Mesa (11.2): Revert "nv50/ir: normalize cube coordinates after derivatives have been computed"
Emil Velikov
evelikov at kemper.freedesktop.org
Mon Apr 4 11:52:52 UTC 2016
Module: Mesa
Branch: 11.2
Commit: 2c7214bc03b0a0b5ac4256d871dce5527b069726
URL: http://cgit.freedesktop.org/mesa/mesa/commit/?id=2c7214bc03b0a0b5ac4256d871dce5527b069726
Author: Emil Velikov <emil.l.velikov at gmail.com>
Date: Mon Apr 4 12:24:02 2016 +0100
Revert "nv50/ir: normalize cube coordinates after derivatives have been computed"
This reverts commit 6eeb284e4f74a2fe5ae6cba90f97f219935e24df.
Requested by Ilia over at mesa-stable
https://lists.freedesktop.org/archives/mesa-stable/2016-April/003939.html
---
.../drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp | 12 ++++++++
.../nouveau/codegen/nv50_ir_lowering_gm107.cpp | 18 +----------
.../nouveau/codegen/nv50_ir_lowering_nv50.cpp | 34 +--------------------
.../nouveau/codegen/nv50_ir_lowering_nvc0.cpp | 35 +---------------------
4 files changed, 15 insertions(+), 84 deletions(-)
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 539ad02..d06e9ef 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_from_tgsi.cpp
@@ -1989,6 +1989,7 @@ Converter::loadProjTexCoords(Value *dst[4], Value *src[4], unsigned int mask)
void
Converter::handleTEX(Value *dst[4], int R, int S, int L, int C, int Dx, int Dy)
{
+ Value *val;
Value *arg[4], *src[8];
Value *lod = NULL, *shd = NULL;
unsigned int s, c, d;
@@ -2031,6 +2032,17 @@ Converter::handleTEX(Value *dst[4], int R, int S, int L, int C, int Dx, int Dy)
shd = src[n - 1];
}
+ if (tgt.isCube()) {
+ for (c = 0; c < 3; ++c)
+ src[c] = mkOp1v(OP_ABS, TYPE_F32, getSSA(), arg[c]);
+ val = getScratch();
+ mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
+ mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
+ mkOp1(OP_RCP, TYPE_F32, val, val);
+ for (c = 0; c < 3; ++c)
+ src[c] = mkOp2v(OP_MUL, TYPE_F32, getSSA(), arg[c], val);
+ }
+
for (c = 0, d = 0; c < 4; ++c) {
if (dst[c]) {
texi->setDef(d++, dst[c]);
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_gm107.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_gm107.cpp
index a5deaef..0b90378 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_gm107.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_gm107.cpp
@@ -67,7 +67,6 @@ GM107LoweringPass::handleManualTXD(TexInstruction *i)
tmp = bld.getScratch();
for (l = 0; l < 4; ++l) {
- Value *src[3], *val;
// mov coordinates from lane l to all lanes
bld.mkOp(OP_QUADON, TYPE_NONE, NULL);
for (c = 0; c < dim; ++c) {
@@ -93,25 +92,10 @@ GM107LoweringPass::handleManualTXD(TexInstruction *i)
add->lanes = 1; /* abused for .ndv */
}
- // normalize cube coordinates if necessary
- if (i->tex.target.isCube()) {
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), crd[c]);
- val = bld.getScratch();
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
- bld.mkOp1(OP_RCP, TYPE_F32, val, val);
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(), crd[c], val);
- } else {
- for (c = 0; c < dim; ++c)
- src[c] = crd[c];
- }
-
// texture
bld.insert(tex = cloneForward(func, i));
for (c = 0; c < dim; ++c)
- tex->setSrc(c + array, src[c]);
+ tex->setSrc(c + array, crd[c]);
bld.mkOp(OP_QUADPOP, TYPE_NONE, NULL);
// save results
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 b7270df..f56528b 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nv50.cpp
@@ -724,23 +724,6 @@ NV50LoweringPreSSA::handleTEX(TexInstruction *i)
const int dref = arg;
const int lod = i->tex.target.isShadow() ? (arg + 1) : arg;
- /* Only normalize in the non-explicit derivatives case.
- */
- if (i->tex.target.isCube() && i->op != OP_TXD) {
- Value *src[3], *val;
- int c;
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), i->getSrc(c));
- val = bld.getScratch();
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
- bld.mkOp1(OP_RCP, TYPE_F32, val, val);
- for (c = 0; c < 3; ++c) {
- i->setSrc(c, bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(),
- i->getSrc(c), val));
- }
- }
-
// handle MS, which means looking up the MS params for this texture, and
// adjusting the input coordinates to point at the right sample.
if (i->tex.target.isMS()) {
@@ -958,7 +941,6 @@ NV50LoweringPreSSA::handleTXD(TexInstruction *i)
bld.mkOp(OP_QUADON, TYPE_NONE, NULL);
for (l = 0; l < 4; ++l) {
- Value *src[3], *val;
// mov coordinates from lane l to all lanes
for (c = 0; c < dim; ++c)
bld.mkQuadop(0x00, crd[c], l, i->getSrc(c), zero);
@@ -968,24 +950,10 @@ NV50LoweringPreSSA::handleTXD(TexInstruction *i)
// add dPdy from lane l to lanes dy
for (c = 0; c < dim; ++c)
bld.mkQuadop(qOps[l][1], crd[c], l, i->dPdy[c].get(), crd[c]);
- // normalize cube coordinates if necessary
- if (i->tex.target.isCube()) {
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), crd[c]);
- val = bld.getScratch();
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
- bld.mkOp1(OP_RCP, TYPE_F32, val, val);
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(), crd[c], val);
- } else {
- for (c = 0; c < dim; ++c)
- src[c] = crd[c];
- }
// texture
bld.insert(tex = cloneForward(func, i));
for (c = 0; c < dim; ++c)
- tex->setSrc(c, src[c]);
+ tex->setSrc(c, crd[c]);
// save results
for (c = 0; i->defExists(c); ++c) {
Instruction *mov;
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 50454cb..8da599e 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -615,24 +615,6 @@ NVC0LoweringPass::handleTEX(TexInstruction *i)
const int lyr = arg - (i->tex.target.isMS() ? 2 : 1);
const int chipset = prog->getTarget()->getChipset();
- /* Only normalize in the non-explicit derivatives case. For explicit
- * derivatives, this is handled in handleManualTXD.
- */
- if (i->tex.target.isCube() && i->dPdx[0].get() == NULL) {
- Value *src[3], *val;
- int c;
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), i->getSrc(c));
- val = bld.getScratch();
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
- bld.mkOp1(OP_RCP, TYPE_F32, val, val);
- for (c = 0; c < 3; ++c) {
- i->setSrc(c, bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(),
- i->getSrc(c), val));
- }
- }
-
// Arguments to the TEX instruction are a little insane. Even though the
// encoding is identical between SM20 and SM30, the arguments mean
// different things between Fermi and Kepler+. A lot of arguments are
@@ -893,7 +875,6 @@ NVC0LoweringPass::handleManualTXD(TexInstruction *i)
bld.mkOp(OP_QUADON, TYPE_NONE, NULL);
for (l = 0; l < 4; ++l) {
- Value *src[3], *val;
// mov coordinates from lane l to all lanes
for (c = 0; c < dim; ++c)
bld.mkQuadop(0x00, crd[c], l, i->getSrc(c + array), zero);
@@ -903,24 +884,10 @@ NVC0LoweringPass::handleManualTXD(TexInstruction *i)
// add dPdy from lane l to lanes dy
for (c = 0; c < dim; ++c)
bld.mkQuadop(qOps[l][1], crd[c], l, i->dPdy[c].get(), crd[c]);
- // normalize cube coordinates
- if (i->tex.target.isCube()) {
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp1v(OP_ABS, TYPE_F32, bld.getSSA(), crd[c]);
- val = bld.getScratch();
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[0], src[1]);
- bld.mkOp2(OP_MAX, TYPE_F32, val, src[2], val);
- bld.mkOp1(OP_RCP, TYPE_F32, val, val);
- for (c = 0; c < 3; ++c)
- src[c] = bld.mkOp2v(OP_MUL, TYPE_F32, bld.getSSA(), crd[c], val);
- } else {
- for (c = 0; c < dim; ++c)
- src[c] = crd[c];
- }
// texture
bld.insert(tex = cloneForward(func, i));
for (c = 0; c < dim; ++c)
- tex->setSrc(c + array, src[c]);
+ tex->setSrc(c + array, crd[c]);
// save results
for (c = 0; i->defExists(c); ++c) {
Instruction *mov;
More information about the mesa-commit
mailing list