[Mesa-dev] [PATCH] nv50/ir: check if the target supports the new offset before inlining

[Ilia Mirkin]

Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=93300
Signed-off-by: Ilia Mirkin <imirkin at alum.mit.edu>
---
 src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp  |  9 ++++++---
 src/gallium/drivers/nouveau/codegen/nv50_ir_target.h      |  2 ++
 .../drivers/nouveau/codegen/nv50_ir_target_nv50.cpp       | 15 +++++++++++++++
 src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.h |  2 ++
 4 files changed, 25 insertions(+), 3 deletions(-)

diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
index 1384a25..7b7352e 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_peephole.cpp
@@ -290,20 +290,23 @@ IndirectPropagation::visit(BasicBlock *bb)
             continue;
          if (insn->op == OP_ADD && !isFloatType(insn->dType)) {
             if (insn->src(0).getFile() != targ->nativeFile(FILE_ADDRESS) ||
-                !insn->src(1).getImmediate(imm))
+                !insn->src(1).getImmediate(imm) ||
+                !targ->insnCanLoadOffset(i, s, imm.reg.data.s32))
                continue;
             i->setIndirect(s, 0, insn->getSrc(0));
             i->setSrc(s, cloneShallow(func, i->getSrc(s)));
             i->src(s).get()->reg.data.offset += imm.reg.data.u32;
          } else if (insn->op == OP_SUB && !isFloatType(insn->dType)) {
             if (insn->src(0).getFile() != targ->nativeFile(FILE_ADDRESS) ||
-                !insn->src(1).getImmediate(imm))
+                !insn->src(1).getImmediate(imm) ||
+                !targ->insnCanLoadOffset(i, s, -imm.reg.data.s32))
                continue;
             i->setIndirect(s, 0, insn->getSrc(0));
             i->setSrc(s, cloneShallow(func, i->getSrc(s)));
             i->src(s).get()->reg.data.offset -= imm.reg.data.u32;
          } else if (insn->op == OP_MOV) {
-            if (!insn->src(0).getImmediate(imm))
+            if (!insn->src(0).getImmediate(imm) ||
+                !targ->insnCanLoadOffset(i, s, imm.reg.data.s32))
                continue;
             i->setIndirect(s, 0, NULL);
             i->setSrc(s, cloneShallow(func, i->getSrc(s)));
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h
index 4e33997..673f881 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target.h
@@ -191,6 +191,8 @@ public:
 
    virtual bool insnCanLoad(const Instruction *insn, int s,
                             const Instruction *ld) const = 0;
+   virtual bool insnCanLoadOffset(const Instruction *insn, int s,
+                                  int offset) const { return true; }
    virtual bool isOpSupported(operation, DataType) const = 0;
    virtual bool isAccessSupported(DataFile, DataType) const = 0;
    virtual bool isModSupported(const Instruction *,
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp
index 2d5baed..101082e 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.cpp
@@ -385,6 +385,21 @@ TargetNV50::insnCanLoad(const Instruction *i, int s,
 }
 
 bool
+TargetNV50::insnCanLoadOffset(const Instruction *i, int s, int offset) const
+{
+   if (!i->src(s).isIndirect(0))
+      return true;
+   offset += i->src(s).get()->reg.data.offset;
+   if (i->op == OP_LOAD || i->op == OP_STORE) {
+      // There are some restrictions in theory, but in practice they're never
+      // going to be hit. When we enable shared/global memory, this will
+      // become more important.
+      return true;
+   }
+   return offset >= 0 && offset <= (int32_t)(127 * i->src(s).get()->reg.size);
+}
+
+bool
 TargetNV50::isAccessSupported(DataFile file, DataType ty) const
 {
    if (ty == TYPE_B96 || ty == TYPE_NONE)
diff --git a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.h b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.h
index 0cbf180..00243d7 100644
--- a/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.h
+++ b/src/gallium/drivers/nouveau/codegen/nv50_ir_target_nv50.h
@@ -46,6 +46,8 @@ public:
 
    virtual bool insnCanLoad(const Instruction *insn, int s,
                             const Instruction *ld) const;
+   virtual bool insnCanLoadOffset(const Instruction *insn, int s,
+                                  int offset) const;
    virtual bool isOpSupported(operation, DataType) const;
    virtual bool isAccessSupported(DataFile, DataType) const;
    virtual bool isModSupported(const Instruction *, int s, Modifier) const;
-- 
2.4.10