KonstantinSeurer
/
mesa
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

nvc0: add support for indirect compute on Fermi 

When indirect compute is used, the size of the grid (in blocks) is
stored as three integers inside a buffer. This requires a macro to
set up GRIDDIM_YX and GRIDDIM_Z.

Changes from v2:
 - do not launch the grid if the number of groups for a dimension is 0

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
This commit is contained in:
Samuel Pitoiset 2016-01-12 23:51:00 +01:00
parent fa7333a742
commit c6293877f0
6 changed files with 81 additions and 20 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
src/gallium/drivers/nouveau/nvc0/mme/Makefile
View File

@ -1,5 +1,5 @@
ENVYAS?=envyas
TARGETS=com9097.mme.h
TARGETS=com9097.mme.h com90c0.mme.h
all: $(TARGETS)

24
src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme Normal file
View File

@ -0,0 +1,24 @@
/* NVC0_COMPUTE_MACRO_LAUNCH_GRID_INDIRECT
*
* arg = num_groups_x
* parm[0] = num_groups_y
* parm[1] = num_groups_z
*/
.section #mme90c0_launch_grid_indirect
parm $r2 maddr 0x108e /* GRIDDIM_YX */
braz $r1 #fail
parm $r3
braz annul $r2 #fail
braz annul $r3 #fail
send (extrinsrt $r1 $r2 0x0 0x10 0x10) /* num_groups_y << 16 | num_groups_x */
send $r3
maddrsend 0xa7 /* COMPUTE_BEGIN */
maddrsend 0x282 /* UNKA08 */
maddr 0xda /* LAUNCH */
send 0x1000
maddrsend 0x281 /* COMPUTE_END */
exit maddr 0xd8 /* UNK360 */
send 0x1
fail:
exit
nop

19
src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme.h Normal file
View File

@ -0,0 +1,19 @@
uint32_t mme90c0_launch_grid_indirect[] = {
0x04238251,
0x00034807,
0x00000301,
/* 0x000e: fail */
0x0002d027,
0x00029827,
0x84008842,
0x00001841,
0x0029c071,
0x00a08071,
0x00368021,
0x04000041,
0x00a04071,
0x003600a1,
0x00004041,
0x00000091,
0x00000011,
};

52
src/gallium/drivers/nouveau/nvc0/nvc0_compute.c
View File

@ -360,14 +360,6 @@ nvc0_launch_grid(struct pipe_context *pipe, const struct pipe_grid_info *info)
BEGIN_NVC0(push, NVC0_COMPUTE(CP_GPR_ALLOC), 1);
PUSH_DATA (push, cp->num_gprs);
/* grid/block setup */
BEGIN_NVC0(push, NVC0_COMPUTE(GRIDDIM_YX), 2);
PUSH_DATA (push, (info->grid[1] << 16) | info->grid[0]);
PUSH_DATA (push, info->grid[2]);
BEGIN_NVC0(push, NVC0_COMPUTE(BLOCKDIM_YX), 2);
PUSH_DATA (push, (info->block[1] << 16) | info->block[0]);
PUSH_DATA (push, info->block[2]);
/* launch preliminary setup */
BEGIN_NVC0(push, NVC0_COMPUTE(GRIDID), 1);
PUSH_DATA (push, 0x1);
@ -376,17 +368,39 @@ nvc0_launch_grid(struct pipe_context *pipe, const struct pipe_grid_info *info)
BEGIN_NVC0(push, NVC0_COMPUTE(FLUSH), 1);
PUSH_DATA (push, NVC0_COMPUTE_FLUSH_GLOBAL | NVC0_COMPUTE_FLUSH_UNK8);
/* kernel launching */
BEGIN_NVC0(push, NVC0_COMPUTE(COMPUTE_BEGIN), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_COMPUTE(0x0a08), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, NVC0_COMPUTE(LAUNCH), 1);
PUSH_DATA (push, 0x1000);
BEGIN_NVC0(push, NVC0_COMPUTE(COMPUTE_END), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_COMPUTE(0x0360), 1);
PUSH_DATA (push, 0x1);
/* block setup */
BEGIN_NVC0(push, NVC0_COMPUTE(BLOCKDIM_YX), 2);
PUSH_DATA (push, (info->block[1] << 16) | info->block[0]);
PUSH_DATA (push, info->block[2]);
if (unlikely(info->indirect)) {
struct nv04_resource *res = nv04_resource(info->indirect);
uint32_t offset = res->offset + info->indirect_offset;
unsigned macro = NVC0_COMPUTE_MACRO_LAUNCH_GRID_INDIRECT;
nouveau_pushbuf_space(push, 16, 0, 1);
PUSH_REFN(push, res->bo, NOUVEAU_BO_RD | res->domain);
PUSH_DATA(push, NVC0_FIFO_PKHDR_1I(1, macro, 3));
nouveau_pushbuf_data(push, res->bo, offset,
NVC0_IB_ENTRY_1_NO_PREFETCH | 3 * 4);
} else {
/* grid setup */
BEGIN_NVC0(push, NVC0_COMPUTE(GRIDDIM_YX), 2);
PUSH_DATA (push, (info->grid[1] << 16) | info->grid[0]);
PUSH_DATA (push, info->grid[2]);
/* kernel launching */
BEGIN_NVC0(push, NVC0_COMPUTE(COMPUTE_BEGIN), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_COMPUTE(0x0a08), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, NVC0_COMPUTE(LAUNCH), 1);
PUSH_DATA (push, 0x1000);
BEGIN_NVC0(push, NVC0_COMPUTE(COMPUTE_END), 1);
PUSH_DATA (push, 0);
BEGIN_NVC0(push, SUBC_COMPUTE(0x0360), 1);
PUSH_DATA (push, 0x1);
}
/* Invalidate all 3D constbufs because they are aliased with COMPUTE. */
nvc0->dirty |= NVC0_NEW_CONSTBUF;

2
src/gallium/drivers/nouveau/nvc0/nvc0_macros.h
View File

@ -35,4 +35,6 @@
#define NVC0_3D_MACRO_QUERY_BUFFER_WRITE 0x00003858
#define NVC0_COMPUTE_MACRO_LAUNCH_GRID_INDIRECT 0x00003860
#endif /* __NVC0_MACROS_H__ */

2
src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
View File

@ -36,6 +36,7 @@
#include "nvc0/nvc0_screen.h"
#include "nvc0/mme/com9097.mme.h"
#include "nvc0/mme/com90c0.mme.h"
static boolean
nvc0_screen_is_format_supported(struct pipe_screen *pscreen,
@ -1074,6 +1075,7 @@ nvc0_screen_create(struct nouveau_device *dev)
MK_MACRO(NVC0_3D_MACRO_DRAW_ARRAYS_INDIRECT_COUNT, mme9097_draw_arrays_indirect_count);
MK_MACRO(NVC0_3D_MACRO_DRAW_ELEMENTS_INDIRECT_COUNT, mme9097_draw_elts_indirect_count);
MK_MACRO(NVC0_3D_MACRO_QUERY_BUFFER_WRITE, mme9097_query_buffer_write);
MK_MACRO(NVC0_COMPUTE_MACRO_LAUNCH_GRID_INDIRECT, mme90c0_launch_grid_indirect);
BEGIN_NVC0(push, NVC0_3D(RASTERIZE_ENABLE), 1);
PUSH_DATA (push, 1);