[Mesa-dev] [PATCH 07/12] nvc0: add support for indirect compute on Fermi

[Samuel Pitoiset]

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.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset at gmail.com>
---
 src/gallium/drivers/nouveau/nvc0/mme/Makefile      |  2 +-
 src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme   | 19 +++++++++++++++++++
 src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme.h | 13 +++++++++++++
 src/gallium/drivers/nouveau/nvc0/nvc0_compute.c    | 18 +++++++++++++++---
 src/gallium/drivers/nouveau/nvc0/nvc0_macros.h     |  2 ++
 src/gallium/drivers/nouveau/nvc0/nvc0_screen.c     |  2 ++
 6 files changed, 52 insertions(+), 4 deletions(-)
 create mode 100644 src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme
 create mode 100644 src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme.h

diff --git a/src/gallium/drivers/nouveau/nvc0/mme/Makefile b/src/gallium/drivers/nouveau/nvc0/mme/Makefile
index 1c0f583..52fb0a5 100644
--- a/src/gallium/drivers/nouveau/nvc0/mme/Makefile
+++ b/src/gallium/drivers/nouveau/nvc0/mme/Makefile
@@ -1,5 +1,5 @@
 ENVYAS?=envyas
-TARGETS=com9097.mme.h
+TARGETS=com9097.mme.h com90c0.mme.h
 
 all: $(TARGETS)
 
diff --git a/src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme b/src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme
new file mode 100644
index 0000000..ee7f726
--- /dev/null
+++ b/src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme
@@ -0,0 +1,19 @@
+/* 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
+   parm $r3
+   mov $r4 (or $r1 $r2)
+   mov $r4 (or $r3 $r4)
+   braz $r4 #fail
+   maddr 0x108e /* GRIDDIM_YX */
+   mov $r4 (extrshl $r2 $r0 0x10 0x10)
+   exit send (or $r4 $r1) /* (num_groups_y << 16) | num_groups_x */
+   send $r3
+fail:
+   exit
+
diff --git a/src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme.h b/src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme.h
new file mode 100644
index 0000000..89076cf
--- /dev/null
+++ b/src/gallium/drivers/nouveau/nvc0/mme/com90c0.mme.h
@@ -0,0 +1,13 @@
+uint32_t mme90c0_launch_grid_indirect[] = {
+	0x00000201,
+	0x00000301,
+/* 0x0009: fail */
+	0x00128c10,
+	0x00131c10,
+	0x00016007,
+	0x04238021,
+	0x84008413,
+	0x001260c0,
+	0x00001841,
+	0x00000091,
+};
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c b/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c
index e63bdcb..dbf2148 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_compute.c
@@ -452,9 +452,21 @@ nvc0_launch_grid(struct pipe_context *pipe, const struct pipe_grid_info *info)
    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]);
+   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 {
+      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]);
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_macros.h b/src/gallium/drivers/nouveau/nvc0/nvc0_macros.h
index 49e176c..57262fe 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_macros.h
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_macros.h
@@ -35,4 +35,6 @@
 
 #define NVC0_3D_MACRO_QUERY_BUFFER_WRITE			0x00003858
 
+#define NVC0_COMPUTE_MACRO_LAUNCH_GRID_INDIRECT       0x00003860
+
 #endif /* __NVC0_MACROS_H__ */
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index 84e4253..85be1cc 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -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,
@@ -1053,6 +1054,7 @@ 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);
-- 
2.6.4