nv50,nvc0: enable TEX_LZ and TXF_LZ

There should be minimal gain, if any, for nvc0, but nv50 may end up
noticing more often that the lod argument is uniform. This, in turn,
will remove the need for some unnecessary transformations, which were
being hit due to the checks being done pre-ssa.

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

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 fdaa628fb24b55abb12768867f687854fee7379f..7aaeedf8dde45e7dc059f932378e0078414820b3 100644 (file)
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -305,6 +305,8 @@ unsigned int Instruction::srcMask(unsigned int s) const
    case TGSI_OPCODE_TXD:
    case TGSI_OPCODE_TXL:
    case TGSI_OPCODE_TXP:
+   case TGSI_OPCODE_TEX_LZ:
+   case TGSI_OPCODE_TXF_LZ:
    case TGSI_OPCODE_LODQ:
    {
       const struct tgsi_instruction_texture *tex = &insn->Texture;
@@ -313,6 +315,8 @@ unsigned int Instruction::srcMask(unsigned int s) const
 
       mask = 0x7;
       if (insn->Instruction.Opcode != TGSI_OPCODE_TEX &&
+          insn->Instruction.Opcode != TGSI_OPCODE_TEX_LZ &&
+          insn->Instruction.Opcode != TGSI_OPCODE_TXF_LZ &&
           insn->Instruction.Opcode != TGSI_OPCODE_TXD)
          mask |= 0x8; /* bias, lod or proj */
 
@@ -820,6 +824,7 @@ static nv50_ir::operation translateOpcode(uint opcode)
    NV50_IR_OPCODE_CASE(DIV, DIV);
 
    NV50_IR_OPCODE_CASE(TXL, TXL);
+   NV50_IR_OPCODE_CASE(TEX_LZ, TXL);
 
    NV50_IR_OPCODE_CASE(CEIL, CEIL);
    NV50_IR_OPCODE_CASE(I2F, CVT);
@@ -833,6 +838,7 @@ static nv50_ir::operation translateOpcode(uint opcode)
    NV50_IR_OPCODE_CASE(XOR, XOR);
    NV50_IR_OPCODE_CASE(SAD, SAD);
    NV50_IR_OPCODE_CASE(TXF, TXF);
+   NV50_IR_OPCODE_CASE(TXF_LZ, TXF);
    NV50_IR_OPCODE_CASE(TXQ, TXQ);
    NV50_IR_OPCODE_CASE(TXQS, TXQ);
    NV50_IR_OPCODE_CASE(TG4, TXG);
@@ -2283,7 +2289,9 @@ Converter::handleTEX(Value *dst[4], int R, int S, int L, int C, int Dx, int Dy)
    for (s = 0; s < tgt.getArgCount(); ++s)
       arg[s] = src[s] = fetchSrc(0, s);
 
-   if (texi->op == OP_TXL || texi->op == OP_TXB)
+   if (tgsi.getOpcode() == TGSI_OPCODE_TEX_LZ)
+      lod = loadImm(NULL, 0);
+   else if (texi->op == OP_TXL || texi->op == OP_TXB)
       lod = fetchSrc(L >> 4, L & 3);
 
    if (C == 0x0f)
@@ -2375,7 +2383,10 @@ Converter::handleTXF(Value *dst[4], int R, int L_M)
    }
    for (c = 0; c < (texi->tex.target.getArgCount() - ms); ++c)
       texi->setSrc(c, fetchSrc(0, c));
-   texi->setSrc(c++, fetchSrc(L_M >> 4, L_M & 3)); // lod or ms
+   if (!ms && tgsi.getOpcode() == TGSI_OPCODE_TXF_LZ)
+      texi->setSrc(c++, loadImm(NULL, 0));
+   else
+      texi->setSrc(c++, fetchSrc(L_M >> 4, L_M & 3)); // lod or ms
 
    setTexRS(texi, c, R, -1);
 
@@ -3415,6 +3426,7 @@ Converter::handleInstruction(const struct tgsi_full_instruction *insn)
       mkOp(OP_DISCARD, TYPE_NONE, NULL);
       break;
    case TGSI_OPCODE_TEX:
+   case TGSI_OPCODE_TEX_LZ:
    case TGSI_OPCODE_TXB:
    case TGSI_OPCODE_TXL:
    case TGSI_OPCODE_TXP:
@@ -3443,6 +3455,7 @@ Converter::handleInstruction(const struct tgsi_full_instruction *insn)
    case TGSI_OPCODE_SAMPLE_C_LZ:
       handleTEX(dst0, 1, 2, 0x30, 0x30, 0x30, 0x40);
       break;
+   case TGSI_OPCODE_TXF_LZ:
    case TGSI_OPCODE_TXF:
       handleTXF(dst0, 1, 0x03);
       break;

diff --git a/src/gallium/drivers/nouveau/nv50/nv50_screen.c b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
index f50dcfa3043eddeea91c6782d58c9cba70c7d3f4..249947ada042b8bfd6c4a679cb90b2d07d1ac02e 100644 (file)
--- a/src/gallium/drivers/nouveau/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nouveau/nv50/nv50_screen.c
@@ -198,6 +198,7 @@ nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_CULL_DISTANCE:
    case PIPE_CAP_TGSI_ARRAY_COMPONENTS:
    case PIPE_CAP_TGSI_MUL_ZERO_WINS:
+   case PIPE_CAP_TGSI_TEX_TXF_LZ:
       return 1;
    case PIPE_CAP_SEAMLESS_CUBE_MAP:
       return 1; /* class_3d >= NVA0_3D_CLASS; */
@@ -262,7 +263,6 @@ nv50_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_DOUBLES:
    case PIPE_CAP_INT64:
    case PIPE_CAP_INT64_DIVMOD:
-   case PIPE_CAP_TGSI_TEX_TXF_LZ:
       return 0;
 
    case PIPE_CAP_VENDOR_ID:

diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index 3e4c4f44ba92abe2fd9f1eefc3e747029fe0fedf..cfe4f677daa3c2ef7b7aa8f441e16c1e25b86dbd 100644 (file)
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -246,6 +246,7 @@ nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_TGSI_MUL_ZERO_WINS:
    case PIPE_CAP_DOUBLES:
    case PIPE_CAP_INT64:
+   case PIPE_CAP_TGSI_TEX_TXF_LZ:
       return 1;
    case PIPE_CAP_COMPUTE:
       return (class_3d < GP100_3D_CLASS);
@@ -284,7 +285,6 @@ nvc0_screen_get_param(struct pipe_screen *pscreen, enum pipe_cap param)
    case PIPE_CAP_NATIVE_FENCE_FD:
    case PIPE_CAP_GLSL_OPTIMIZE_CONSERVATIVELY:
    case PIPE_CAP_INT64_DIVMOD:
-   case PIPE_CAP_TGSI_TEX_TXF_LZ:
       return 0;
 
    case PIPE_CAP_VENDOR_ID: