[Mesa-dev] [PATCH v3 06/16] anv/gpu_memcpy: Add a lighter-weight GPU memcpy function

[Nanley Chery]

On Mon, Jul 10, 2017 at 09:35:25AM -0700, Jason Ekstrand wrote:
> On Wed, Jun 28, 2017 at 2:14 PM, Nanley Chery <nanleychery at gmail.com> wrote:
> 
> > We'll be performing a GPU memcpy in more places to copy small amounts of
> > data. Add an alternate function that thrashes less state.
> >
> > v2:
> > - Make a new function (Jason Ekstrand).
> > - Move the #define into the function.
> > v3:
> > - Update the function name (Jason).
> > - Update comments.
> >
> > Signed-off-by: Nanley Chery <nanley.g.chery at intel.com>
> > ---
> >  src/intel/vulkan/anv_genX.h        |  5 +++++
> >  src/intel/vulkan/genX_gpu_memcpy.c | 40 ++++++++++++++++++++++++++++++
> > ++++++++
> >  2 files changed, 45 insertions(+)
> >
> > diff --git a/src/intel/vulkan/anv_genX.h b/src/intel/vulkan/anv_genX.h
> > index 8da5e075dc..0b7322e281 100644
> > --- a/src/intel/vulkan/anv_genX.h
> > +++ b/src/intel/vulkan/anv_genX.h
> > @@ -69,5 +69,10 @@ void genX(cmd_buffer_so_memcpy)(struct anv_cmd_buffer
> > *cmd_buffer,
> >                                  struct anv_bo *src, uint32_t src_offset,
> >                                  uint32_t size);
> >
> > +void genX(cmd_buffer_mi_memcpy)(struct anv_cmd_buffer *cmd_buffer,
> > +                                struct anv_bo *dst, uint32_t dst_offset,
> > +                                struct anv_bo *src, uint32_t src_offset,
> > +                                uint32_t size);
> > +
> >  void genX(blorp_exec)(struct blorp_batch *batch,
> >                        const struct blorp_params *params);
> > diff --git a/src/intel/vulkan/genX_gpu_memcpy.c
> > b/src/intel/vulkan/genX_gpu_memcpy.c
> > index 5ef35e6283..9c6b46de94 100644
> > --- a/src/intel/vulkan/genX_gpu_memcpy.c
> > +++ b/src/intel/vulkan/genX_gpu_memcpy.c
> > @@ -52,6 +52,46 @@ gcd_pow2_u64(uint64_t a, uint64_t b)
> >  }
> >
> >  void
> > +genX(cmd_buffer_mi_memcpy)(struct anv_cmd_buffer *cmd_buffer,
> > +                           struct anv_bo *dst, uint32_t dst_offset,
> > +                           struct anv_bo *src, uint32_t src_offset,
> > +                           uint32_t size)
> > +{
> > +   /* This memcpy operates in units of dwords. */
> > +   assert(size % 4 == 0);
> > +   assert(dst_offset % 4 == 0);
> > +   assert(src_offset % 4 == 0);
> > +
> > +   for (uint32_t i = 0; i < size; i += 4) {
> > +      const struct anv_address src_addr =
> > +         (struct anv_address) { src, src_offset + i};
> > +      const struct anv_address dst_addr =
> > +         (struct anv_address) { dst, dst_offset + i};
> > +#if GEN_GEN >= 8
> > +      anv_batch_emit(&cmd_buffer->batch, GENX(MI_COPY_MEM_MEM), cp) {
> > +         cp.DestinationMemoryAddress = dst_addr;
> > +         cp.SourceMemoryAddress = src_addr;
> > +      }
> > +#else
> > +      /* IVB does not have a general purpose register for command streamer
> > +       * commands. Therefore, we use an alternate temporary register.
> > +       */
> > +#define TEMP_REG 0x2400 /* MI_PREDICATE_SRC0 */
> >
> 
> Using the predicate register seems a bit sketchy.  Vulkan doesn't support
> predication today so it's probably safe but I don't know what form
> predication will take in the future (there's a decent chance it'll get
> added) so I have no idea if this will end up being safe.  Why not use one
> of the indirect dispatch/draw registers?  Those will be safe because we
> only ever set them immediately before 3DPRIMITIVE or GPGPU_WALKER.
> 
> --Jason
> 
> 

I don't mind using any alternate register, so long as it doesn't lose
any bits (like the SO_WRITE_OFFSET registers). The register, Load
Indirect Base Vertex (0x2440), looks like it will work just fine.

Thanks,
Nanley

> > +      anv_batch_emit(&cmd_buffer->batch, GENX(MI_LOAD_REGISTER_MEM),
> > load) {
> > +         load.RegisterAddress = TEMP_REG;
> > +         load.MemoryAddress = src_addr;
> > +      }
> > +      anv_batch_emit(&cmd_buffer->batch, GENX(MI_STORE_REGISTER_MEM),
> > store) {
> > +         store.RegisterAddress = TEMP_REG;
> > +         store.MemoryAddress = dst_addr;
> > +      }
> > +#undef TEMP_REG
> > +#endif
> > +   }
> > +   return;
> > +}
> > +
> > +void
> >  genX(cmd_buffer_so_memcpy)(struct anv_cmd_buffer *cmd_buffer,
> >                             struct anv_bo *dst, uint32_t dst_offset,
> >                             struct anv_bo *src, uint32_t src_offset,
> > --
> > 2.13.1
> >
> > _______________________________________________
> > mesa-dev mailing list
> > mesa-dev at lists.freedesktop.org
> > https://lists.freedesktop.org/mailman/listinfo/mesa-dev
> >