RPCS3/llvm-mirror
1
0
Fork 0
mirror of https://github.com/RPCS3/llvm-mirror.git synced 2024-10-19 02:52:53 +02:00
Code Issues Projects Releases Wiki Activity

[AMDGPU] Change the l1 flush instruction for AMDPAL/MESA3D.

Browse Source 
This commit changes which l1 flush instruction is used for AMDPAL and
MESA3d workloads to flush the entire l1 cache instead of just the
volatile lines.

Differential Revision: https://reviews.llvm.org/D55367

llvm-svn: 348771
This commit is contained in:
Neil Henning 2018-12-10 16:35:53 +00:00
parent 007fe3928c
commit ad77e61e7c
3 changed files with 451 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

8
lib/Target/AMDGPU/SIMemoryLegalizer.cpp
View File

@ -812,6 +812,12 @@ bool SIGfx7CacheControl::insertCacheInvalidate(MachineBasicBlock::iterator &MI,
MachineBasicBlock &MBB = *MI->getParent();
DebugLoc DL = MI->getDebugLoc();
const GCNSubtarget &STM = MBB.getParent()->getSubtarget<GCNSubtarget>();
const unsigned Flush = STM.isAmdPalOS() || STM.isMesa3DOS()
? AMDGPU::BUFFER_WBINVL1
: AMDGPU::BUFFER_WBINVL1_VOL;
if (Pos == Position::AFTER)
++MI;
@ -819,7 +825,7 @@ bool SIGfx7CacheControl::insertCacheInvalidate(MachineBasicBlock::iterator &MI,
switch (Scope) {
case SIAtomicScope::SYSTEM:
case SIAtomicScope::AGENT:
BuildMI(MBB, MI, DL, TII->get(AMDGPU::BUFFER_WBINVL1_VOL));
BuildMI(MBB, MI, DL, TII->get(Flush));
Changed = true;
break;
case SIAtomicScope::WORKGROUP:

222
test/CodeGen/AMDGPU/memory-legalizer-amdpal.ll Normal file
View File

@ -0,0 +1,222 @@
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx800 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--amdpal -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; FUNC-LABEL: {{^}}system_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN-NEXT: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_acquire() {
entry:
fence acquire
ret void
}
; FUNC-LABEL: {{^}}system_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_release() {
entry:
fence release
ret void
}
; FUNC-LABEL: {{^}}system_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_acq_rel() {
entry:
fence acq_rel
ret void
}
; FUNC-LABEL: {{^}}system_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_seq_cst() {
entry:
fence seq_cst
ret void
}
; FUNC-LABEL: {{^}}singlethread_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_acquire() {
entry:
fence syncscope("singlethread") acquire
ret void
}
; FUNC-LABEL: {{^}}singlethread_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_release() {
entry:
fence syncscope("singlethread") release
ret void
}
; FUNC-LABEL: {{^}}singlethread_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_acq_rel() {
entry:
fence syncscope("singlethread") acq_rel
ret void
}
; FUNC-LABEL: {{^}}singlethread_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_seq_cst() {
entry:
fence syncscope("singlethread") seq_cst
ret void
}
; FUNC-LABEL: {{^}}agent_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN-NEXT: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_acquire() {
entry:
fence syncscope("agent") acquire
ret void
}
; FUNC-LABEL: {{^}}agent_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_release() {
entry:
fence syncscope("agent") release
ret void
}
; FUNC-LABEL: {{^}}agent_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_acq_rel() {
entry:
fence syncscope("agent") acq_rel
ret void
}
; FUNC-LABEL: {{^}}agent_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_seq_cst() {
entry:
fence syncscope("agent") seq_cst
ret void
}
; FUNC-LABEL: {{^}}workgroup_acquire:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_acquire() {
entry:
fence syncscope("workgroup") acquire
ret void
}
; FUNC-LABEL: {{^}}workgroup_release:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_release() {
entry:
fence syncscope("workgroup") release
ret void
}
; FUNC-LABEL: {{^}}workgroup_acq_rel:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_acq_rel() {
entry:
fence syncscope("workgroup") acq_rel
ret void
}
; FUNC-LABEL: {{^}}workgroup_seq_cst:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_seq_cst() {
entry:
fence syncscope("workgroup") seq_cst
ret void
}
; FUNC-LABEL: {{^}}wavefront_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_acquire() {
entry:
fence syncscope("wavefront") acquire
ret void
}
; FUNC-LABEL: {{^}}wavefront_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_release() {
entry:
fence syncscope("wavefront") release
ret void
}
; FUNC-LABEL: {{^}}wavefront_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_acq_rel() {
entry:
fence syncscope("wavefront") acq_rel
ret void
}
; FUNC-LABEL: {{^}}wavefront_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_seq_cst() {
entry:
fence syncscope("wavefront") seq_cst
ret void
}

222
test/CodeGen/AMDGPU/memory-legalizer-mesa3d.ll Normal file
View File

@ -0,0 +1,222 @@
; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx600 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx700 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx800 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; RUN: llc -mtriple=amdgcn--mesa3d -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=FUNC,GCN %s
; FUNC-LABEL: {{^}}system_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN-NEXT: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_acquire() {
entry:
fence acquire
ret void
}
; FUNC-LABEL: {{^}}system_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_release() {
entry:
fence release
ret void
}
; FUNC-LABEL: {{^}}system_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_acq_rel() {
entry:
fence acq_rel
ret void
}
; FUNC-LABEL: {{^}}system_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @system_seq_cst() {
entry:
fence seq_cst
ret void
}
; FUNC-LABEL: {{^}}singlethread_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_acquire() {
entry:
fence syncscope("singlethread") acquire
ret void
}
; FUNC-LABEL: {{^}}singlethread_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_release() {
entry:
fence syncscope("singlethread") release
ret void
}
; FUNC-LABEL: {{^}}singlethread_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_acq_rel() {
entry:
fence syncscope("singlethread") acq_rel
ret void
}
; FUNC-LABEL: {{^}}singlethread_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @singlethread_seq_cst() {
entry:
fence syncscope("singlethread") seq_cst
ret void
}
; FUNC-LABEL: {{^}}agent_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN-NEXT: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_acquire() {
entry:
fence syncscope("agent") acquire
ret void
}
; FUNC-LABEL: {{^}}agent_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_release() {
entry:
fence syncscope("agent") release
ret void
}
; FUNC-LABEL: {{^}}agent_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_acq_rel() {
entry:
fence syncscope("agent") acq_rel
ret void
}
; FUNC-LABEL: {{^}}agent_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_waitcnt vmcnt(0){{$}}
; GCN: buffer_wbinvl1{{$}}
; GCN: s_endpgm
define amdgpu_kernel void @agent_seq_cst() {
entry:
fence syncscope("agent") seq_cst
ret void
}
; FUNC-LABEL: {{^}}workgroup_acquire:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_acquire() {
entry:
fence syncscope("workgroup") acquire
ret void
}
; FUNC-LABEL: {{^}}workgroup_release:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_release() {
entry:
fence syncscope("workgroup") release
ret void
}
; FUNC-LABEL: {{^}}workgroup_acq_rel:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_acq_rel() {
entry:
fence syncscope("workgroup") acq_rel
ret void
}
; FUNC-LABEL: {{^}}workgroup_seq_cst:
; GCN: %bb.0
; GCN-NOT: s_waitcnt vmcnt(0){{$}}
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @workgroup_seq_cst() {
entry:
fence syncscope("workgroup") seq_cst
ret void
}
; FUNC-LABEL: {{^}}wavefront_acquire:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_acquire() {
entry:
fence syncscope("wavefront") acquire
ret void
}
; FUNC-LABEL: {{^}}wavefront_release:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_release() {
entry:
fence syncscope("wavefront") release
ret void
}
; FUNC-LABEL: {{^}}wavefront_acq_rel:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_acq_rel() {
entry:
fence syncscope("wavefront") acq_rel
ret void
}
; FUNC-LABEL: {{^}}wavefront_seq_cst:
; GCN: %bb.0
; GCN-NOT: ATOMIC_FENCE
; GCN: s_endpgm
define amdgpu_kernel void @wavefront_seq_cst() {
entry:
fence syncscope("wavefront") seq_cst
ret void
}