summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorIlia Mirkin <[email protected]>2016-09-08 02:13:56 -0400
committerIlia Mirkin <[email protected]>2016-09-10 13:40:04 -0400
commita8c0c7301cccd6bba1842ae8aa901cd45d9d5ffd (patch)
tree307042aa435ada8d7c3c7af53b452cf7b02844e6
parenta22aee5ad1154918aff77af5102fecd2aa39a631 (diff)
gm107/ir: allow indirect inputs to be loaded by frag shader
Looks like the GM107 IPA op does not allow a separate offset when using an indirect register. Instead we must use AL2P like we do for indirect vertex operations on Kepler+. Signed-off-by: Ilia Mirkin <[email protected]> Reviewed-by: Samuel Pitoiset <[email protected]>
-rw-r--r--src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp25
-rw-r--r--src/gallium/drivers/nouveau/nvc0/nvc0_screen.c1
2 files changed, 21 insertions, 5 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 2604296d46e..3c3d61112b9 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_lowering_nvc0.cpp
@@ -2695,13 +2695,30 @@ NVC0LoweringPass::visit(Instruction *i)
/* Kepler+ has a special opcode to compute a new base address to be used
* for indirect loads.
+ *
+ * Maxwell+ has an additional similar requirement for indirect
+ * interpolation ops in frag shaders.
*/
- if (targ->getChipset() >= NVISA_GK104_CHIPSET && !i->perPatch &&
- (i->op == OP_VFETCH || i->op == OP_EXPORT) && i->src(0).isIndirect(0)) {
+ bool doAfetch = false;
+ if (targ->getChipset() >= NVISA_GK104_CHIPSET &&
+ !i->perPatch &&
+ (i->op == OP_VFETCH || i->op == OP_EXPORT) &&
+ i->src(0).isIndirect(0)) {
+ doAfetch = true;
+ }
+ if (targ->getChipset() >= NVISA_GM107_CHIPSET &&
+ (i->op == OP_LINTERP || i->op == OP_PINTERP) &&
+ i->src(0).isIndirect(0)) {
+ doAfetch = true;
+ }
+
+ if (doAfetch) {
+ Value *addr = cloneShallow(func, i->getSrc(0));
Instruction *afetch = bld.mkOp1(OP_AFETCH, TYPE_U32, bld.getSSA(),
- cloneShallow(func, i->getSrc(0)));
+ i->getSrc(0));
afetch->setIndirect(0, 0, i->getIndirect(0, 0));
- i->src(0).get()->reg.data.offset = 0;
+ addr->reg.data.offset = 0;
+ i->setSrc(0, addr);
i->setIndirect(0, 0, afetch->getDef(0));
}
diff --git a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
index 0103031da66..1757cbb93a5 100644
--- a/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
+++ b/src/gallium/drivers/nouveau/nvc0/nvc0_screen.c
@@ -349,7 +349,6 @@ nvc0_screen_get_shader_param(struct pipe_screen *pscreen, unsigned shader,
case PIPE_SHADER_CAP_INDIRECT_OUTPUT_ADDR:
return shader != PIPE_SHADER_FRAGMENT;
case PIPE_SHADER_CAP_INDIRECT_INPUT_ADDR:
- return shader != PIPE_SHADER_FRAGMENT || class_3d < GM107_3D_CLASS;
case PIPE_SHADER_CAP_INDIRECT_TEMP_ADDR:
case PIPE_SHADER_CAP_INDIRECT_CONST_ADDR:
return 1;