nvc0: add ARB_shader_draw_parameters support

Signed-off-by: Ilia Mirkin <[email protected]>

6 files changed, 42 insertions, 11 deletions

diff --git a/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme b/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme
index b2060d1fa53..8c647d0c66c 100644
--- a/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme
+++ b/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme
@@ -241,8 +241,10 @@ locn_0f_ts:
    parm $r2 /* instance_count */
    parm $r4 maddr 0x5f7 /* INDEX_BATCH_FIRST, start */
    parm $r4 send $r4 /* index_bias, send start */
+   maddr 0x8e4 /* CB_DATA */
    braz $r2 #dei_end
-   parm $r5 /* start_instance */
+   parm $r5 send $r4 /* start_instance, send index_bias */
+   send $r5 /* send start_instance */
    read $r6 0x50d /* VB_ELEMENT_BASE */
    read $r7 0x50e /* VB_INSTANCE_BASE */
    maddr 0x150d /* VB_ELEMENT,INSTANCE_BASE */
@@ -283,8 +285,10 @@ dei_end:
    parm $r2 /* count */
    parm $r3 /* instance_count */
    parm $r4 maddr 0x35d /* VERTEX_BUFFER_FIRST, start */
-   parm $r4 send $r4 /* start_instance */
    braz $r3 #dai_end
+   parm $r4 send $r4 /* start_instance */
+   maddrsend 0x8e4 /* CB_DATA, send 0 as base_vertex */
+   send $r4 /* send start_instance */
    read $r6 0x50e /* VB_INSTANCE_BASE */
    maddr 0x50e /* VB_INSTANCE_BASE */
    mov $r5 0x1
diff --git a/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme.h b/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme.h
index bac9042c2df..acad303ce60 100644
--- a/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme.h
+++ b/src/gallium/drivers/nouveau/nvc0/mme/com9097.mme.h
@@ -128,11 +128,13 @@ uint32_t mme9097_draw_elts_indirect[] = {
 	0x00000301,
 	0x00000201,
 	0x017dc451,
-/* 0x000e: dei_again */
 	0x00002431,
-	0x0005d007,
-	0x00000501,
-/* 0x001b: dei_end */
+/* 0x0010: dei_again */
+	0x02390021,
+	0x00061007,
+	0x00002531,
+/* 0x001d: dei_end */
+	0x00002841,
 	0x01434615,
 	0x01438715,
 	0x05434021,
@@ -161,11 +163,13 @@ uint32_t mme9097_draw_elts_indirect[] = {
 uint32_t mme9097_draw_arrays_indirect[] = {
 	0x00000201,
 	0x00000301,
-/* 0x0009: dai_again */
+/* 0x000b: dai_again */
 	0x00d74451,
+	0x00049807,
 	0x00002431,
-/* 0x0013: dai_end */
-	0x0003d807,
+/* 0x0015: dai_end */
+	0x02390071,
+	0x00002041,
 	0x01438615,
 	0x01438021,
 	0x00004511,
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_program.c b/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
index 67a25acf778..730d6feac69 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_program.c
@@ -533,8 +533,9 @@ nvc0_program_translate(struct nvc0_program *prog, uint16_t chipset,
    info->bin.source = (void *)prog->pipe.tokens;
 
    info->io.genUserClip = prog->vp.num_ucps;
+   info->io.auxCBSlot = 15;
    info->io.ucpBase = 256;
-   info->io.ucpCBSlot = 15;
+   info->io.drawInfoBase = 256 + 128;
 
    if (prog->type == PIPE_SHADER_COMPUTE) {
       if (chipset >= NVISA_GK104_CHIPSET) {
@@ -583,6 +584,7 @@ nvc0_program_translate(struct nvc0_program *prog, uint16_t chipset,
    prog->num_barriers = info->numBarriers;
 
    prog->vp.need_vertex_id = info->io.vertexId < PIPE_MAX_SHADER_INPUTS;
+   prog->vp.need_draw_parameters = info->prop.vp.usesDrawParameters;
 
    if (info->io.edgeFlagOut < PIPE_MAX_ATTRIBS)
       info->out[info->io.edgeFlagOut].mask = 0; /* for headergen */
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_program.h b/src/gallium/drivers/nouveau/nvc0/nvc0_program.h
index 9c45e7b3e31..8b8d221edfc 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_program.h
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_program.h
@@ -42,6 +42,7 @@ struct nvc0_program {
       uint8_t num_ucps; /* also set to max if ClipDistance is used */
       uint8_t edgeflag; /* attribute index of edgeflag input */
       bool need_vertex_id;
+      bool need_draw_parameters;
    } vp;
    struct {
       uint8_t early_z;
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index 500510a0870..f029d164436 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -184,6 +184,7 @@ nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_FORCE_PERSAMPLE_INTERP:
    case PIPE_CAP_SHAREABLE_SHADERS:
    case PIPE_CAP_CLEAR_TEXTURE:
+   case PIPE_CAP_DRAW_PARAMETERS:
       return 1;
    case PIPE_CAP_SEAMLESS_CUBE_MAP_PER_TEXTURE:
       return (class_3d >= NVE4_3D_CLASS) ? 1 : 0;
@@ -206,7 +207,6 @@ nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_VERTEXID_NOBASE:
    case PIPE_CAP_RESOURCE_FROM_USER_MEMORY:
    case PIPE_CAP_DEVICE_RESET_STATUS_QUERY:
-   case PIPE_CAP_DRAW_PARAMETERS:
       return 0;
 
    case PIPE_CAP_VENDOR_ID:
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
index 54443bdccc0..273451e638d 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_vbo.c
@@ -814,6 +814,14 @@ nvc0_draw_indirect(struct nvc0_context *nvc0, const struct pipe_draw_info *info)
    if (buf->fence_wr && !nouveau_fence_signalled(buf->fence_wr))
       IMMED_NVC0(push, SUBC_3D(NV10_SUBCHAN_REF_CNT), 0);
 
+   /* Queue things up to let the macros write params to the driver constbuf */
+   BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
+   PUSH_DATA (push, 512);
+   PUSH_DATAh(push, nvc0->screen->uniform_bo->offset + (5 << 16) + (0 << 9));
+   PUSH_DATA (push, nvc0->screen->uniform_bo->offset + (5 << 16) + (0 << 9));
+   BEGIN_NVC0(push, NVC0_3D(CB_POS), 1);
+   PUSH_DATA (push, 256 + 128);
+
    PUSH_SPACE(push, 8);
    if (info->indexed) {
       assert(nvc0->idxbuf.buffer);
@@ -901,6 +909,18 @@ nvc0_draw_vbo(struct pipe_context *pipe, const struct pipe_draw_info *info)
    /* 8 as minimum to avoid immediate double validation of new buffers */
    nvc0_state_validate(nvc0, ~0, 8);
 
+   if (nvc0->vertprog->vp.need_draw_parameters) {
+      BEGIN_NVC0(push, NVC0_3D(CB_SIZE), 3);
+      PUSH_DATA (push, 512);
+      PUSH_DATAh(push, nvc0->screen->uniform_bo->offset + (5 << 16) + (0 << 9));
+      PUSH_DATA (push, nvc0->screen->uniform_bo->offset + (5 << 16) + (0 << 9));
+      BEGIN_1IC0(push, NVC0_3D(CB_POS), 1 + 3);
+      PUSH_DATA (push, 256 + 128);
+      PUSH_DATA (push, info->index_bias);
+      PUSH_DATA (push, info->start_instance);
+      PUSH_DATA (push, info->drawid);
+   }
+
    push->kick_notify = nvc0_draw_vbo_kick_notify;
 
    /* TODO: Instead of iterating over all the buffer resources looking for