AMDGPU/GlobalISel: Implement computeKnownBits for groupstaticsize

diff --git a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
index 867a722..ce669bb 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
+++ b/llvm/lib/Target/AMDGPU/AMDGPUSubtarget.h
@@ -78,7 +78,7 @@ protected:
   bool EnablePromoteAlloca;
   bool HasTrigReducedRange;
   unsigned MaxWavesPerEU;
-  int LocalMemorySize;
+  unsigned LocalMemorySize;
   char WavefrontSizeLog2;
 
 public:
@@ -202,7 +202,7 @@ public:
     return WavefrontSizeLog2;
   }
 
-  int getLocalMemorySize() const {
+  unsigned getLocalMemorySize() const {
     return LocalMemorySize;
   }
 

diff --git a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
index f64a378..b6f78ab 100644 (file)
--- a/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/SIISelLowering.cpp
@@ -11425,6 +11425,13 @@ void SITargetLowering::computeKnownBitsForTargetInstr(
       Known.Zero.setHighBits(Size - getSubtarget()->getWavefrontSizeLog2());
       break;
     }
+    case Intrinsic::amdgcn_groupstaticsize: {
+      // We can report everything over the maximum size as 0. We can't report
+      // based on the actual size because we don't know if it's accurate or not
+      // at any given point.
+      Known.Zero.setHighBits(countLeadingZeros(getSubtarget()->getLocalMemorySize()));
+      break;
+    }
     default:
       break;
     }

diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/postlegalizercombiner-and.mir b/llvm/test/CodeGen/AMDGPU/GlobalISel/postlegalizercombiner-and.mir
index 7bfca9a..b7491ab 100644 (file)
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/postlegalizercombiner-and.mir
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/postlegalizercombiner-and.mir
@@ -1,5 +1,5 @@
 # NOTE: Assertions have been autogenerated by utils/update_mir_test_checks.py
-# RUN: llc -mtriple=amdgcn-amd-amdhsa -run-pass=amdgpu-postlegalizer-combiner -verify-machineinstrs %s -o - | FileCheck %s
+# RUN: llc -mtriple=amdgcn-amd-amdhsa -mcpu=gfx900 -run-pass=amdgpu-postlegalizer-combiner -verify-machineinstrs %s -o - | FileCheck %s
 
 ---
 name:  remove_and_255_zextload
@@ -182,3 +182,86 @@ body:             |
     $vgpr0 = COPY %and
 
 ...
+
+# Test known bits for groupstaticsize is the maximum LDS size.
+---
+name:  remove_and_65535_groupstaticsize
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: remove_and_65535_groupstaticsize
+    ; CHECK: liveins: $vgpr0_vgpr1
+    ; CHECK: %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    ; CHECK: %mask:_(s32) = G_CONSTANT i32 65535
+    ; CHECK: %and:_(s32) = G_AND %lds_size, %mask
+    ; CHECK: $vgpr0 = COPY %and(s32)
+    %ptr:_(p1) = COPY $vgpr0_vgpr1
+    %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    %mask:_(s32) = G_CONSTANT i32 65535
+    %and:_(s32) = G_AND %lds_size, %mask
+    $vgpr0 = COPY %and
+
+...
+
+---
+name:  remove_and_131071_groupstaticsize
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: remove_and_131071_groupstaticsize
+    ; CHECK: liveins: $vgpr0_vgpr1
+    ; CHECK: %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    ; CHECK: $vgpr0 = COPY %lds_size(s32)
+    %ptr:_(p1) = COPY $vgpr0_vgpr1
+    %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    %mask:_(s32) = G_CONSTANT i32 131071
+    %and:_(s32) = G_AND %lds_size, %mask
+    $vgpr0 = COPY %and
+
+...
+
+---
+name:  no_remove_and_65536_groupstaticsize
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: no_remove_and_65536_groupstaticsize
+    ; CHECK: liveins: $vgpr0_vgpr1
+    ; CHECK: %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    ; CHECK: %mask:_(s32) = G_CONSTANT i32 65536
+    ; CHECK: %and:_(s32) = G_AND %lds_size, %mask
+    ; CHECK: $vgpr0 = COPY %and(s32)
+    %ptr:_(p1) = COPY $vgpr0_vgpr1
+    %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    %mask:_(s32) = G_CONSTANT i32 65536
+    %and:_(s32) = G_AND %lds_size, %mask
+    $vgpr0 = COPY %and
+
+...
+
+---
+name:  no_remove_and_32767_groupstaticsize
+legalized:       true
+tracksRegLiveness: true
+body:             |
+  bb.0:
+    liveins: $vgpr0_vgpr1
+    ; CHECK-LABEL: name: no_remove_and_32767_groupstaticsize
+    ; CHECK: liveins: $vgpr0_vgpr1
+    ; CHECK: %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    ; CHECK: %mask:_(s32) = G_CONSTANT i32 32767
+    ; CHECK: %and:_(s32) = G_AND %lds_size, %mask
+    ; CHECK: $vgpr0 = COPY %and(s32)
+    %ptr:_(p1) = COPY $vgpr0_vgpr1
+    %lds_size:_(s32) = G_INTRINSIC intrinsic(@llvm.amdgcn.groupstaticsize)
+    %mask:_(s32) = G_CONSTANT i32 32767
+    %and:_(s32) = G_AND %lds_size, %mask
+    $vgpr0 = COPY %and
+
+...