gpu: nvgpu: optimize mem_desc accessor loops

Instead of going via gk20a_mem_{wr,rd}32() on each iteration, do direct
memcpy/memset with sysmem, and minimize the enter/exit overhead with
vidmem.

JIRA DNVGPU-23

Change-Id: I5437e35f8393a746777a40636c1e9b5d93ced1f6
Signed-off-by: Konsta Holtta <kholtta@nvidia.com>
Reviewed-on: http://git-master/r/1159524
Reviewed-by: Terje Bergstrom <tbergstrom@nvidia.com>
Tested-by: Terje Bergstrom <tbergstrom@nvidia.com>

1 files changed, 2 insertions, 2 deletions

diff --git a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
index d1628b07..23420fef 100644
--- a/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
+++ b/drivers/gpu/nvgpu/gk20a/mm_gk20a.h
@@ -458,9 +458,9 @@ void gk20a_mem_wr(struct gk20a *g, struct mem_desc *mem, u32 offset, u32 data);
 /* memcpy from cpu, offset and size in bytes (32b-aligned) */
 void gk20a_mem_wr_n(struct gk20a *g, struct mem_desc *mem, u32 offset,
                 void *src, u32 size);
-/* size and offset in bytes (32b-aligned), filled with u32s */
+/* size and offset in bytes (32b-aligned), filled with the constant byte c */
 void gk20a_memset(struct gk20a *g, struct mem_desc *mem, u32 offset,
-                u32 value, u32 size);
+                u32 c, u32 size);
 
 #if 0 /*related to addr bits above, concern below TBD on which is accurate */
 #define bar1_instance_block_shift_gk20a() (max_physaddr_bits_gk20a() -\