[AMDGPU] Make the iterative schedulers selectable via amdgpu-sched-strategy (llvm#135042)

Currently, the only way for users to try these schedulers is via
`-misched=` . However, this overrides the default scheduler for all
targets. This causes problems for various toolchains / drivers which
spawn jobs for both x86 and AMDGPU -- e.g. hipcc. On the other hand,
`amdgpu-sched-strategy` only changes the scheduler for AMDGPU target.

Author: jrbyrnes

llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp

@@ -424,10 +424,10 @@ static cl::opt<bool>
                        cl::desc("Enable loop data prefetch on AMDGPU"),
                        cl::Hidden, cl::init(false));
 
-static cl::opt<bool> EnableMaxIlpSchedStrategy(
-    "amdgpu-enable-max-ilp-scheduling-strategy",
-    cl::desc("Enable scheduling strategy to maximize ILP for a single wave."),
-    cl::Hidden, cl::init(false));
+static cl::opt<std::string>
+    AMDGPUSchedStrategy("amdgpu-sched-strategy",
+                        cl::desc("Select custom AMDGPU scheduling strategy."),
+                        cl::Hidden, cl::init(""));
 
 static cl::opt<bool> EnableRewritePartialRegUses(
     "amdgpu-enable-rewrite-partial-reg-uses",
@@ -1288,9 +1288,24 @@ ScheduleDAGInstrs *GCNPassConfig::createMachineScheduler(
   if (ST.enableSIScheduler())
     return createSIMachineScheduler(C);
 
-  if (EnableMaxIlpSchedStrategy)
+  Attribute SchedStrategyAttr =
+      C->MF->getFunction().getFnAttribute("amdgpu-sched-strategy");
+  StringRef SchedStrategy = SchedStrategyAttr.isValid()
+                                ? SchedStrategyAttr.getValueAsString()
+                                : AMDGPUSchedStrategy;
+
+  if (SchedStrategy == "max-ilp")
     return createGCNMaxILPMachineScheduler(C);
 
+  if (SchedStrategy == "iterative-ilp")
+    return createIterativeILPMachineScheduler(C);
+
+  if (SchedStrategy == "iterative-minreg")
+    return createMinRegScheduler(C);
+
+  if (SchedStrategy == "iterative-maxocc")
+    return createIterativeGCNMaxOccupancyMachineScheduler(C);
+
   return createGCNMaxOccupancyMachineScheduler(C);
 }
 

llvm/test/CodeGen/AMDGPU/schedule-ilp-liveness-tracking.mir

@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
-# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -amdgpu-enable-max-ilp-scheduling-strategy -verify-machineinstrs -run-pass=machine-scheduler -verify-misched -o - %s | FileCheck %s
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx90a -amdgpu-sched-strategy=max-ilp -verify-machineinstrs -run-pass=machine-scheduler -verify-misched -o - %s | FileCheck %s
 
 ---
 name:            max-ilp-liveness-tracking

llvm/test/CodeGen/AMDGPU/schedule-ilp.ll

@@ -1,6 +1,6 @@
 ; RUN: llc -mtriple=amdgcn -mcpu=tonga -misched=gcn-iterative-ilp -verify-machineinstrs < %s | FileCheck %s
 ; RUN: llc -mtriple=amdgcn -mcpu=tonga -misched=gcn-max-ilp -verify-machineinstrs < %s | FileCheck %s
-; RUN: llc -mtriple=amdgcn -mcpu=tonga -amdgpu-enable-max-ilp-scheduling-strategy -verify-machineinstrs < %s | FileCheck %s
+; RUN: llc -mtriple=amdgcn -mcpu=tonga -amdgpu-sched-strategy=max-ilp -verify-machineinstrs < %s | FileCheck %s
 
 ; CHECK: NumVgprs: {{[0-9][0-9][0-9]$}}
 

llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit2.ll

@@ -1,7 +1,11 @@
 ; RUN: llc -mtriple=amdgcn -mcpu=tahiti -enable-amdgpu-aa=0 -misched=gcn-iterative-minreg -verify-machineinstrs < %s | FileCheck --check-prefix=SI-MINREG %s
 ; RUN: llc -mtriple=amdgcn -mcpu=tahiti -enable-amdgpu-aa=0 -misched=gcn-iterative-max-occupancy-experimental -verify-machineinstrs < %s | FileCheck --check-prefix=SI-MAXOCC %s
+; RUN: llc -mtriple=amdgcn -mcpu=tahiti -enable-amdgpu-aa=0 -amdgpu-sched-strategy=iterative-minreg -verify-machineinstrs < %s | FileCheck --check-prefix=SI-MINREG %s
+; RUN: llc -mtriple=amdgcn -mcpu=tahiti -enable-amdgpu-aa=0 -amdgpu-sched-strategy=iterative-maxocc -verify-machineinstrs < %s | FileCheck --check-prefix=SI-MAXOCC %s
 ; RUN: llc -mtriple=amdgcn -mcpu=fiji -enable-amdgpu-aa=0 -misched=gcn-iterative-minreg -verify-machineinstrs < %s | FileCheck --check-prefix=VI %s
 ; RUN: llc -mtriple=amdgcn -mcpu=fiji -enable-amdgpu-aa=0 -misched=gcn-iterative-max-occupancy-experimental -verify-machineinstrs < %s | FileCheck --check-prefix=VI %s
+; RUN: llc -mtriple=amdgcn -mcpu=fiji -enable-amdgpu-aa=0 -amdgpu-sched-strategy=iterative-minreg -verify-machineinstrs < %s | FileCheck --check-prefix=VI %s
+; RUN: llc -mtriple=amdgcn -mcpu=fiji -enable-amdgpu-aa=0 -amdgpu-sched-strategy=iterative-maxocc -verify-machineinstrs < %s | FileCheck --check-prefix=VI %s
 
 ; SI-MINREG: NumSgprs: {{[1-9]$}}
 ; SI-MINREG: NumVgprs: {{[1-9]$}}

llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit3.ll

@@ -1,5 +1,6 @@
 ; RUN: llc -mtriple=amdgcn -mcpu=tonga -verify-machineinstrs < %s | FileCheck --check-prefix=MISCHED %s
 ; RUN: llc -mtriple=amdgcn -mcpu=tonga -misched=gcn-iterative-ilp -verify-machineinstrs < %s | FileCheck --check-prefix=GCN-ILP %s
+; RUN: llc -mtriple=amdgcn -mcpu=tonga -amdgpu-sched-strategy=iterative-ilp -verify-machineinstrs < %s | FileCheck --check-prefix=GCN-ILP %s
 
 ; Test the scheduler when only one wave is requested. The result should be high register usage and max ILP.