nv50/ir: change texture offsets to ValueRefs, allow nonconst

This allows us to have non-constant offsets for textureGatherOffset and
textureGatherOffsets.

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

8 files changed, 61 insertions, 20 deletions

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir.cpp
index a24a66c77a3..13f8cf20c76 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir.cpp
@@ -911,6 +911,9 @@ TexInstruction::~TexInstruction()
       dPdx[c].set(NULL);
       dPdy[c].set(NULL);
    }
+   for (int n = 0; n < 4; ++n)
+      for (int c = 0; c < 3; ++c)
+         offset[n][c].set(NULL);
 }
 
 TexInstruction *
@@ -930,6 +933,10 @@ TexInstruction::clone(ClonePolicy<Function>& pol, Instruction *i) const
       }
    }
 
+   for (int n = 0; n < tex->tex.useOffsets; ++n)
+      for (int c = 0; c < 3; ++c)
+         tex->offset[n][c].set(offset[n][c]);
+
    return tex;
 }
 
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir.h b/src/gallium/drivers/nouveau/codegen/nv50_ir.h
index 919d3a4c7bd..8872cfb7025 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir.h
@@ -913,13 +913,14 @@ public:
       bool derivAll;
 
       int8_t useOffsets; // 0, 1, or 4 for textureGatherOffsets
-      int8_t offset[4][3];
+      int8_t offset[3]; // only used on nv50
 
       enum TexQuery query;
    } tex;
 
    ValueRef dPdx[3];
    ValueRef dPdy[3];
+   ValueRef offset[4][3];
 };
 
 class CmpInstruction : public Instruction
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
index 63d55252b54..5992c543d69 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_gk110.cpp
@@ -1104,12 +1104,14 @@ CodeEmitterGK110::emitTEX(const TexInstruction *i)
       // ?
    }
 
-   if (i->tex.useOffsets) {
+   if (i->tex.useOffsets == 1) {
       switch (i->op) {
       case OP_TXF: code[1] |= 0x200; break;
       default: code[1] |= 0x800; break;
       }
    }
+   if (i->tex.useOffsets == 4)
+      code[1] |= 0x1000;
 }
 
 void
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
index 9eccd9f0ccd..077eba8e3d4 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nv50.cpp
@@ -1475,9 +1475,9 @@ CodeEmitterNV50::emitTEX(const TexInstruction *i)
       code[0] |= 0x08000000;
    } else
    if (i->tex.useOffsets) {
-      code[1] |= (i->tex.offset[0][0] & 0xf) << 24;
-      code[1] |= (i->tex.offset[0][1] & 0xf) << 20;
-      code[1] |= (i->tex.offset[0][2] & 0xf) << 16;
+      code[1] |= (i->tex.offset[0] & 0xf) << 24;
+      code[1] |= (i->tex.offset[1] & 0xf) << 20;
+      code[1] |= (i->tex.offset[2] & 0xf) << 16;
    }
 
    code[0] |= (i->tex.mask & 0x3) << 25;
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
index 11a7c2b2bc7..b1f76cf8043 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_emit_nvc0.cpp
@@ -1208,8 +1208,10 @@ CodeEmitterNVC0::emitTEX(const TexInstruction *i)
        i->tex.target == TEX_TARGET_2D_MS_ARRAY)
       code[1] |= 1 << 23;
 
-   if (i->tex.useOffsets) // in vecSrc0.w
+   if (i->tex.useOffsets == 1)
       code[1] |= 1 << 22;
+   if (i->tex.useOffsets == 4)
+      code[1] |= 1 << 23;
 
    srcId(i, src1, 26);
 }
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 402cecfb5f8..3e44bf9fdba 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -1763,11 +1763,11 @@ Converter::handleTEX(Value *dst[4], int R, int S, int L, int C, int Dx, int Dy)
    if (tgsi.getOpcode() == TGSI_OPCODE_TG4 && !tgt.isShadow())
       texi->tex.gatherComp = tgsi.getSrc(1).getValueU32(0, info);
 
+   texi->tex.useOffsets = tgsi.getNumTexOffsets();
    for (s = 0; s < tgsi.getNumTexOffsets(); ++s) {
       for (c = 0; c < 3; ++c) {
-         texi->tex.offset[s][c] = tgsi.getTexOffset(s).getValueU32(c, info);
-         if (texi->tex.offset[s][c])
-            texi->tex.useOffsets = s + 1;
+         texi->offset[s][c].set(fetchSrc(tgsi.getTexOffset(s), c, NULL));
+         texi->offset[s][c].setInsn(texi);
       }
    }
 
@@ -1800,11 +1800,11 @@ Converter::handleTXF(Value *dst[4], int R, int L_M)
 
    setTexRS(texi, c, R, -1);
 
+   texi->tex.useOffsets = tgsi.getNumTexOffsets();
    for (s = 0; s < tgsi.getNumTexOffsets(); ++s) {
       for (c = 0; c < 3; ++c) {
-         texi->tex.offset[s][c] = tgsi.getTexOffset(s).getValueU32(c, info);
-         if (texi->tex.offset[s][c])
-            texi->tex.useOffsets = s + 1;
+         texi->offset[s][c].set(fetchSrc(tgsi.getTexOffset(s), c, NULL));
+         texi->offset[s][c].setInsn(texi);
       }
    }
 
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
index 69e88e6902d..eafc0a73bc6 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
@@ -695,6 +695,14 @@ NV50LoweringPreSSA::handleTEX(TexInstruction *i)
    // texel offsets are 3 immediate fields in the instruction,
    // nv50 cannot do textureGatherOffsets
    assert(i->tex.useOffsets <= 1);
+   if (i->tex.useOffsets) {
+      for (int c = 0; c < 3; ++c) {
+         ImmediateValue val;
+         assert(i->offset[0][c].getImmediate(val));
+         i->tex.offset[c] = val.reg.data.u32;
+         i->offset[0][c].set(NULL);
+      }
+   }
 
    return true;
 }
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 c7e9063fe65..dc1eb5e80a5 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -739,21 +739,42 @@ NVC0LoweringPass::handleTEX(TexInstruction *i)
 
    // offset is last source (lod 1st, dc 2nd)
    if (i->tex.useOffsets) {
-      uint32_t value = 0;
       int n, c;
       int s = i->srcCount(0xff, true);
       if (i->srcExists(s)) // move potential predicate out of the way
          i->moveSources(s, 1);
+      if (i->tex.useOffsets == 4 && i->srcExists(s + 1))
+         i->moveSources(s + 1, 1);
       if (i->op == OP_TXG) {
-         assert(i->tex.useOffsets == 1);
-         for (c = 0; c < 3; ++c)
-            value |= (i->tex.offset[0][c] & 0xff) << (c * 8);
+         // Either there is 1 offset, which goes into the 2 low bytes of the
+         // first source, or there are 4 offsets, which go into 2 sources (8
+         // values, 1 byte each).
+         Value *offs[2] = {NULL, NULL};
+         for (n = 0; n < i->tex.useOffsets; n++) {
+            for (c = 0; c < 2; ++c) {
+               if ((n % 2) == 0 && c == 0)
+                  offs[n / 2] = i->offset[n][c].get();
+               else
+                  bld.mkOp3(OP_INSBF, TYPE_U32,
+                            offs[n / 2],
+                            i->offset[n][c].get(),
+                            bld.mkImm(0x800 | ((n * 16 + c * 8) % 32)),
+                            offs[n / 2]);
+            }
+         }
+         i->setSrc(s, offs[0]);
+         if (offs[1])
+            i->setSrc(s + 1, offs[1]);
       } else {
-         for (n = 0; n < i->tex.useOffsets; ++n)
-            for (c = 0; c < 3; ++c)
-               value |= (i->tex.offset[n][c] & 0xf) << (n * 12 + c * 4);
+         unsigned imm = 0;
+         assert(i->tex.useOffsets == 1);
+         for (c = 0; c < 3; ++c) {
+            ImmediateValue val;
+            assert(i->offset[0][c].getImmediate(val));
+            imm |= (val.reg.data.u32 & 0xf) << (c * 4);
+         }
+         i->setSrc(s, bld.loadImm(NULL, imm));
       }
-      i->setSrc(s, bld.loadImm(NULL, value));
    }
 
    if (chipset >= NVISA_GK104_CHIPSET) {