ir: add indirect support for images on Kepler

Samuel Pitoiset Tue, 26 Apr 2016 09:10:40 -0700


On 04/26/2016 03:22 AM, Ilia Mirkin wrote:

On Mon, Apr 25, 2016 at 4:15 PM, Samuel Pitoiset
<[email protected]> wrote:

This fixes arb_shader_image_load_store-indexing and
arb_shader_image_load_store-max-images.

Signed-off-by: Samuel Pitoiset <[email protected]>
Reviewed-by: Ilia Mirkin <[email protected]>
---
 .../nouveau/codegen/nv50_ir_lowering_nvc0.cpp      | 40 +++++++++++++++-------
 1 file changed, 28 insertions(+), 12 deletions(-)

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 1f26a03..d2bd970 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -1576,9 +1576,17 @@ NVC0LoweringPass::adjustCoordinatesMS(TexInstruction 
*tex)
    Value *s = tex->getSrc(arg - 1);

    Value *tx = bld.getSSA(), *ty = bld.getSSA(), *ts = bld.getSSA();
+   Value *ind = NULL;

-   Value *ms_x = loadSuInfo32(NULL, base + NVE4_SU_INFO_MS(0));
-   Value *ms_y = loadSuInfo32(NULL, base + NVE4_SU_INFO_MS(1));
+   if (tex->tex.rIndirectSrc >= 0) {
+      // FIXME: out of bounds
+      ind = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(),
+                       tex->getIndirectR(), bld.mkImm(6));


Probably not the only place, but I guess you're kind of assuming that
r == 0 here? I think in practice it will be, but nothing guarantees
it. Either fix it up, or add an assert?


Yeah, this was my guess, but adding an assert is more secure.


Also, why are you setting r to 0xff here?

No ideas, I thought that setting r to 0xff was needed but after a secondlook at the code, it's just useless. I have removed it.

+      tex->tex.r = 0xff;
+   }
+
+   Value *ms_x = loadSuInfo32(ind, base + NVE4_SU_INFO_MS(0));
+   Value *ms_y = loadSuInfo32(ind, base + NVE4_SU_INFO_MS(1));

    bld.mkOp2(OP_SHL, TYPE_U32, tx, x, ms_x);
    bld.mkOp2(OP_SHL, TYPE_U32, ty, y, ms_y);
@@ -1617,6 +1625,7 @@ NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction 
*su)
    Value *src[3];
    Value *bf, *eau, *off;
    Value *addr, *pred;
+   Value *ind = NULL;

    off = bld.getScratch(4);
    bf = bld.getScratch(4);
@@ -1627,6 +1636,13 @@ 
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)

    adjustCoordinatesMS(su);

+   if (su->tex.rIndirectSrc >= 0) {
+      // FIXME: out of bounds
+      ind = bld.mkOp2v(OP_SHL, TYPE_U32, bld.getSSA(),
+                       su->getIndirectR(), bld.mkImm(6));
+      su->tex.r = 0xff;
+   }
+
    // calculate clamped coordinates
    for (c = 0; c < arg; ++c) {
       int dimc = c;
@@ -1638,9 +1654,9 @@ NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction 
*su)

       src[c] = bld.getScratch();
       if (c == 0 && raw)
-         v = loadSuInfo32(NULL, base + NVE4_SU_INFO_RAW_X);
+         v = loadSuInfo32(ind, base + NVE4_SU_INFO_RAW_X);
       else
-         v = loadSuInfo32(NULL, base + NVE4_SU_INFO_DIM(dimc));
+         v = loadSuInfo32(ind, base + NVE4_SU_INFO_DIM(dimc));
       bld.mkOp3(OP_SUCLAMP, TYPE_S32, src[c], su->getSrc(c), v, zero)
          ->subOp = getSuClampSubOp(su, dimc);
    }
@@ -1662,16 +1678,16 @@ 
NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction *su)
          bld.mkOp2(OP_AND, TYPE_U32, off, src[0], bld.loadImm(NULL, 0xffff));
    } else
    if (dim == 3) {
-      v = loadSuInfo32(NULL, base + NVE4_SU_INFO_UNK1C);
+      v = loadSuInfo32(ind, base + NVE4_SU_INFO_UNK1C);
       bld.mkOp3(OP_MADSP, TYPE_U32, off, src[2], v, src[1])
          ->subOp = NV50_IR_SUBOP_MADSP(4,2,8); // u16l u16l u16l

-      v = loadSuInfo32(NULL, base + NVE4_SU_INFO_PITCH);
+      v = loadSuInfo32(ind, base + NVE4_SU_INFO_PITCH);
       bld.mkOp3(OP_MADSP, TYPE_U32, off, off, v, src[0])
          ->subOp = NV50_IR_SUBOP_MADSP(0,2,8); // u32 u16l u16l
    } else {
       assert(dim == 2);
-      v = loadSuInfo32(NULL, base + NVE4_SU_INFO_PITCH);
+      v = loadSuInfo32(ind, base + NVE4_SU_INFO_PITCH);
       bld.mkOp3(OP_MADSP, TYPE_U32, off, src[1], v, src[0])
          ->subOp = (su->tex.target.isArray() || su->tex.target.isCube()) ?
          NV50_IR_SUBOP_MADSP_SD : NV50_IR_SUBOP_MADSP(4,2,8); // u16l u16l u16l
@@ -1682,7 +1698,7 @@ NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction 
*su)
       if (raw) {
          bf = src[0];
       } else {
-         v = loadSuInfo32(NULL, base + NVE4_SU_INFO_FMT);
+         v = loadSuInfo32(ind, base + NVE4_SU_INFO_FMT);
          bld.mkOp3(OP_VSHL, TYPE_U32, bf, src[0], v, zero)
             ->subOp = NV50_IR_SUBOP_V1(7,6,8|2);
       }
@@ -1699,7 +1715,7 @@ NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction 
*su)
       case 2:
          z = off;
          if (!su->tex.target.isArray() && !su->tex.target.isCube()) {
-            z = loadSuInfo32(NULL, base + NVE4_SU_INFO_UNK1C);
+            z = loadSuInfo32(ind, base + NVE4_SU_INFO_UNK1C);
             subOp = NV50_IR_SUBOP_SUBFM_3D;
          }
          break;
@@ -1714,7 +1730,7 @@ NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction 
*su)
    }

    // part 2
-   v = loadSuInfo32(NULL, base + NVE4_SU_INFO_ADDR);
+   v = loadSuInfo32(ind, base + NVE4_SU_INFO_ADDR);

    if (su->tex.target == TEX_TARGET_BUFFER) {
       eau = v;
@@ -1723,7 +1739,7 @@ NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction 
*su)
    }
    // add array layer offset
    if (su->tex.target.isArray() || su->tex.target.isCube()) {
-      v = loadSuInfo32(NULL, base + NVE4_SU_INFO_ARRAY);
+      v = loadSuInfo32(ind, base + NVE4_SU_INFO_ARRAY);
       if (dim == 1)
          bld.mkOp3(OP_MADSP, TYPE_U32, eau, src[1], v, eau)
             ->subOp = NV50_IR_SUBOP_MADSP(4,0,0); // u16 u24 u32
@@ -1763,7 +1779,7 @@ NVC0LoweringPass::processSurfaceCoordsNVE4(TexInstruction 
*su)

    // let's just set it 0 for raw access and hope it works
    v = raw ?
-      bld.mkImm(0) : loadSuInfo32(NULL, base + NVE4_SU_INFO_FMT);
+      bld.mkImm(0) : loadSuInfo32(ind, base + NVE4_SU_INFO_FMT);

    // get rid of old coordinate sources, make space for fmt info and predicate
    su->moveSources(arg, 3 - arg);
--
2.8.0

_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev


--
-Samuel
_______________________________________________
mesa-dev mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/mesa-dev

Re: [Mesa-dev] [PATCH v2 18/24] nvc0/ir: add indirect support for images on Kepler

Reply via email to