From d5398b62da1913e7224c826da0dbd5fa88436f18 Mon Sep 17 00:00:00 2001
From: Rhys Perry <pendingchaos02@gmail.com>
Date: Thu, 27 Apr 2023 16:22:52 +0100
Subject: [PATCH] aco/ra: create M0-affinities for s_sendmsg
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

v2 by Timur Kristóf:
Do not add the affinity for instructions that can't write m0
reliably, such as readlane-like instructions on GFX8.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/22690>
---
 src/amd/compiler/aco_ir.cpp                  | 21 ++++++++++++++++++++
 src/amd/compiler/aco_ir.h                    |  1 +
 src/amd/compiler/aco_register_allocation.cpp |  8 ++++++++
 3 files changed, 30 insertions(+)

diff --git a/src/amd/compiler/aco_ir.cpp b/src/amd/compiler/aco_ir.cpp
index ef04a89c780..03f6931be1c 100644
--- a/src/amd/compiler/aco_ir.cpp
+++ b/src/amd/compiler/aco_ir.cpp
@@ -479,6 +479,27 @@ can_use_opsel(amd_gfx_level gfx_level, aco_opcode op, int idx)
    }
 }
 
+bool
+can_write_m0(amd_gfx_level gfx_level, const aco_ptr<Instruction>& instr)
+{
+   if (instr->isSALU())
+      return true;
+
+   if (instr->isVALU())
+      return gfx_level >= GFX9;
+
+   switch (instr->opcode) {
+   case aco_opcode::p_parallelcopy:
+   case aco_opcode::p_extract:
+   case aco_opcode::p_insert:
+      return true;
+   case aco_opcode::p_reload:
+      return gfx_level >= GFX9;
+   default:
+      return false;
+   }
+}
+
 bool
 instr_is_16bit(amd_gfx_level gfx_level, aco_opcode op)
 {
diff --git a/src/amd/compiler/aco_ir.h b/src/amd/compiler/aco_ir.h
index 6ae4571abd8..a39e949847f 100644
--- a/src/amd/compiler/aco_ir.h
+++ b/src/amd/compiler/aco_ir.h
@@ -1804,6 +1804,7 @@ bool instr_is_16bit(amd_gfx_level gfx_level, aco_opcode op);
 uint8_t get_gfx11_true16_mask(aco_opcode op);
 bool can_use_SDWA(amd_gfx_level gfx_level, const aco_ptr<Instruction>& instr, bool pre_ra);
 bool can_use_DPP(const aco_ptr<Instruction>& instr, bool pre_ra, bool dpp8);
+bool can_write_m0(amd_gfx_level gfx_level, const aco_ptr<Instruction>& instr);
 /* updates "instr" and returns the old instruction (or NULL if no update was needed) */
 aco_ptr<Instruction> convert_to_SDWA(amd_gfx_level gfx_level, aco_ptr<Instruction>& instr);
 aco_ptr<Instruction> convert_to_DPP(aco_ptr<Instruction>& instr, bool dpp8);
diff --git a/src/amd/compiler/aco_register_allocation.cpp b/src/amd/compiler/aco_register_allocation.cpp
index ca9b8702419..fba6e1c012c 100644
--- a/src/amd/compiler/aco_register_allocation.cpp
+++ b/src/amd/compiler/aco_register_allocation.cpp
@@ -55,6 +55,7 @@ struct assignment {
       struct {
          bool assigned : 1;
          bool vcc : 1;
+         bool m0 : 1;
       };
       uint8_t _ = 0;
    };
@@ -1653,6 +1654,11 @@ get_reg(ra_ctx& ctx, RegisterFile& reg_file, Temp temp,
       if (get_reg_specified(ctx, reg_file, temp.regClass(), instr, vcc))
          return vcc;
    }
+   if (ctx.assignments[temp.id()].m0) {
+      if (get_reg_specified(ctx, reg_file, temp.regClass(), instr, m0) &&
+          can_write_m0(ctx.program->gfx_level, instr))
+         return m0;
+   }
 
    std::optional<PhysReg> res;
 
@@ -2461,6 +2467,8 @@ get_affinities(ra_ctx& ctx, std::vector<IDSet>& live_out_per_block)
             if (!instr->definitions[1].isKill() && instr->operands[0].isTemp() &&
                 instr->operands[1].isFixed() && instr->operands[1].physReg() == exec)
                ctx.assignments[instr->operands[0].tempId()].vcc = true;
+         } else if (instr->opcode == aco_opcode::s_sendmsg) {
+            ctx.assignments[instr->operands[0].tempId()].m0 = true;
          }
 
          /* add operands to live variables */