nvc0: add ARB_shader_draw_parameters support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>

14 files changed, 74 insertions, 15 deletions

diff --git a/docs/relnotes/11.2.0.html b/docs/relnotes/11.2.0.html
index 23bb31c6235..d31da8ba4a6 100644
--- a/docs/relnotes/11.2.0.html
+++ b/docs/relnotes/11.2.0.html
@@ -47,6 +47,7 @@ Note: some of the new features are only available with certain drivers.
 <li>GL_ARB_base_instance on freedreno/a4xx</li>
 <li>GL_ARB_compute_shader on i965</li>
 <li>GL_ARB_copy_image on r600</li>
+<li>GL_ARB_shader_draw_parameters on i965, nvc0</li>
 <li>GL_ARB_tessellation_shader on i965 and r600 (evergreen/cayman only)</li>
 <li>GL_ARB_texture_buffer_object_rgb32 on freedreno/a4xx</li>
 <li>GL_ARB_texture_buffer_range on freedreno/a4xx</li>
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir.h b/src/gallium/drivers/nouveau/codegen/nv50_ir.h
index d09a0ab0610..d1fdd75495f 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir.h
@@ -390,6 +390,9 @@ enum SVSemantic
    SV_VERTEX_STRIDE,
    SV_INVOCATION_INFO,
    SV_THREAD_KILL,
+   SV_BASEVERTEX,
+   SV_BASEINSTANCE,
+   SV_DRAWID,
    SV_UNDEFINED,
    SV_LAST
 };
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
index b49bf9d53bc..4504240ac5e 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_driver.h
@@ -124,6 +124,7 @@ struct nv50_ir_prog_info
    union {
       struct {
          uint32_t inputMask[4]; /* mask of attributes read (1 bit per scalar) */
+         bool usesDrawParameters;
       } vp;
       struct {
          uint8_t inputPatchSize;
@@ -160,8 +161,9 @@ struct nv50_ir_prog_info
       uint8_t clipDistances;     /* number of clip distance outputs */
       uint8_t cullDistances;     /* number of cull distance outputs */
       int8_t genUserClip;        /* request user clip planes for ClipVertex */
+      uint8_t auxCBSlot;         /* constant buffer index of UCP/draw data */
       uint16_t ucpBase;          /* base address for UCPs */
-      uint8_t ucpCBSlot;         /* constant buffer index of UCP data */
+      uint16_t drawInfoBase;     /* base address for draw parameters */
       uint8_t pointSize;         /* output index for PointSize */
       uint8_t instanceId;        /* system value index of InstanceID */
       uint8_t vertexId;          /* system value index of VertexID */
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
index b23386040a7..beb67fe20f1 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -377,6 +377,9 @@ static nv50_ir::SVSemantic translateSysVal(uint sysval)
    case TGSI_SEMANTIC_TESSINNER:  return nv50_ir::SV_TESS_INNER;
    case TGSI_SEMANTIC_VERTICESIN: return nv50_ir::SV_VERTEX_COUNT;
    case TGSI_SEMANTIC_HELPER_INVOCATION: return nv50_ir::SV_THREAD_KILL;
+   case TGSI_SEMANTIC_BASEVERTEX: return nv50_ir::SV_BASEVERTEX;
+   case TGSI_SEMANTIC_BASEINSTANCE: return nv50_ir::SV_BASEINSTANCE;
+   case TGSI_SEMANTIC_DRAWID:     return nv50_ir::SV_DRAWID;
    default:
       assert(0);
       return nv50_ir::SV_CLOCK;
@@ -1128,6 +1131,11 @@ bool Source::scanDeclaration(const struct tgsi_full_declaration *decl)
       case TGSI_SEMANTIC_SAMPLEPOS:
          info->prop.fp.sampleInterp = 1;
          break;
+      case TGSI_SEMANTIC_BASEVERTEX:
+      case TGSI_SEMANTIC_BASEINSTANCE:
+      case TGSI_SEMANTIC_DRAWID:
+         info->prop.vp.usesDrawParameters = true;
+         break;
       default:
          break;
       }
@@ -3252,7 +3260,7 @@ Converter::handleUserClipPlanes()
 
    for (c = 0; c < 4; ++c) {
       for (i = 0; i < info->io.genUserClip; ++i) {
-         Symbol *sym = mkSymbol(FILE_MEMORY_CONST, info->io.ucpCBSlot,
+         Symbol *sym = mkSymbol(FILE_MEMORY_CONST, info->io.auxCBSlot,
                                 TYPE_F32, info->io.ucpBase + i * 16 + c * 4);
          Value *ucp = mkLoadv(TYPE_F32, sym, NULL);
          if (c == 0)
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
index e67bf3eca84..6530078b938 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -1576,6 +1576,17 @@ NVC0LoweringPass::handleRDSV(Instruction *i)
       ld = bld.mkOp1(OP_PIXLD, TYPE_U32, i->getDef(0), bld.mkImm(0));
       ld->subOp = NV50_IR_SUBOP_PIXLD_COVMASK;
       break;
+   case SV_BASEVERTEX:
+   case SV_BASEINSTANCE:
+   case SV_DRAWID:
+      ld = bld.mkLoad(TYPE_U32, i->getDef(0),
+                      bld.mkSymbol(FILE_MEMORY_CONST,
+                                   prog->driver->io.auxCBSlot,
+                                   TYPE_U32,
+                                   prog->driver->io.drawInfoBase +
+                                   4 * (sv - SV_BASEVERTEX)),
+                      NULL);
+      break;
    default:
       if (prog->getType() == Program::TYPE_TESSELLATION_EVAL && !i->perPatch)
          vtx = bld.mkOp1v(OP_PFETCH, TYPE_U32, bld.getSSA(), bld.mkImm(0));
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp
index 19637ce33f5..014c652eede 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nvc0.cpp
@@ -295,6 +295,9 @@ TargetNVC0::getSVAddress(DataFile shaderFile, const Symbol *sym) const
    case SV_SAMPLE_INDEX:   return 0;
    case SV_SAMPLE_POS:     return 0;
    case SV_SAMPLE_MASK:    return 0;
+   case SV_BASEVERTEX:     return 0;
+   case SV_BASEINSTANCE:   return 0;
+   case SV_DRAWID:         return 0;
    default:
       return 0xffffffff;
    }
diff --git a/src/gallium/drivers/nouveau/nouveau_compiler.c b/src/gallium/drivers/nouveau/nouveau_compiler.c
index 670b0c8b135..cd44aa1e1d9 100644
--- a/src/gallium/drivers/nouveau/nouveau_compiler.c
+++ b/src/gallium/drivers/nouveau/nouveau_compiler.c
@@ -112,7 +112,7 @@ nouveau_codegen(int chipset, int type, struct tgsi_token tokens[],
    info.bin.sourceRep = NV50_PROGRAM_IR_TGSI;
    info.bin.source = tokens;
 
-   info.io.ucpCBSlot = 15;
+   info.io.auxCBSlot = 15;
    info.io.ucpBase = NV50_CB_AUX_UCP_OFFSET;
 
    info.io.resInfoCBSlot = 15;
diff --git a/src/gallium/drivers/nouveau/nv50/nv50_program.c b/src/gallium/drivers/nouveau/nv50/nv50_program.c
index a4b8ddfda95..b63584e0a09 100644
--- a/src/gallium/drivers/nouveau/nv50/nv50_program.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_program.c
@@ -335,7 +335,7 @@ nv50_program_translate(struct nv50_program *prog, uint16_t chipset,
    info->bin.sourceRep = NV50_PROGRAM_IR_TGSI;
    info->bin.source = (void *)prog->pipe.tokens;
 
-   info->io.ucpCBSlot = 15;
+   info->io.auxCBSlot = 15;
    info->io.ucpBase = NV50_CB_AUX_UCP_OFFSET;
    info->io.genUserClip = prog->vp.clpd_nr;
 
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