[AMDGPU] Fix CS scratch setup on pre-GCN3 ASICs
Summary:
Prior to GCN3 s_load_dword offsets are in dwords rather than bytes.
Thus the scratch buffer descriptor offset must be adjusted for pre-GCN3 ASICs.
Reviewers: nhaehnle, tpr
Reviewed By: nhaehnle
Subscribers: sheredom, arsenm, kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits
Differential Revision: https://reviews.llvm.org/D56496
git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@353530 91177308-0d34-0410-b5e6-96231b3b80d8
diff --git a/lib/Target/AMDGPU/SIFrameLowering.cpp b/lib/Target/AMDGPU/SIFrameLowering.cpp
index 175e2bd..5368847 100644
--- a/lib/Target/AMDGPU/SIFrameLowering.cpp
+++ b/lib/Target/AMDGPU/SIFrameLowering.cpp
@@ -422,9 +422,11 @@
MachineMemOperand::MODereferenceable,
16, 4);
unsigned Offset = Fn.getCallingConv() == CallingConv::AMDGPU_CS ? 16 : 0;
+ const GCNSubtarget &Subtarget = MF.getSubtarget<GCNSubtarget>();
+ unsigned EncodedOffset = AMDGPU::getSMRDEncodedOffset(Subtarget, Offset);
BuildMI(MBB, I, DL, LoadDwordX4, ScratchRsrcReg)
.addReg(Rsrc01)
- .addImm(Offset) // offset
+ .addImm(EncodedOffset) // offset
.addImm(0) // glc
.addReg(ScratchRsrcReg, RegState::ImplicitDefine)
.addMemOperand(MMO);
diff --git a/test/CodeGen/AMDGPU/amdpal.ll b/test/CodeGen/AMDGPU/amdpal.ll
index 8d4c09c..c6d082d 100644
--- a/test/CodeGen/AMDGPU/amdpal.ll
+++ b/test/CodeGen/AMDGPU/amdpal.ll
@@ -1,4 +1,5 @@
-; RUN: llc < %s -mtriple=amdgcn--amdpal -mcpu=tahiti | FileCheck --check-prefix=PAL --enable-var-scope %s
+; RUN: llc < %s -mtriple=amdgcn--amdpal -mcpu=tahiti | FileCheck --check-prefixes=PAL,CI --enable-var-scope %s
+; RUN: llc < %s -mtriple=amdgcn--amdpal -mcpu=tonga | FileCheck --check-prefixes=PAL,VI --enable-var-scope %s
; PAL-NOT: .AMDGPU.config
; PAL-LABEL: {{^}}simple:
@@ -55,11 +56,13 @@
; Check code sequence for amdpal use of scratch for alloca in a compute shader.
; The scratch descriptor is loaded from offset 0x10 of the GIT, rather than offset
; 0 in a graphics shader.
+; Prior to GCN3 s_load_dword offsets are dwords, so the offset will be 0x4.
; PAL-LABEL: {{^}}scratch2_cs:
; PAL: s_movk_i32 s{{[0-9]+}}, 0x1234
; PAL: s_mov_b32 s[[GITPTR:[0-9]+]], s0
-; PAL: s_load_dwordx4 s{{\[}}[[SCRATCHDESC:[0-9]+]]:{{[0-9]+]}}, s{{\[}}[[GITPTR]]:{{[0-9]+\]}}, 0x10
+; CI: s_load_dwordx4 s{{\[}}[[SCRATCHDESC:[0-9]+]]:{{[0-9]+]}}, s{{\[}}[[GITPTR]]:{{[0-9]+\]}}, 0x4
+; VI: s_load_dwordx4 s{{\[}}[[SCRATCHDESC:[0-9]+]]:{{[0-9]+]}}, s{{\[}}[[GITPTR]]:{{[0-9]+\]}}, 0x10
; PAL: buffer_store{{.*}}, s{{\[}}[[SCRATCHDESC]]:
define amdgpu_cs void @scratch2_cs(i32 inreg, i32 inreg, i32 inreg, <3 x i32> inreg, i32 inreg, <3 x i32> %coord, <2 x i32> %in, i32 %extra, i32 %idx) #0 {