nvc0/ir: implement lowering of surface ops for nve4

6 files changed, 30 insertions, 13 deletions

diff --git a/src/gallium/drivers/nv50/codegen/nv50_ir.cpp b/src/gallium/drivers/nv50/codegen/nv50_ir.cpp
index 6a391211dc9..3c05d05b81f 100644
--- a/src/gallium/drivers/nv50/codegen/nv50_ir.cpp
+++ b/src/gallium/drivers/nv50/codegen/nv50_ir.cpp
@@ -932,7 +932,7 @@ const struct TexInstruction::Target::Desc TexInstruction::Target::descTable[] =
 {
    { "1D",                1, 1, false, false, false },
    { "2D",                2, 2, false, false, false },
-   { "2D_MS",             2, 2, false, false, false },
+   { "2D_MS",             2, 3, false, false, false },
    { "3D",                3, 3, false, false, false },
    { "CUBE",              2, 3, false, true,  false },
    { "1D_SHADOW",         1, 1, false, false, true  },
@@ -940,7 +940,7 @@ const struct TexInstruction::Target::Desc TexInstruction::Target::descTable[] =
    { "CUBE_SHADOW",       2, 3, false, true,  true  },
    { "1D_ARRAY",          1, 2, true,  false, false },
    { "2D_ARRAY",          2, 3, true,  false, false },
-   { "2D_MS_ARRAY",       2, 3, true,  false, false },
+   { "2D_MS_ARRAY",       2, 4, true,  false, false },
    { "CUBE_ARRAY",        2, 4, true,  true,  false },
    { "1D_ARRAY_SHADOW",   1, 2, true,  false, true  },
    { "2D_ARRAY_SHADOW",   2, 3, true,  false, true  },
@@ -1137,6 +1137,7 @@ nv50_ir_generate_code(struct nv50_ir_prog_info *info)
    nv50_ir::Program *prog = new nv50_ir::Program(type, targ);
    if (!prog)
       return -1;
+   prog->driver = info;
    prog->dbgFlags = info->dbgFlags;
    prog->optLevel = info->optLevel;
 
diff --git a/src/gallium/drivers/nv50/codegen/nv50_ir.h b/src/gallium/drivers/nv50/codegen/nv50_ir.h
index 637196e4518..bdea48bbdf3 100644
--- a/src/gallium/drivers/nv50/codegen/nv50_ir.h
+++ b/src/gallium/drivers/nv50/codegen/nv50_ir.h
@@ -829,6 +829,9 @@ public:
       }
 
       inline bool operator==(TexTarget targ) const { return target == targ; }
+      inline bool operator!=(TexTarget targ) const { return target != targ; }
+
+      enum TexTarget getEnum() const { return target; }
 
    private:
       struct Desc
@@ -1149,6 +1152,8 @@ public:
 
    void *targetPriv; // e.g. to carry information between passes
 
+   const struct nv50_ir_prog_info *driver; // for driver configuration
+
    void releaseInstruction(Instruction *);
    void releaseValue(Value *);
 };
diff --git a/src/gallium/drivers/nv50/codegen/nv50_ir_driver.h b/src/gallium/drivers/nv50/codegen/nv50_ir_driver.h
index 446befa5fb7..deee60cd6a7 100644
--- a/src/gallium/drivers/nv50/codegen/nv50_ir_driver.h
+++ b/src/gallium/drivers/nv50/codegen/nv50_ir_driver.h
@@ -169,7 +169,7 @@ struct nv50_ir_prog_info
       uint8_t cullDistanceMask;  /* clip distance mode (1 bit per output) */
       int8_t genUserClip;        /* request user clip planes for ClipVertex */
       uint16_t ucpBase;          /* base address for UCPs */
-      uint8_t ucpBinding;        /* constant buffer index of UCP data */
+      uint8_t ucpCBSlot;         /* constant buffer index of UCP data */
       uint8_t pointSize;         /* output index for PointSize */
       uint8_t instanceId;        /* system value index of InstanceID */
       uint8_t vertexId;          /* system value index of VertexID */
@@ -179,6 +179,11 @@ struct nv50_ir_prog_info
       uint8_t sampleMask;        /* output index of SampleMask */
       uint8_t backFaceColor[2];  /* input/output indices of back face colour */
       uint8_t globalAccess;      /* 1 for read, 2 for wr, 3 for rw */
+      uint8_t resInfoCBSlot;     /* cX[] used for tex handles, surface info */
+      uint16_t texBindBase;      /* base address for tex handles (nve4) */
+      uint16_t suInfoBase;       /* base address for surface info (nve4) */
+      uint8_t msInfoCBSlot;      /* cX[] used for multisample info */
+      uint16_t msInfoBase;       /* base address for multisample info */
    } io;
 
    /* driver callback to assign input/output locations */
diff --git a/src/gallium/drivers/nv50/codegen/nv50_ir_from_tgsi.cpp b/src/gallium/drivers/nv50/codegen/nv50_ir_from_tgsi.cpp
index 32915174e01..69c05c1464c 100644
--- a/src/gallium/drivers/nv50/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nv50/codegen/nv50_ir_from_tgsi.cpp
@@ -2285,7 +2285,7 @@ Converter::handleUserClipPlanes()
 
    for (c = 0; c < 4; ++c) {
       for (i = 0; i < info->io.genUserClip; ++i) {
-         Symbol *sym = mkSymbol(FILE_MEMORY_CONST, info->io.ucpBinding,
+         Symbol *sym = mkSymbol(FILE_MEMORY_CONST, info->io.ucpCBSlot,
                                 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/nv50/codegen/nv50_ir_ra.cpp b/src/gallium/drivers/nv50/codegen/nv50_ir_ra.cpp
index ed991dac30b..bf2380d2648 100644
--- a/src/gallium/drivers/nv50/codegen/nv50_ir_ra.cpp
+++ b/src/gallium/drivers/nv50/codegen/nv50_ir_ra.cpp
@@ -1852,17 +1852,23 @@ RegAlloc::InsertConstraintsPass::condenseSrcs(Instruction *insn,
 void
 RegAlloc::InsertConstraintsPass::texConstraintNVE0(TexInstruction *tex)
 {
-   textureMask(tex);
+   if (isTextureOp(tex->op))
+      textureMask(tex);
    condenseDefs(tex);
 
-   int n = tex->srcCount(0xff, true);
-   if (n > 4) {
-      condenseSrcs(tex, 0, 3);
-      if (n > 5) // NOTE: first call modified positions already
-         condenseSrcs(tex, 4 - (4 - 1), n - 1 - (4 - 1));
+   if (tex->op == OP_SUSTB || tex->op == OP_SUSTP) {
+      condenseSrcs(tex, 3, (3 + typeSizeof(tex->dType) / 4) - 1);
    } else
-   if (n > 1) {
-      condenseSrcs(tex, 0, n - 1);
+   if (isTextureOp(tex->op)) {
+      int n = tex->srcCount(0xff, true);
+      if (n > 4) {
+         condenseSrcs(tex, 0, 3);
+         if (n > 5) // NOTE: first call modified positions already
+            condenseSrcs(tex, 4 - (4 - 1), n - 1 - (4 - 1));
+      } else
+      if (n > 1) {
+         condenseSrcs(tex, 0, n - 1);
+      }
    }
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 10810bf3e07..c17ffdc4111 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -315,7 +315,7 @@ 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.ucpBinding = 15;
+   info->io.ucpCBSlot = 15;
    info->io.ucpBase = 0;
    info->io.genUserClip = prog->vp.clpd_nr;