[llvm] 62abc8c - [AMDGPU] Set GFX11 null export target based on export attributes

[Jay Foad via llvm-commits]

Author: Carl Ritson
Date: 2022-06-21T09:40:31+01:00
New Revision: 62abc8c2005e3d086c44db041c6430e42bf3e4f3

URL: https://github.com/llvm/llvm-project/commit/62abc8c2005e3d086c44db041c6430e42bf3e4f3
DIFF: https://github.com/llvm/llvm-project/commit/62abc8c2005e3d086c44db041c6430e42bf3e4f3.diff

LOG: [AMDGPU] Set GFX11 null export target based on export attributes

If shader only has depth exports use MRTZ otherwise use MRT0.

Differential Revision: https://reviews.llvm.org/D128185

Added: 
    

Modified: 
    llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp
    llvm/test/CodeGen/AMDGPU/early-term.mir

Removed: 
    


################################################################################
diff  --git a/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp b/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp
index 4fa8ec7111343..47095ae22027c 100644
--- a/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SILateBranchLowering.cpp
@@ -72,16 +72,22 @@ static void generateEndPgm(MachineBasicBlock &MBB,
   bool IsPS = F.getCallingConv() == CallingConv::AMDGPU_PS;
 
   // Check if hardware has been configured to expect color or depth exports.
-  bool HasExports =
-      AMDGPU::getHasColorExport(F) || AMDGPU::getHasDepthExport(F);
+  bool HasColorExports = AMDGPU::getHasColorExport(F);
+  bool HasDepthExports = AMDGPU::getHasDepthExport(F);
+  bool HasExports = HasColorExports || HasDepthExports;
 
   // Prior to GFX10, hardware always expects at least one export for PS.
   bool MustExport = !AMDGPU::isGFX10Plus(TII->getSubtarget());
 
   if (IsPS && (HasExports || MustExport)) {
     // Generate "null export" if hardware is expecting PS to export.
+    const GCNSubtarget &ST = MBB.getParent()->getSubtarget<GCNSubtarget>();
+    int Target =
+        ST.hasNullExportTarget()
+            ? AMDGPU::Exp::ET_NULL
+            : (HasColorExports ? AMDGPU::Exp::ET_MRT0 : AMDGPU::Exp::ET_MRTZ);
     BuildMI(MBB, I, DL, TII->get(AMDGPU::EXP_DONE))
-        .addImm(AMDGPU::Exp::ET_NULL)
+        .addImm(Target)
         .addReg(AMDGPU::VGPR0, RegState::Undef)
         .addReg(AMDGPU::VGPR0, RegState::Undef)
         .addReg(AMDGPU::VGPR0, RegState::Undef)

diff  --git a/llvm/test/CodeGen/AMDGPU/early-term.mir b/llvm/test/CodeGen/AMDGPU/early-term.mir
index a8a245b45af6d..60dd1289f1066 100644
--- a/llvm/test/CodeGen/AMDGPU/early-term.mir
+++ b/llvm/test/CodeGen/AMDGPU/early-term.mir
@@ -26,7 +26,12 @@
     ret void
   }
 
+  define amdgpu_ps void @early_term_depth_only() #1 {
+    ret void
+  }
+
   attributes #0 = { "amdgpu-color-export"="0" "amdgpu-depth-export"="0" }
+  attributes #1 = { "amdgpu-color-export"="0" "amdgpu-depth-export"="1" }
 ...
 
 ---
@@ -255,3 +260,41 @@ body: |
     EXP_DONE 0, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
     S_ENDPGM 0
 ...
+
+---
+name: early_term_depth_only
+tracksRegLiveness: true
+liveins:
+  - { reg: '$sgpr0' }
+  - { reg: '$sgpr1' }
+body: |
+  ; GCN-LABEL: name: early_term_depth_only
+  ; GCN: bb.0:
+  ; GCN:   successors: %bb.1(0x80000000), %bb.2(0x00000000)
+  ; GCN:   liveins: $sgpr0, $sgpr1
+  ; GCN:   $vgpr0 = V_MOV_B32_e32 0, implicit $exec
+  ; GCN:   dead $sgpr0 = S_AND_B32 $sgpr0, killed $sgpr1, implicit-def $scc
+  ; GCN:   S_CBRANCH_SCC0 %bb.2, implicit $scc
+  ; GCN: bb.1:
+  ; GCN:   liveins: $vgpr0
+  ; GCN:   EXP_DONE 8, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+  ; GCN:   S_ENDPGM 0
+  ; GCN: bb.2:
+  ; GCN:   $exec = S_MOV_B64 0
+  ; GFX9:  EXP_DONE 9, undef $vgpr0, undef $vgpr0, undef $vgpr0, undef $vgpr0, 1, 0, 0, implicit $exec
+  ; GFX10: EXP_DONE 9, undef $vgpr0, undef $vgpr0, undef $vgpr0, undef $vgpr0, 1, 0, 0, implicit $exec
+  ; GFX11: EXP_DONE 8, undef $vgpr0, undef $vgpr0, undef $vgpr0, undef $vgpr0, 1, 0, 0, implicit $exec
+  ; GCN:   S_ENDPGM 0
+  bb.0:
+    liveins: $sgpr0, $sgpr1
+    successors: %bb.1
+
+    $vgpr0 = V_MOV_B32_e32 0, implicit $exec
+    dead $sgpr0 = S_AND_B32 $sgpr0, killed $sgpr1, implicit-def $scc
+    SI_EARLY_TERMINATE_SCC0 implicit $scc, implicit $exec
+
+  bb.1:
+    liveins: $vgpr0
+    EXP_DONE 8, $vgpr0, $vgpr0, $vgpr0, $vgpr0, -1, -1, 15, implicit $exec
+    S_ENDPGM 0
+...