We use bfdotadd() in four callsites for various helper functions. Currently
this all assumes that we have the FPCR.EBF=0 semantics. For FPCR.EBF=1
we will need to:
* call a different routine to bfdotadd() because we need to do a
fused multiply-add rather than separate multiply and add steps
* use a different float_status that honours the FPCR rounding mode
and denormal-flushing fields
* pass in an extra float_status that has been set up to perform
round-to-odd rounding
To prepare for this, refactor all the callsites so that instead of
for (...) {
x = bfdotadd(...);
}
they are:
float_status fpst, fpst_odd;
if (is_ebf(env, &fpst, &fpst_odd)) {
for (...) {
x = bfdotadd_ebf(..., &fpst, &fpst_odd);
}
} else {
for (...) {
x = bfdotadd(..., &fpst);
}
}
For the moment the is_ebf() function always returns false, sets up
fpst for EBF=0 semantics and never sets up fpst_odd; bfdotadd_ebf()
will assert if called. We'll fill in the handling for EBF=1 in the
next commit.
This change should be a zero-behaviour-change refactor.
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
To implement the FEAT_EBF16 semantics, we are going to need
the CPUARMState env pointer in every helper function which calls
bfdotadd().
Pass the env pointer through from generated code to the sme_bfmopa
helper. (We'll add the code that uses it when we've adjusted
all the helpers to have access to the env pointer.)
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
The FMOPA (widening) SME instruction takes pairs of half-precision
floating point values, widens them to single-precision, does a
two-way dot product and accumulates the results into a
single-precision destination. We don't quite correctly handle the
FPCR bits FZ and FZ16 which control flushing of denormal inputs and
outputs. This is because at the moment we pass a single float_status
value to the helper function, which then uses that configuration for
all the fp operations it does. However, because the inputs to this
operation are float16 and the outputs are float32 we need to use the
fp_status_f16 for the float16 input widening but the normal fp_status
for everything else. Otherwise we will apply the flushing control
FPCR.FZ16 to the 32-bit output rather than the FPCR.FZ control, and
incorrectly flush a denormal output to zero when we should not (or
vice-versa).
(In commit 207d30b5fd we tried to fix the FZ handling but
didn't get it right, switching from "use FPCR.FZ for everything" to
"use FPCR.FZ16 for everything".)
Pass the CPU env to the sme_fmopa_h helper instead of an fp_status
pointer, and have the helper pass an extra fp_status into the
f16_dotadd() function so that we can use the right status for the
right parts of this operation.
Cc: qemu-stable@nongnu.org
Fixes: 207d30b5fd ("target/arm: Use FPST_F16 for SME FMOPA (widening)")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2373
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
The UMOPA/UMOPS instructions are supposed to multiply unsigned 8 or
16 bit elements and accumulate the products into a 64-bit element.
In the Arm ARM pseudocode, this is done with the usual
infinite-precision signed arithmetic. However our implementation
doesn't quite get it right, because in the DEF_IMOP_64() macro we do:
sum += (NTYPE)(n >> 0) * (MTYPE)(m >> 0);
where NTYPE and MTYPE are uint16_t or int16_t. In the uint16_t case,
the C usual arithmetic conversions mean the values are converted to
"int" type and the multiply is done as a 32-bit multiply. This means
that if the inputs are, for example, 0xffff and 0xffff then the
result is 0xFFFE0001 as an int, which is then promoted to uint64_t
for the accumulation into sum; this promotion incorrectly sign
extends the multiply.
Avoid the incorrect sign extension by casting to int64_t before
the multiply, so we do the multiply as 64-bit signed arithmetic,
which is a type large enough that the multiply can never
overflow into the sign bit.
(The equivalent 8-bit operations in DEF_IMOP_32() are fine, because
the 8-bit multiplies can never overflow into the sign bit of a
32-bit integer.)
Cc: qemu-stable@nongnu.org
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2372
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
Message-id: 20240722172957.1041231-3-peter.maydell@linaro.org
Avoid a race condition with munmap in another thread.
Use around blocks that exclusively use "host_fn".
Keep the blocks as small as possible, but without setting
and clearing for every operation on one page.
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
We made a copy above because the fp exception flags
are not propagated back to the FPST register, but
then failed to use the copy.
Cc: qemu-stable@nongnu.org
Fixes: 558e956c71 ("target/arm: Implement FMOPA, FMOPS (non-widening)")
Signed-off-by: Daniyal Khan <danikhan632@gmail.com>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Message-id: 20240717060149.204788-2-richard.henderson@linaro.org
[rth: Split from a larger patch]
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Reviewed-by: Alex Bennée <alex.bennee@linaro.org>
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
While the 8-bit input elements are sequential in the input vector,
the 32-bit output elements are not sequential in the output matrix.
Do not attempt to compute 2 32-bit outputs at the same time.
Cc: qemu-stable@nongnu.org
Fixes: 23a5e3859f ("target/arm: Implement SME integer outer product")
Resolves: https://gitlab.com/qemu-project/qemu/-/issues/2083
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Reviewed-by: Philippe Mathieu-Daudé <philmd@linaro.org>
Message-id: 20240305163931.242795-1-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
The TBI and TCMA bits are located within mtedesc, not desc.
Cc: qemu-stable@nongnu.org
Reviewed-by: Peter Maydell <peter.maydell@linaro.org>
Signed-off-by: Richard Henderson <richard.henderson@linaro.org>
Tested-by: Gustavo Romero <gustavo.romero@linaro.org>
Message-id: 20240207025210.8837-7-richard.henderson@linaro.org
Signed-off-by: Peter Maydell <peter.maydell@linaro.org>
Peter Maydell