AMDGPU: Rework how private buffer passed for HSA

[oota-llvm.git] / lib / Target / AMDGPU / AMDGPUAsmPrinter.cpp

diff --git a/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp b/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
index 314ef721c1fcd395cb391f2a47205ced4e809aa2..b96c48f0561de2cbdc4871c795b92d7b42052372 100644 (file)
--- a/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
+++ b/lib/Target/AMDGPU/AMDGPUAsmPrinter.cpp
@@ -452,18 +452,27 @@ void AMDGPUAsmPrinter::getSIProgramInfo(SIProgramInfo &ProgInfo,
       S_00B848_FLOAT_MODE(ProgInfo.FloatMode) |
       S_00B848_PRIV(ProgInfo.Priv) |
       S_00B848_DX10_CLAMP(ProgInfo.DX10Clamp) |
-      S_00B848_IEEE_MODE(ProgInfo.DebugMode) |
+      S_00B848_DEBUG_MODE(ProgInfo.DebugMode) |
       S_00B848_IEEE_MODE(ProgInfo.IEEEMode);
 
+  // 0 = X, 1 = XY, 2 = XYZ
+  unsigned TIDIGCompCnt = 0;
+  if (MFI->hasWorkItemIDZ())
+    TIDIGCompCnt = 2;
+  else if (MFI->hasWorkItemIDY())
+    TIDIGCompCnt = 1;
+
   ProgInfo.ComputePGMRSrc2 =
       S_00B84C_SCRATCH_EN(ProgInfo.ScratchBlocks > 0) |
-      S_00B84C_USER_SGPR(MFI->NumUserSGPRs) |
-      S_00B84C_TGID_X_EN(1) |
-      S_00B84C_TGID_Y_EN(1) |
-      S_00B84C_TGID_Z_EN(1) |
-      S_00B84C_TG_SIZE_EN(1) |
-      S_00B84C_TIDIG_COMP_CNT(2) |
-      S_00B84C_LDS_SIZE(ProgInfo.LDSBlocks);
+      S_00B84C_USER_SGPR(MFI->getNumUserSGPRs()) |
+      S_00B84C_TGID_X_EN(MFI->hasWorkGroupIDX()) |
+      S_00B84C_TGID_Y_EN(MFI->hasWorkGroupIDY()) |
+      S_00B84C_TGID_Z_EN(MFI->hasWorkGroupIDZ()) |
+      S_00B84C_TG_SIZE_EN(MFI->hasWorkGroupInfo()) |
+      S_00B84C_TIDIG_COMP_CNT(TIDIGCompCnt) |
+      S_00B84C_EXCP_EN_MSB(0) |
+      S_00B84C_LDS_SIZE(ProgInfo.LDSBlocks) |
+      S_00B84C_EXCP_EN(0);
 }
 
 static unsigned getRsrcReg(unsigned ShaderType) {
@@ -524,9 +533,44 @@ void AMDGPUAsmPrinter::EmitAmdKernelCodeT(const MachineFunction &MF,
   header.compute_pgm_resource_registers =
       KernelInfo.ComputePGMRSrc1 |
       (KernelInfo.ComputePGMRSrc2 << 32);
-  header.code_properties =
-      AMD_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR |
-      AMD_CODE_PROPERTY_IS_PTR64;
+  header.code_properties = AMD_CODE_PROPERTY_IS_PTR64;
+
+  if (MFI->hasPrivateSegmentBuffer()) {
+    header.code_properties |=
+      AMD_CODE_PROPERTY_ENABLE_SGPR_PRIVATE_SEGMENT_BUFFER;
+  }
+
+  if (MFI->hasDispatchPtr())
+    header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_PTR;
+
+  if (MFI->hasQueuePtr())
+    header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_QUEUE_PTR;
+
+  if (MFI->hasKernargSegmentPtr())
+    header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_KERNARG_SEGMENT_PTR;
+
+  if (MFI->hasDispatchID())
+    header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_ID;
+
+  if (MFI->hasFlatScratchInit())
+    header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_FLAT_SCRATCH_INIT;
+
+  // TODO: Private segment size
+
+  if (MFI->hasGridWorkgroupCountX()) {
+    header.code_properties |=
+      AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_X;
+  }
+
+  if (MFI->hasGridWorkgroupCountY()) {
+    header.code_properties |=
+      AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_Y;
+  }
+
+  if (MFI->hasGridWorkgroupCountZ()) {
+    header.code_properties |=
+      AMD_CODE_PROPERTY_ENABLE_SGPR_GRID_WORKGROUP_COUNT_Z;
+  }
 
   if (MFI->hasDispatchPtr())
     header.code_properties |= AMD_CODE_PROPERTY_ENABLE_SGPR_DISPATCH_PTR;