[llvm] [AMDGPU] New AMDGPUInsertSingleUseVDST pass (PR #72388)

[Jay Foad via llvm-commits]

================
@@ -0,0 +1,1023 @@
+# NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py UTC_ARGS: --version 2
+# RUN: llc -march=amdgcn -mcpu=gfx1150 -mattr=-wavefrontsize32,+wavefrontsize64 -verify-machineinstrs -run-pass=amdgpu-insert-single-use-vdst %s -o - | FileCheck %s
+
+---
+name: valu_dep_1
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0
+    ; CHECK-LABEL: name: valu_dep_1
+    ; CHECK: liveins: $vgpr0
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_2
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1
+    ; CHECK-LABEL: name: valu_dep_2
+    ; CHECK: liveins: $vgpr0, $vgpr1
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_3
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2
+    ; CHECK-LABEL: name: valu_dep_3
+    ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_4
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+    ; CHECK-LABEL: name: valu_dep_4
+    ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+    $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: valu_dep_5
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+    ; CHECK-LABEL: name: valu_dep_5
+    ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2, $vgpr3, $vgpr4
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr4 = V_ADD_U32_e32 $vgpr4, $vgpr4, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr1 = V_ADD_U32_e32 $vgpr1, $vgpr1, implicit $exec
+    $vgpr2 = V_ADD_U32_e32 $vgpr2, $vgpr2, implicit $exec
+    $vgpr3 = V_ADD_U32_e32 $vgpr3, $vgpr3, implicit $exec
+    $vgpr4 = V_ADD_U32_e32 $vgpr4, $vgpr4, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: multiple_uses_1
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2
+    ; CHECK-LABEL: name: multiple_uses_1
+    ; CHECK: liveins: $vgpr0, $vgpr1, $vgpr2
+    ; CHECK-NEXT: {{  $}}
+    ; CHECK-NEXT: $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    ; CHECK-NEXT: S_SINGLEUSE_VDST 1
+    ; CHECK-NEXT: $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr0 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr1 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+    $vgpr2 = V_ADD_U32_e32 $vgpr0, $vgpr0, implicit $exec
+...
+
+---
+name: multiple_uses_2
+tracksRegLiveness: true
+body: |
+  bb.0:
+    liveins: $vgpr0, $vgpr1, $vgpr2
----------------
jayfoad wrote:

Done

https://github.com/llvm/llvm-project/pull/72388