: Alex Bennée <alex.bennee@linaro.org>
9
Message-id: 20180512003217.9105-7-richard.henderson@linaro.org
10
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11
---
12
target/arm/translate-a64.c | 65 ++++++++++++++++++++++++++++++++++++++
13
1 file changed, 65 insertions(+)
14
15
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
16
index XXXXXXX..XXXXXXX 100644
17
--- a/target/arm/translate-a64.c
18
+++ b/target/arm/translate-a64.c
19
@@ -XXX,XX +XXX,XX @@ static void handle_fp_2src_double(DisasContext *s, int opcode,
20
tcg_temp_free_i64(tcg_res);
21
}
22
23
+/* Floating-point data-processing (2 source) - half precision */
24
+static void handle_fp_2src_half(DisasContext *s, int opcode,
25
+ int rd, int rn, int rm)
26
+{
27
+ TCGv_i32 tcg_op1;
28
+ TCGv_i32 tcg_op2;
29
+ TCGv_i32 tcg_res;
30
+ TCGv_ptr fpst;
31
+
32
+ tcg_res = tcg_temp_new_i32();
33
+ fpst = get_fpstatus_ptr(true);
34
+ tcg_op1 = read_fp_hreg(s, rn);
35
+ tcg_op2 = read_fp_hreg(s, rm);
36
+
37
+ switch (opcode) {
38
+ case 0x0: /* FMUL */
39
+ gen_helper_advsimd_mulh(tcg_res, tcg_op1, tcg_op2, fpst);
40
+ break;
41
+ case 0x1: /* FDIV */
42
+ gen_helper_advsimd_divh(tcg_res, tcg_op1, tcg_op2, fpst);
43
+ break;
44
+ case 0x2: /* FADD */
45
+ gen_helper_advsimd_addh(tcg_res, tcg_op1, tcg_op2, fpst);
46
+ break;
47
+ case 0x3: /* FSUB */
48
+ gen_helper_advsimd_subh(tcg_res, tcg_op1, tcg_op2, fpst);
49
+ break;
50
+ case 0x4: /* FMAX */
51
+ gen_helper_advsimd_maxh(tcg_res, tcg_op1, tcg_op2, fpst);
52
+ break;
53
+ case 0x5: /* FMIN */
54
+ gen_helper_advsimd_minh(tcg_res, tcg_op1, tcg_op2, fpst);
55
+ break;
56
+ case 0x6: /* FMAXNM */
57
+ gen_helper_advsimd_maxnumh(tcg_res, tcg_op1, tcg_op2, fpst);
58
+ break;
59
+ case 0x7: /* FMINNM */
60
+ gen_helper_advsimd_minnumh(tcg_res, tcg_op1, tcg_op2, fpst);
61
+ break;
62
+ case 0x8: /* FNMUL */
63
+ gen_helper_advsimd_mulh(tcg_res, tcg_op1, tcg_op2, fpst);
64
+ tcg_gen_xori_i32(tcg_res, tcg_res, 0x8000);
65
+ break;
66
+ default:
67
+ g_assert_not_reached();
68
+ }
69
+
70
+ write_fp_sreg(s, rd, tcg_res);
71
+
72
+ tcg_temp_free_ptr(fpst);
73
+ tcg_temp_free_i32(tcg_op1);
74
+ tcg_temp_free_i32(tcg_op2);
75
+ tcg_temp_free_i32(tcg_res);
76
+}
77
+
78
/* Floating point data-processing (2 source)
79
* 31 30 29 28 24 23 22 21 20 16 15 12 11 10 9 5 4 0
80
* +---+---+---+-----------+------+---+------+--------+-----+------+------+
81
@@ -XXX,XX +XXX,XX @@ static void disas_fp_2src(DisasContext *s, uint32_t insn)
82
}
83
handle_fp_2src_double(s, opcode, rd, rn, rm);
84
break;
85
+ case 3:
86
+ if (!arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
87
+ unallocated_encoding(s);
88
+ return;
89
+ }
90
+ if (!fp_access_check(s)) {
91
+ return;
92
+ }
93
+ handle_fp_2src_half(s, opcode, rd, rn, rm);
94
+ break;
95
default:
96
unallocated_encoding(s);
97
}
98
--
99
2.17.0
100
101
diff view generated by jsdifflib
Deleted patch
1
From: Richard Henderson <richard.henderson@linaro.org>
2
1
3
We missed all of the scalar fp16 fma operations.
4
5
Cc: qemu-stable@nongnu.org
6
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
7
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
8
Tested-by: Alex Bennée <alex.bennee@linaro.org>
9
Message-id: 20180512003217.9105-8-richard.henderson@linaro.org
10
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
11
---
12
target/arm/translate-a64.c | 48 ++++++++++++++++++++++++++++++++++++++
13
1 file changed, 48 insertions(+)
14
15
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
16
index XXXXXXX..XXXXXXX 100644
17
--- a/target/arm/translate-a64.c
18
+++ b/target/arm/translate-a64.c
19
@@ -XXX,XX +XXX,XX @@ static void handle_fp_3src_double(DisasContext *s, bool o0, bool o1,
20
tcg_temp_free_i64(tcg_res);
21
}
22
23
+/* Floating-point data-processing (3 source) - half precision */
24
+static void handle_fp_3src_half(DisasContext *s, bool o0, bool o1,
25
+ int rd, int rn, int rm, int ra)
26
+{
27
+ TCGv_i32 tcg_op1, tcg_op2, tcg_op3;
28
+ TCGv_i32 tcg_res = tcg_temp_new_i32();
29
+ TCGv_ptr fpst = get_fpstatus_ptr(true);
30
+
31
+ tcg_op1 = read_fp_hreg(s, rn);
32
+ tcg_op2 = read_fp_hreg(s, rm);
33
+ tcg_op3 = read_fp_hreg(s, ra);
34
+
35
+ /* These are fused multiply-add, and must be done as one
36
+ * floating point operation with no rounding between the
37
+ * multiplication and addition steps.
38
+ * NB that doing the negations here as separate steps is
39
+ * correct : an input NaN should come out with its sign bit
40
+ * flipped if it is a negated-input.
41
+ */
42
+ if (o1 == true) {
43
+ tcg_gen_xori_i32(tcg_op3, tcg_op3, 0x8000);
44
+ }
45
+
46
+ if (o0 != o1) {
47
+ tcg_gen_xori_i32(tcg_op1, tcg_op1, 0x8000);
48
+ }
49
+
50
+ gen_helper_advsimd_muladdh(tcg_res, tcg_op1, tcg_op2, tcg_op3, fpst);
51
+
52
+ write_fp_sreg(s, rd, tcg_res);
53
+
54
+ tcg_temp_free_ptr(fpst);
55
+ tcg_temp_free_i32(tcg_op1);
56
+ tcg_temp_free_i32(tcg_op2);
57
+ tcg_temp_free_i32(tcg_op3);
58
+ tcg_temp_free_i32(tcg_res);
59
+}
60
+
61
/* Floating point data-processing (3 source)
62
* 31 30 29 28 24 23 22 21 20 16 15 14 10 9 5 4 0
63
* +---+---+---+-----------+------+----+------+----+------+------+------+
64
@@ -XXX,XX +XXX,XX @@ static void disas_fp_3src(DisasContext *s, uint32_t insn)
65
}
66
handle_fp_3src_double(s, o0, o1, rd, rn, rm, ra);
67
break;
68
+ case 3:
69
+ if (!arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
70
+ unallocated_encoding(s);
71
+ return;
72
+ }
73
+ if (!fp_access_check(s)) {
74
+ return;
75
+ }
76
+ handle_fp_3src_half(s, o0, o1, rd, rn, rm, ra);
77
+ break;
78
default:
79
unallocated_encoding(s);
80
}
81
--
82
2.17.0
83
84
diff view generated by jsdifflib
Deleted patch
1
From: Alex Bennée <alex.bennee@linaro.org>
2
1
3
These where missed out from the rest of the half-precision work.
4
5
Cc: qemu-stable@nongnu.org
6
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
7
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
8
Tested-by: Alex Bennée <alex.bennee@linaro.org>
9
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
10
Message-id: 20180512003217.9105-9-richard.henderson@linaro.org
11
[rth: Diagnose lack of FP16 before fp_access_check]
12
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
13
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
14
---
15
target/arm/helper-a64.h | 2 +
16
target/arm/helper-a64.c | 10 +++++
17
target/arm/translate-a64.c | 88 ++++++++++++++++++++++++++++++--------
18
3 files changed, 83 insertions(+), 17 deletions(-)
19
20
diff --git a/target/arm/helper-a64.h b/target/arm/helper-a64.h
21
index XXXXXXX..XXXXXXX 100644
22
--- a/target/arm/helper-a64.h
23
+++ b/target/arm/helper-a64.h
24
@@ -XXX,XX +XXX,XX @@
25
DEF_HELPER_FLAGS_2(udiv64, TCG_CALL_NO_RWG_SE, i64, i64, i64)
26
DEF_HELPER_FLAGS_2(sdiv64, TCG_CALL_NO_RWG_SE, s64, s64, s64)
27
DEF_HELPER_FLAGS_1(rbit64, TCG_CALL_NO_RWG_SE, i64, i64)
28
+DEF_HELPER_3(vfp_cmph_a64, i64, f16, f16, ptr)
29
+DEF_HELPER_3(vfp_cmpeh_a64, i64, f16, f16, ptr)
30
DEF_HELPER_3(vfp_cmps_a64, i64, f32, f32, ptr)
31
DEF_HELPER_3(vfp_cmpes_a64, i64, f32, f32, ptr)
32
DEF_HELPER_3(vfp_cmpd_a64, i64, f64, f64, ptr)
33
diff --git a/target/arm/helper-a64.c b/target/arm/helper-a64.c
34
index XXXXXXX..XXXXXXX 100644
35
--- a/target/arm/helper-a64.c
36
+++ b/target/arm/helper-a64.c
37
@@ -XXX,XX +XXX,XX @@ static inline uint32_t float_rel_to_flags(int res)
38
return flags;
39
}
40
41
+uint64_t HELPER(vfp_cmph_a64)(float16 x, float16 y, void *fp_status)
42
+{
43
+ return float_rel_to_flags(float16_compare_quiet(x, y, fp_status));
44
+}
45
+
46
+uint64_t HELPER(vfp_cmpeh_a64)(float16 x, float16 y, void *fp_status)
47
+{
48
+ return float_rel_to_flags(float16_compare(x, y, fp_status));
49
+}
50
+
51
uint64_t HELPER(vfp_cmps_a64)(float32 x, float32 y, void *fp_status)
52
{
53
return float_rel_to_flags(float32_compare_quiet(x, y, fp_status));
54
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
55
index XXXXXXX..XXXXXXX 100644
56
--- a/target/arm/translate-a64.c
57
+++ b/target/arm/translate-a64.c
58
@@ -XXX,XX +XXX,XX @@ static void disas_data_proc_reg(DisasContext *s, uint32_t insn)
59
}
60
}
61
62
-static void handle_fp_compare(DisasContext *s, bool is_double,
63
+static void handle_fp_compare(DisasContext *s, int size,
64
unsigned int rn, unsigned int rm,
65
bool cmp_with_zero, bool signal_all_nans)
66
{
67
TCGv_i64 tcg_flags = tcg_temp_new_i64();
68
- TCGv_ptr fpst = get_fpstatus_ptr(false);
69
+ TCGv_ptr fpst = get_fpstatus_ptr(size == MO_16);
70
71
- if (is_double) {
72
+ if (size == MO_64) {
73
TCGv_i64 tcg_vn, tcg_vm;
74
75
tcg_vn = read_fp_dreg(s, rn);
76
@@ -XXX,XX +XXX,XX @@ static void handle_fp_compare(DisasContext *s, bool is_double,
77
tcg_temp_free_i64(tcg_vn);
78
tcg_temp_free_i64(tcg_vm);
79
} else {
80
- TCGv_i32 tcg_vn, tcg_vm;
81
+ TCGv_i32 tcg_vn = tcg_temp_new_i32();
82
+ TCGv_i32 tcg_vm = tcg_temp_new_i32();
83
84
- tcg_vn = read_fp_sreg(s, rn);
85
+ read_vec_element_i32(s, tcg_vn, rn, 0, size);
86
if (cmp_with_zero) {
87
- tcg_vm = tcg_const_i32(0);
88
+ tcg_gen_movi_i32(tcg_vm, 0);
89
} else {
90
- tcg_vm = read_fp_sreg(s, rm);
91
+ read_vec_element_i32(s, tcg_vm, rm, 0, size);
92
}
93
- if (signal_all_nans) {
94
- gen_helper_vfp_cmpes_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
95
- } else {
96
- gen_helper_vfp_cmps_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
97
+
98
+ switch (size) {
99
+ case MO_32:
100
+ if (signal_all_nans) {
101
+ gen_helper_vfp_cmpes_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
102
+ } else {
103
+ gen_helper_vfp_cmps_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
104
+ }
105
+ break;
106
+ case MO_16:
107
+ if (signal_all_nans) {
108
+ gen_helper_vfp_cmpeh_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
109
+ } else {
110
+ gen_helper_vfp_cmph_a64(tcg_flags, tcg_vn, tcg_vm, fpst);
111
+ }
112
+ break;
113
+ default:
114
+ g_assert_not_reached();
115
}
116
+
117
tcg_temp_free_i32(tcg_vn);
118
tcg_temp_free_i32(tcg_vm);
119
}
120
@@ -XXX,XX +XXX,XX @@ static void handle_fp_compare(DisasContext *s, bool is_double,
121
static void disas_fp_compare(DisasContext *s, uint32_t insn)
122
{
123
unsigned int mos, type, rm, op, rn, opc, op2r;
124
+ int size;
125
126
mos = extract32(insn, 29, 3);
127
- type = extract32(insn, 22, 2); /* 0 = single, 1 = double */
128
+ type = extract32(insn, 22, 2);
129
rm = extract32(insn, 16, 5);
130
op = extract32(insn, 14, 2);
131
rn = extract32(insn, 5, 5);
132
opc = extract32(insn, 3, 2);
133
op2r = extract32(insn, 0, 3);
134
135
- if (mos || op || op2r || type > 1) {
136
+ if (mos || op || op2r) {
137
+ unallocated_encoding(s);
138
+ return;
139
+ }
140
+
141
+ switch (type) {
142
+ case 0:
143
+ size = MO_32;
144
+ break;
145
+ case 1:
146
+ size = MO_64;
147
+ break;
148
+ case 3:
149
+ size = MO_16;
150
+ if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
151
+ break;
152
+ }
153
+ /* fallthru */
154
+ default:
155
unallocated_encoding(s);
156
return;
157
}
158
@@ -XXX,XX +XXX,XX @@ static void disas_fp_compare(DisasContext *s, uint32_t insn)
159
return;
160
}
161
162
- handle_fp_compare(s, type, rn, rm, opc & 1, opc & 2);
163
+ handle_fp_compare(s, size, rn, rm, opc & 1, opc & 2);
164
}
165
166
/* Floating point conditional compare
167
@@ -XXX,XX +XXX,XX @@ static void disas_fp_ccomp(DisasContext *s, uint32_t insn)
168
unsigned int mos, type, rm, cond, rn, op, nzcv;
169
TCGv_i64 tcg_flags;
170
TCGLabel *label_continue = NULL;
171
+ int size;
172
173
mos = extract32(insn, 29, 3);
174
- type = extract32(insn, 22, 2); /* 0 = single, 1 = double */
175
+ type = extract32(insn, 22, 2);
176
rm = extract32(insn, 16, 5);
177
cond = extract32(insn, 12, 4);
178
rn = extract32(insn, 5, 5);
179
op = extract32(insn, 4, 1);
180
nzcv = extract32(insn, 0, 4);
181
182
- if (mos || type > 1) {
183
+ if (mos) {
184
+ unallocated_encoding(s);
185
+ return;
186
+ }
187
+
188
+ switch (type) {
189
+ case 0:
190
+ size = MO_32;
191
+ break;
192
+ case 1:
193
+ size = MO_64;
194
+ break;
195
+ case 3:
196
+ size = MO_16;
197
+ if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
198
+ break;
199
+ }
200
+ /* fallthru */
201
+ default:
202
unallocated_encoding(s);
203
return;
204
}
205
@@ -XXX,XX +XXX,XX @@ static void disas_fp_ccomp(DisasContext *s, uint32_t insn)
206
gen_set_label(label_match);
207
}
208
209
- handle_fp_compare(s, type, rn, rm, false, op);
210
+ handle_fp_compare(s, size, rn, rm, false, op);
211
212
if (cond < 0x0e) {
213
gen_set_label(label_continue);
214
--
215
2.17.0
216
217
diff view generated by jsdifflib
Deleted patch
1
From: Alex Bennée <alex.bennee@linaro.org>
2
1
3
These were missed out from the rest of the half-precision work.
4
5
Cc: qemu-stable@nongnu.org
6
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
7
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
8
Tested-by: Alex Bennée <alex.bennee@linaro.org>
9
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
10
Message-id: 20180512003217.9105-10-richard.henderson@linaro.org
11
[rth: Fix erroneous check vs type]
12
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
13
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
14
---
15
target/arm/translate-a64.c | 31 +++++++++++++++++++++++++------
16
1 file changed, 25 insertions(+), 6 deletions(-)
17
18
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
19
index XXXXXXX..XXXXXXX 100644
20
--- a/target/arm/translate-a64.c
21
+++ b/target/arm/translate-a64.c
22
@@ -XXX,XX +XXX,XX @@ static void disas_fp_csel(DisasContext *s, uint32_t insn)
23
unsigned int mos, type, rm, cond, rn, rd;
24
TCGv_i64 t_true, t_false, t_zero;
25
DisasCompare64 c;
26
+ TCGMemOp sz;
27
28
mos = extract32(insn, 29, 3);
29
- type = extract32(insn, 22, 2); /* 0 = single, 1 = double */
30
+ type = extract32(insn, 22, 2);
31
rm = extract32(insn, 16, 5);
32
cond = extract32(insn, 12, 4);
33
rn = extract32(insn, 5, 5);
34
rd = extract32(insn, 0, 5);
35
36
- if (mos || type > 1) {
37
+ if (mos) {
38
+ unallocated_encoding(s);
39
+ return;
40
+ }
41
+
42
+ switch (type) {
43
+ case 0:
44
+ sz = MO_32;
45
+ break;
46
+ case 1:
47
+ sz = MO_64;
48
+ break;
49
+ case 3:
50
+ sz = MO_16;
51
+ if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
52
+ break;
53
+ }
54
+ /* fallthru */
55
+ default:
56
unallocated_encoding(s);
57
return;
58
}
59
@@ -XXX,XX +XXX,XX @@ static void disas_fp_csel(DisasContext *s, uint32_t insn)
60
return;
61
}
62
63
- /* Zero extend sreg inputs to 64 bits now. */
64
+ /* Zero extend sreg & hreg inputs to 64 bits now. */
65
t_true = tcg_temp_new_i64();
66
t_false = tcg_temp_new_i64();
67
- read_vec_element(s, t_true, rn, 0, type ? MO_64 : MO_32);
68
- read_vec_element(s, t_false, rm, 0, type ? MO_64 : MO_32);
69
+ read_vec_element(s, t_true, rn, 0, sz);
70
+ read_vec_element(s, t_false, rm, 0, sz);
71
72
a64_test_cc(&c, cond);
73
t_zero = tcg_const_i64(0);
74
@@ -XXX,XX +XXX,XX @@ static void disas_fp_csel(DisasContext *s, uint32_t insn)
75
tcg_temp_free_i64(t_false);
76
a64_free_cc(&c);
77
78
- /* Note that sregs write back zeros to the high bits,
79
+ /* Note that sregs & hregs write back zeros to the high bits,
80
and we've already done the zero-extension. */
81
write_fp_dreg(s, rd, t_true);
82
tcg_temp_free_i64(t_true);
83
--
84
2.17.0
85
86
diff view generated by jsdifflib
Deleted patch
1
From: Alex Bennée <alex.bennee@linaro.org>
2
1
3
All the hard work is already done by vfp_expand_imm, we just need to
4
make sure we pick up the correct size.
5
6
Cc: qemu-stable@nongnu.org
7
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
8
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
9
Tested-by: Alex Bennée <alex.bennee@linaro.org>
10
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
11
Message-id: 20180512003217.9105-11-richard.henderson@linaro.org
12
[rth: Merge unallocated_encoding check with TCGMemOp conversion.]
13
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
14
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
15
---
16
target/arm/translate-a64.c | 20 +++++++++++++++++---
17
1 file changed, 17 insertions(+), 3 deletions(-)
18
19
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
20
index XXXXXXX..XXXXXXX 100644
21
--- a/target/arm/translate-a64.c
22
+++ b/target/arm/translate-a64.c
23
@@ -XXX,XX +XXX,XX @@ static void disas_fp_imm(DisasContext *s, uint32_t insn)
24
{
25
int rd = extract32(insn, 0, 5);
26
int imm8 = extract32(insn, 13, 8);
27
- int is_double = extract32(insn, 22, 2);
28
+ int type = extract32(insn, 22, 2);
29
uint64_t imm;
30
TCGv_i64 tcg_res;
31
+ TCGMemOp sz;
32
33
- if (is_double > 1) {
34
+ switch (type) {
35
+ case 0:
36
+ sz = MO_32;
37
+ break;
38
+ case 1:
39
+ sz = MO_64;
40
+ break;
41
+ case 3:
42
+ sz = MO_16;
43
+ if (arm_dc_feature(s, ARM_FEATURE_V8_FP16)) {
44
+ break;
45
+ }
46
+ /* fallthru */
47
+ default:
48
unallocated_encoding(s);
49
return;
50
}
51
@@ -XXX,XX +XXX,XX @@ static void disas_fp_imm(DisasContext *s, uint32_t insn)
52
return;
53
}
54
55
- imm = vfp_expand_imm(MO_32 + is_double, imm8);
56
+ imm = vfp_expand_imm(sz, imm8);
57
58
tcg_res = tcg_const_i64(imm);
59
write_fp_dreg(s, rd, tcg_res);
60
--
61
2.17.0
62
63
diff view generated by jsdifflib
Deleted patch
1
From: Alex Bennée <alex.bennee@linaro.org>
2
1
3
We are meant to explicitly pass fpst, not cpu_env.
4
5
Cc: qemu-stable@nongnu.org
6
Signed-off-by: Alex Bennée <alex.bennee@linaro.org>
7
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
8
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
9
Tested-by: Alex Bennée <alex.bennee@linaro.org>
10
Message-id: 20180512003217.9105-12-richard.henderson@linaro.org
11
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
12
---
13
target/arm/translate-a64.c | 3 ++-
14
1 file changed, 2 insertions(+), 1 deletion(-)
15
16
diff --git a/target/arm/translate-a64.c b/target/arm/translate-a64.c
17
index XXXXXXX..XXXXXXX 100644
18
--- a/target/arm/translate-a64.c
19
+++ b/target/arm/translate-a64.c
20
@@ -XXX,XX +XXX,XX @@ static void handle_fp_1src_half(DisasContext *s, int opcode, int rd, int rn)
21
tcg_gen_xori_i32(tcg_res, tcg_op, 0x8000);
22
break;
23
case 0x3: /* FSQRT */
24
- gen_helper_sqrt_f16(tcg_res, tcg_op, cpu_env);
25
+ fpst = get_fpstatus_ptr(true);
26
+ gen_helper_sqrt_f16(tcg_res, tcg_op, fpst);
27
break;
28
case 0x8: /* FRINTN */
29
case 0x9: /* FRINTP */
30
--
31
2.17.0
32
33
diff view generated by jsdifflib