[AMDGPU] Make the iterative schedulers selectable via amdgpu-sched-strategy

[jrbyrnes]

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.

[llvmbot]

@llvm/pr-subscribers-backend-amdgpu

Author: Jeffrey Byrnes (jrbyrnes)

Changes

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.

---

Full diff: https://github.com/llvm/llvm-project/pull/135042.diff

3 Files Affected:

• (modified) llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp (+9)
• (modified) llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit2.ll (+4)
• (modified) llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit3.ll (+1)

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 4b5c70f09155f..8d16b4f2bc818 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -1085,6 +1085,15 @@ GCNTargetMachine::createMachineScheduler(MachineSchedContext *C) const {
   if (SchedStrategy == "max-memory-clause")
     return createGCNMaxMemoryClauseMachineScheduler(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);
 }
 
diff --git a/llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit2.ll b/llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit2.ll
index 2c9d24ee04ebf..7d771342a598e 100644
--- a/llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit2.ll
+++ b/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]$}}
 ; SI-MINREG: NumVgprs: {{[1-9]$}}
diff --git a/llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit3.ll b/llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit3.ll
index 96b40bca5e2e3..ef24996d00274 100644
--- a/llvm/test/CodeGen/AMDGPU/schedule-regpressure-limit3.ll
+++ b/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.
 

[vpykhtin]

LGTM, Thanks!