gallium: add new LOD opcode

The operation performed is all the same as LODQ, but with the usual
differences between dx10 and GL texture opcodes, that is separate resource
and sampler indices (plus result swizzling, and setting z/w channels
to zero).

Reviewed-by: Jose Fonseca <[email protected]>
Acked-by: Nicolai Hähnle <[email protected]>

1 files changed, 44 insertions, 4 deletions

diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
index 9c019a311d7..afed96c9b1d 100644
--- a/src/gallium/auxiliary/tgsi/tgsi_exec.c
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -2351,15 +2351,22 @@ static void
 exec_lodq(struct tgsi_exec_machine *mach,
           const struct tgsi_full_instruction *inst)
 {
-   uint unit;
+   uint resource_unit, sampler_unit;
    int dim;
    int i;
    union tgsi_exec_channel coords[4];
    const union tgsi_exec_channel *args[ARRAY_SIZE(coords)];
    union tgsi_exec_channel r[2];
 
-   unit = fetch_sampler_unit(mach, inst, 1);
-   dim = tgsi_util_get_texture_coord_dim(inst->Texture.Texture);
+   resource_unit = fetch_sampler_unit(mach, inst, 1);
+   if (inst->Instruction.Opcode == TGSI_OPCODE_LOD) {
+      uint target = mach->SamplerViews[resource_unit].Resource;
+      dim = tgsi_util_get_texture_coord_dim(target);
+      sampler_unit = fetch_sampler_unit(mach, inst, 2);
+   } else {
+      dim = tgsi_util_get_texture_coord_dim(inst->Texture.Texture);
+      sampler_unit = resource_unit;
+   }
    assert(dim <= ARRAY_SIZE(coords));
    /* fetch coordinates */
    for (i = 0; i < dim; i++) {
@@ -2369,7 +2376,7 @@ exec_lodq(struct tgsi_exec_machine *mach,
    for (i = dim; i < ARRAY_SIZE(coords); i++) {
       args[i] = &ZeroVec;
    }
-   mach->Sampler->query_lod(mach->Sampler, unit, unit,
+   mach->Sampler->query_lod(mach->Sampler, resource_unit, sampler_unit,
                             args[0]->f,
                             args[1]->f,
                             args[2]->f,
@@ -2386,6 +2393,35 @@ exec_lodq(struct tgsi_exec_machine *mach,
       store_dest(mach, &r[1], &inst->Dst[0], inst, TGSI_CHAN_Y,
                  TGSI_EXEC_DATA_FLOAT);
    }
+   if (inst->Instruction.Opcode == TGSI_OPCODE_LOD) {
+      unsigned char swizzles[4];
+      unsigned chan;
+      swizzles[0] = inst->Src[1].Register.SwizzleX;
+      swizzles[1] = inst->Src[1].Register.SwizzleY;
+      swizzles[2] = inst->Src[1].Register.SwizzleZ;
+      swizzles[3] = inst->Src[1].Register.SwizzleW;
+
+      for (chan = 0; chan < TGSI_NUM_CHANNELS; chan++) {
+         if (inst->Dst[0].Register.WriteMask & (1 << chan)) {
+            if (swizzles[chan] >= 2) {
+               store_dest(mach, &ZeroVec,
+                          &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+            } else {
+               store_dest(mach, &r[swizzles[chan]],
+                          &inst->Dst[0], inst, chan, TGSI_EXEC_DATA_FLOAT);
+            }
+         }
+      }
+   } else {
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_X) {
+         store_dest(mach, &r[0], &inst->Dst[0], inst, TGSI_CHAN_X,
+                    TGSI_EXEC_DATA_FLOAT);
+      }
+      if (inst->Dst[0].Register.WriteMask & TGSI_WRITEMASK_Y) {
+         store_dest(mach, &r[1], &inst->Dst[0], inst, TGSI_CHAN_Y,
+                    TGSI_EXEC_DATA_FLOAT);
+      }
+   }
 }
 
 static void
@@ -5718,6 +5754,10 @@ exec_instruction(
       assert(0);
       break;
 
+   case TGSI_OPCODE_LOD:
+      exec_lodq(mach, inst);
+      break;
+
    case TGSI_OPCODE_UARL:
       exec_vector_unary(mach, inst, micro_uarl, TGSI_EXEC_DATA_INT, TGSI_EXEC_DATA_UINT);
       break;