| Message ID | 20181011205206.3552-20-richard.henderson@linaro.org |
|---|---|
| State | New |
| Headers | show |
| Series | target/arm: Convert some neon insns to gvec | expand |
On 11/10/2018 22:52, Richard Henderson wrote: > For a sequence of loads or stores from a single register, > little-endian operations can be promoted to an 8-byte op. > This can reduce the number of operations by a factor of 8. > > Signed-off-by: Richard Henderson <richard.henderson@linaro.org> Reviewed-by: Philippe Mathieu-Daudé <philmd@redhat.com> > --- > target/arm/translate.c | 10 ++++++++++ > 1 file changed, 10 insertions(+) > > diff --git a/target/arm/translate.c b/target/arm/translate.c > index 12a744b3c3..09f2d648b7 100644 > --- a/target/arm/translate.c > +++ b/target/arm/translate.c > @@ -5011,6 +5011,16 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn) > if (size == 3 && (interleave | spacing) != 1) { > return 1; > } > + /* For our purposes, bytes are always little-endian. */ > + if (size == 0) { > + endian = MO_LE; > + } > + /* Consecutive little-endian elements from a single register > + * can be promoted to a larger little-endian operation. > + */ > + if (interleave == 1 && endian == MO_LE) { > + size = 3; > + } > tmp64 = tcg_temp_new_i64(); > addr = tcg_temp_new_i32(); > tmp2 = tcg_const_i32(1 << size); >
diff --git a/target/arm/translate.c b/target/arm/translate.c index 12a744b3c3..09f2d648b7 100644 --- a/target/arm/translate.c +++ b/target/arm/translate.c @@ -5011,6 +5011,16 @@ static int disas_neon_ls_insn(DisasContext *s, uint32_t insn) if (size == 3 && (interleave | spacing) != 1) { return 1; } + /* For our purposes, bytes are always little-endian. */ + if (size == 0) { + endian = MO_LE; + } + /* Consecutive little-endian elements from a single register + * can be promoted to a larger little-endian operation. + */ + if (interleave == 1 && endian == MO_LE) { + size = 3; + } tmp64 = tcg_temp_new_i64(); addr = tcg_temp_new_i32(); tmp2 = tcg_const_i32(1 << size);
For a sequence of loads or stores from a single register, little-endian operations can be promoted to an 8-byte op. This can reduce the number of operations by a factor of 8. Signed-off-by: Richard Henderson <richard.henderson@linaro.org> --- target/arm/translate.c | 10 ++++++++++ 1 file changed, 10 insertions(+)