[llvm] r219953 - R600: Fix nonsensical implementation of computeKnownBits for BFE

[Matt Arsenault]

Author: arsenm
Date: Thu Oct 16 15:07:40 2014
New Revision: 219953

URL: http://llvm.org/viewvc/llvm-project?rev=219953&view=rev
Log:
R600: Fix nonsensical implementation of computeKnownBits for BFE

This was resulting in invalid simplifications of sdiv

Modified:
    llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp
    llvm/trunk/test/CodeGen/R600/llvm.AMDGPU.bfe.i32.ll

Modified: llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp?rev=219953&r1=219952&r2=219953&view=diff
==============================================================================
--- llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp (original)
+++ llvm/trunk/lib/Target/R600/AMDGPUISelLowering.cpp Thu Oct 16 15:07:40 2014
@@ -2377,11 +2377,7 @@ void AMDGPUTargetLowering::computeKnownB
     unsigned BitWidth = 32;
     uint32_t Width = CWidth->getZExtValue() & 0x1f;
 
-    // FIXME: This could do a lot more. If offset is 0, should be the same as
-    // sign_extend_inreg implementation, but that involves duplicating it.
-    if (Opc == AMDGPUISD::BFE_I32)
-      KnownOne = APInt::getHighBitsSet(BitWidth, BitWidth - Width);
-    else
+    if (Opc == AMDGPUISD::BFE_U32)
       KnownZero = APInt::getHighBitsSet(BitWidth, BitWidth - Width);
 
     break;

Modified: llvm/trunk/test/CodeGen/R600/llvm.AMDGPU.bfe.i32.ll
URL: http://llvm.org/viewvc/llvm-project/llvm/trunk/test/CodeGen/R600/llvm.AMDGPU.bfe.i32.ll?rev=219953&r1=219952&r2=219953&view=diff
==============================================================================
--- llvm/trunk/test/CodeGen/R600/llvm.AMDGPU.bfe.i32.ll (original)
+++ llvm/trunk/test/CodeGen/R600/llvm.AMDGPU.bfe.i32.ll Thu Oct 16 15:07:40 2014
@@ -424,3 +424,18 @@ define void @bfe_sext_in_reg_i24(i32 add
   store i32 %ashr, i32 addrspace(1)* %out, align 4
   ret void
 }
+
+; FUNC-LABEL: @simplify_demanded_bfe_sdiv
+; SI: BUFFER_LOAD_DWORD [[LOAD:v[0-9]+]]
+; SI: V_BFE_I32 [[BFE:v[0-9]+]], [[LOAD]], 1, 16
+; SI: V_LSHRREV_B32_e32 [[TMP0:v[0-9]+]], 31, [[BFE]]
+; SI: V_ADD_I32_e32 [[TMP1:v[0-9]+]], [[TMP0]], [[BFE]]
+; SI: V_ASHRREV_I32_e32 [[TMP2:v[0-9]+]], 1, [[TMP1]]
+; SI: BUFFER_STORE_DWORD [[TMP2]]
+define void @simplify_demanded_bfe_sdiv(i32 addrspace(1)* %out, i32 addrspace(1)* %in) nounwind {
+  %src = load i32 addrspace(1)* %in, align 4
+  %bfe = call i32 @llvm.AMDGPU.bfe.i32(i32 %src, i32 1, i32 16) nounwind readnone
+  %div = sdiv i32 %bfe, 2
+  store i32 %div, i32 addrspace(1)* %out, align 4
+  ret void
+}