| Message ID | 1353063863-11446-4-git-send-email-aurelien@aurel32.net |
|---|---|
| State | New |
| Headers | show |
> -----Original Message----- > From: qemu-devel-bounces+ericj=mips.com@nongnu.org [mailto:qemu-devel- > bounces+ericj=mips.com@nongnu.org] On Behalf Of Aurelien Jarno > Sent: Friday, November 16, 2012 3:04 AM > To: qemu-devel@nongnu.org > Cc: Aurelien Jarno > Subject: [Qemu-devel] [PATCH 3/7] target-mips: add unions to access DSP > elements > > Instead of playing with bit shifting, add two unions (one for 32-bit > values, one for 64-bit ones) to access all the DSP elements with the > correct type. > > This make the code easier to read and less error prone, and allow GCC > to vectorize the code in some cases. > > Signed-off-by: Aurelien Jarno <aurelien@aurel32.net> > --- > target-mips/dsp_helper.c | 22 ++++++++++++++++++++++ > 1 file changed, 22 insertions(+) > > diff --git a/target-mips/dsp_helper.c b/target-mips/dsp_helper.c > index e7949c2..8015d8d 100644 > --- a/target-mips/dsp_helper.c > +++ b/target-mips/dsp_helper.c > @@ -20,6 +20,28 @@ > #include "cpu.h" > #include "helper.h" > > +/* As the byte ordering doesn't matter, i.e. all columns are treated > + identically, these unions can be used directly. */ > +typedef union { > + uint8_t ub[4]; > + int8_t sb[4]; > + uint16_t uh[2]; > + int16_t sh[2]; > + uint32_t uw[1]; > + int32_t sw[1]; > +} DSP32Value; > + > +typedef union { > + uint8_t ub[8]; > + int8_t sb[8]; > + uint16_t uh[4]; > + int16_t sh[4]; > + uint32_t uw[2]; > + int32_t sw[2]; > + uint64_t ul[1]; > + int64_t sl[1]; > +} DSP64Value; > + > /*** MIPS DSP internal functions begin ***/ > #define MIPSDSP_ABS(x) (((x) >= 0) ? x : -x) > #define MIPSDSP_OVERFLOW(a, b, c, d) (!(!((a ^ b ^ -1) & (a ^ c) & d))) > -- > 1.7.10.4 > Reviewed-by: Eric Johnson <ericj@mips.com>
diff --git a/target-mips/dsp_helper.c b/target-mips/dsp_helper.c index e7949c2..8015d8d 100644 --- a/target-mips/dsp_helper.c +++ b/target-mips/dsp_helper.c @@ -20,6 +20,28 @@ #include "cpu.h" #include "helper.h" +/* As the byte ordering doesn't matter, i.e. all columns are treated + identically, these unions can be used directly. */ +typedef union { + uint8_t ub[4]; + int8_t sb[4]; + uint16_t uh[2]; + int16_t sh[2]; + uint32_t uw[1]; + int32_t sw[1]; +} DSP32Value; + +typedef union { + uint8_t ub[8]; + int8_t sb[8]; + uint16_t uh[4]; + int16_t sh[4]; + uint32_t uw[2]; + int32_t sw[2]; + uint64_t ul[1]; + int64_t sl[1]; +} DSP64Value; + /*** MIPS DSP internal functions begin ***/ #define MIPSDSP_ABS(x) (((x) >= 0) ? x : -x) #define MIPSDSP_OVERFLOW(a, b, c, d) (!(!((a ^ b ^ -1) & (a ^ c) & d)))
Instead of playing with bit shifting, add two unions (one for 32-bit values, one for 64-bit ones) to access all the DSP elements with the correct type. This make the code easier to read and less error prone, and allow GCC to vectorize the code in some cases. Signed-off-by: Aurelien Jarno <aurelien@aurel32.net> --- target-mips/dsp_helper.c | 22 ++++++++++++++++++++++ 1 file changed, 22 insertions(+)