Just general suggestions.
If you don't want to map the buffer and synchronize with the CPU, you
can use the WRITE_DATA packet to write the data.
I wouldn't use buffer_unmap for buffers that can be mapped/unmapped
often. Even though buffer_unmap is a no-op right now, it might change
in the future.
Marek
On Fri, Jan 16, 2015 at 1:48 AM, Tom Stellard <thomas.stell...@amd.com> wrote:
> Instead of passing a pointer to the scratch buffer via user sgprs, we
> now patch the shader with the buffer address using reloc information
> from the LLVM generated ELF.
>
> v2:
> - Make sure not to break older LLVM.
> ---
> src/gallium/drivers/radeonsi/si_compute.c | 42
> +++++++++++++++++++++++++++++++
> 1 file changed, 42 insertions(+)
>
> diff --git a/src/gallium/drivers/radeonsi/si_compute.c
> b/src/gallium/drivers/radeonsi/si_compute.c
> index f911e1f..9571a3e 100644
> --- a/src/gallium/drivers/radeonsi/si_compute.c
> +++ b/src/gallium/drivers/radeonsi/si_compute.c
> @@ -36,9 +36,18 @@
> #if HAVE_LLVM < 0x0305
> #define NUM_USER_SGPRS 2
> #else
> +/* XXX: Even though we don't pass the scratch buffer via user sgprs any more
> + * LLVM still expects that we specify 4 USER_SGPRS so it can remain
> compatible
> + * with older mesa. */
> #define NUM_USER_SGPRS 4
> #endif
>
> +static const char *scratch_rsrc_dword0_symbol =
> + "SCRATCH_RSRC_DWORD0";
> +
> +static const char *scratch_rsrc_dword1_symbol =
> + "SCRATCH_RSRC_DWORD1";
> +
> struct si_compute {
> struct si_context *ctx;
>
> @@ -174,6 +183,35 @@ static unsigned compute_num_waves_for_scratch(
> return scratch_waves;
> }
>
> +static void apply_scratch_relocs(const struct si_screen *sscreen,
> + const struct radeon_shader_binary *binary,
> + struct si_shader *shader, uint64_t scratch_va) {
> + unsigned i;
> + char *ptr;
> + uint32_t scratch_rsrc_dword0 = scratch_va & 0xffffffff;
> + uint32_t scratch_rsrc_dword1 =
> + S_008F04_BASE_ADDRESS_HI(scratch_va >> 32)
> + | S_008F04_STRIDE(shader->scratch_bytes_per_wave / 64);
> +
> + if (!binary->reloc_count) {
> + return;
> + }
> +
> + ptr = sscreen->b.ws->buffer_map(shader->bo->cs_buf, NULL,
> + PIPE_TRANSFER_READ_WRITE);
> + for (i = 0 ; i < binary->reloc_count; i++) {
> + const struct radeon_shader_reloc *reloc = &binary->relocs[i];
> + if (!strcmp(scratch_rsrc_dword0_symbol, reloc->name)) {
> + util_memcpy_cpu_to_le32(ptr + reloc->offset,
> + &scratch_rsrc_dword0, 4);
> + } else if (!strcmp(scratch_rsrc_dword1_symbol, reloc->name)) {
> + util_memcpy_cpu_to_le32(ptr + reloc->offset,
> + &scratch_rsrc_dword1, 4);
> + }
> + }
> + sscreen->b.ws->buffer_unmap(shader->bo->cs_buf);
> +}
> +
> static void si_launch_grid(
> struct pipe_context *ctx,
> const uint *block_layout, const uint *grid_layout,
> @@ -256,6 +294,10 @@ static void si_launch_grid(
> RADEON_USAGE_READWRITE,
> RADEON_PRIO_SHADER_RESOURCE_RW);
>
> + /* Patch the shader with the scratch buffer address. */
> + apply_scratch_relocs(sctx->screen,
> + &program->binary, shader, scratch_buffer_va);
> +
> }
>
> for (i = 0; i < (kernel_args_size / 4); i++) {
> --
> 1.8.5.5
>
> _______________________________________________
> mesa-dev mailing list
> mesa-dev@lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/mesa-dev
_______________________________________________
mesa-dev mailing list
mesa-dev@lists.freedesktop.org
http://lists.freedesktop.org/mailman/listinfo/mesa-dev
