[AMDGPU][MC][DISASSEMBLER][GFX9] Corrected decoding of GLOBAL/SCRATCH…

… opcodes See bug 35433: https://bugs.llvm.org/show_bug.cgi?id=35433 Differential Revision: https://reviews.llvm.org/D40493 Reviewers: artem.tamazov, SamWot, arsenm git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@319050 91177308-0d34-0410-b5e6-96231b3b80d8
qzmfranklin · Nov 27, 2017 · 6f67bb8 · 6f67bb8
1 parent d89f477
commit 6f67bb8
Show file tree

Hide file tree

Showing 4 changed files with 91 additions and 6 deletions.
diff --git a/lib/Target/AMDGPU/FLATInstructions.td b/lib/Target/AMDGPU/FLATInstructions.td
@@ -305,7 +305,7 @@ multiclass FLAT_Global_Atomic_Pseudo<
   def _SADDR : FLAT_AtomicNoRet_Pseudo <opName,
     (outs),
     (ins VReg_64:$vaddr, data_rc:$vdata, SReg_64:$saddr, offset_s13:$offset, slc:$slc),
-    " $vaddr, $vdata$saddr$offset$slc">,
+    " $vaddr, $vdata, $saddr$offset$slc">,
     AtomicNoRet <opName#"_saddr", 0> {
     let has_saddr = 1;
     let enabled_saddr = 1;
@@ -315,7 +315,7 @@ multiclass FLAT_Global_Atomic_Pseudo<
   def _SADDR_RTN : FLAT_AtomicRet_Pseudo <opName,
     (outs vdst_rc:$vdst),
       (ins VReg_64:$vaddr, data_rc:$vdata, SReg_64:$saddr, offset_s13:$offset, slc:$slc),
-    " $vdst, $vaddr, $vdata$saddr$offset glc$slc">,
+    " $vdst, $vaddr, $vdata, $saddr$offset glc$slc">,
     AtomicNoRet <opName#"_saddr", 1> {
      let has_saddr = 1;
      let enabled_saddr = 1;

diff --git a/lib/Target/AMDGPU/InstPrinter/AMDGPUInstPrinter.cpp b/lib/Target/AMDGPU/InstPrinter/AMDGPUInstPrinter.cpp
@@ -72,9 +72,9 @@ void AMDGPUInstPrinter::printU16ImmDecOperand(const MCInst *MI, unsigned OpNo,
   O << formatDec(MI->getOperand(OpNo).getImm() & 0xffff);
 }
 
-void AMDGPUInstPrinter::printS16ImmDecOperand(const MCInst *MI, unsigned OpNo,
+void AMDGPUInstPrinter::printS13ImmDecOperand(const MCInst *MI, unsigned OpNo,
                                               raw_ostream &O) {
-  O << formatDec(static_cast<int16_t>(MI->getOperand(OpNo).getImm()));
+  O << formatDec(SignExtend32<13>(MI->getOperand(OpNo).getImm()));
 }
 
 void AMDGPUInstPrinter::printU32ImmOperand(const MCInst *MI, unsigned OpNo,
@@ -129,7 +129,7 @@ void AMDGPUInstPrinter::printOffsetS13(const MCInst *MI, unsigned OpNo,
   uint16_t Imm = MI->getOperand(OpNo).getImm();
   if (Imm != 0) {
     O << ((OpNo == 0)? "offset:" : " offset:");
-    printS16ImmDecOperand(MI, OpNo, O);
+    printS13ImmDecOperand(MI, OpNo, O);
   }
 }
 

diff --git a/lib/Target/AMDGPU/InstPrinter/AMDGPUInstPrinter.h b/lib/Target/AMDGPU/InstPrinter/AMDGPUInstPrinter.h
@@ -42,7 +42,7 @@ class AMDGPUInstPrinter : public MCInstPrinter {
   void printU4ImmDecOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
   void printU8ImmDecOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
   void printU16ImmDecOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
-  void printS16ImmDecOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
+  void printS13ImmDecOperand(const MCInst *MI, unsigned OpNo, raw_ostream &O);
   void printU32ImmOperand(const MCInst *MI, unsigned OpNo,
                           const MCSubtargetInfo &STI, raw_ostream &O);
   void printNamedBit(const MCInst *MI, unsigned OpNo, raw_ostream &O,

diff --git a/test/MC/Disassembler/AMDGPU/flat_gfx9.txt b/test/MC/Disassembler/AMDGPU/flat_gfx9.txt
@@ -0,0 +1,85 @@
+# RUN: llvm-mc -arch=amdgcn -mcpu=gfx901 -disassemble -show-encoding < %s | FileCheck %s
+
+# CHECK: flat_atomic_add v[0:1], v0    ; encoding: [0x00,0x00,0x08,0xdd,0x00,0x00,0x00,0x00]
+0x00,0x00,0x08,0xdd,0x00,0x00,0x00,0x00
+
+# CHECK: flat_atomic_add v[0:1], v0 offset:7    ; encoding: [0x07,0x00,0x08,0xdd,0x00,0x00,0x00,0x00]
+0x07,0x00,0x08,0xdd,0x00,0x00,0x00,0x00
+
+# CHECK: flat_atomic_add v0, v[0:1], v0 offset:4095 glc    ; encoding: [0xff,0x0f,0x09,0xdd,0x00,0x00,0x00,0x00]
+0xff,0x0f,0x09,0xdd,0x00,0x00,0x00,0x00
+
+# CHECK: flat_atomic_add v[0:1], v0 offset:4095 slc    ; encoding: [0xff,0x0f,0x0a,0xdd,0x00,0x00,0x00,0x00]
+0xff,0x0f,0x0a,0xdd,0x00,0x00,0x00,0x00
+
+# CHECK: flat_load_dword v0, v[0:1]    ; encoding: [0x00,0x00,0x50,0xdc,0x00,0x00,0x00,0x00]
+0x00,0x00,0x50,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: flat_load_dword v0, v[0:1] offset:7    ; encoding: [0x07,0x00,0x50,0xdc,0x00,0x00,0x00,0x00]
+0x07,0x00,0x50,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: flat_load_dword v0, v[0:1] offset:4095 glc    ; encoding: [0xff,0x0f,0x51,0xdc,0x00,0x00,0x00,0x00]
+0xff,0x0f,0x51,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: flat_store_byte v[0:1], v0    ; encoding: [0x00,0x00,0x60,0xdc,0x00,0x00,0x00,0x00]
+0x00,0x00,0x60,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: flat_store_byte v[0:1], v0 offset:7    ; encoding: [0x07,0x00,0x60,0xdc,0x00,0x00,0x00,0x00]
+0x07,0x00,0x60,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: flat_store_byte v[0:1], v0 offset:4095 glc    ; encoding: [0xff,0x0f,0x61,0xdc,0x00,0x00,0x00,0x00]
+0xff,0x0f,0x61,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: global_atomic_add v[2:3], v4, s[0:1]    ; encoding: [0x00,0x80,0x08,0xdd,0x02,0x04,0x00,0x00]
+0x00,0x80,0x08,0xdd,0x02,0x04,0x00,0x00
+
+# CHECK: global_atomic_add v[2:3], v4, s[0:1] offset:4095    ; encoding: [0xff,0x8f,0x08,0xdd,0x02,0x04,0x00,0x00]
+0xff,0x8f,0x08,0xdd,0x02,0x04,0x00,0x00
+
+# CHECK: global_atomic_add v[2:3], v4, s[0:1] offset:-4096    ; encoding: [0x00,0x90,0x08,0xdd,0x02,0x04,0x00,0x00]
+0x00,0x90,0x08,0xdd,0x02,0x04,0x00,0x00
+
+# CHECK: global_atomic_add v0, v[2:3], v4, s[0:1] offset:-1 glc    ; encoding: [0xff,0x9f,0x09,0xdd,0x02,0x04,0x00,0x00]
+0xff,0x9f,0x09,0xdd,0x02,0x04,0x00,0x00
+
+# CHECK: global_load_sbyte v0, v[2:3], s[0:1]    ; encoding: [0x00,0x80,0x44,0xdc,0x02,0x00,0x00,0x00]
+0x00,0x80,0x44,0xdc,0x02,0x00,0x00,0x00
+
+# CHECK: global_load_sbyte v0, v[2:3], s[0:1] offset:4095    ; encoding: [0xff,0x8f,0x44,0xdc,0x02,0x00,0x00,0x00]
+0xff,0x8f,0x44,0xdc,0x02,0x00,0x00,0x00
+
+# CHECK: global_load_sbyte v0, v[2:3], s[0:1] offset:-4096    ; encoding: [0x00,0x90,0x44,0xdc,0x02,0x00,0x00,0x00]
+0x00,0x90,0x44,0xdc,0x02,0x00,0x00,0x00
+
+# CHECK: global_store_dwordx2 v[2:3], v[4:5], s[0:1]    ; encoding: [0x00,0x80,0x74,0xdc,0x02,0x04,0x00,0x00]
+0x00,0x80,0x74,0xdc,0x02,0x04,0x00,0x00
+
+# CHECK: global_store_dwordx2 v[2:3], v[4:5], s[0:1] offset:4095    ; encoding: [0xff,0x8f,0x74,0xdc,0x02,0x04,0x00,0x00]
+0xff,0x8f,0x74,0xdc,0x02,0x04,0x00,0x00
+
+# CHECK: global_store_dwordx2 v[2:3], v[4:5], s[0:1] offset:-4096    ; encoding: [0x00,0x90,0x74,0xdc,0x02,0x04,0x00,0x00]
+0x00,0x90,0x74,0xdc,0x02,0x04,0x00,0x00
+
+# CHECK: scratch_load_dword v0, v0, off offset:-1    ; encoding: [0xff,0x5f,0x50,0xdc,0x00,0x00,0x7f,0x00]
+0xff,0x5f,0x50,0xdc,0x00,0x00,0x7f,0x00
+
+# CHECK: scratch_load_dword v0, off, s0    ; encoding: [0x00,0x40,0x50,0xdc,0x00,0x00,0x00,0x00]
+0x00,0x40,0x50,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: scratch_load_dword v0, off, s0 offset:4095    ; encoding: [0xff,0x4f,0x50,0xdc,0x00,0x00,0x00,0x00]
+0xff,0x4f,0x50,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: scratch_load_dword v0, off, s0 offset:-4096    ; encoding: [0x00,0x50,0x50,0xdc,0x00,0x00,0x00,0x00]
+0x00,0x50,0x50,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: scratch_store_short v0, v0, off offset:-1    ; encoding: [0xff,0x5f,0x68,0xdc,0x00,0x00,0x7f,0x00]
+0xff,0x5f,0x68,0xdc,0x00,0x00,0x7f,0x00
+
+# CHECK: scratch_store_short off, v0, s0    ; encoding: [0x00,0x40,0x68,0xdc,0x00,0x00,0x00,0x00]
+0x00,0x40,0x68,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: scratch_store_short off, v0, s0 offset:4095    ; encoding: [0xff,0x4f,0x68,0xdc,0x00,0x00,0x00,0x00]
+0xff,0x4f,0x68,0xdc,0x00,0x00,0x00,0x00
+
+# CHECK: scratch_store_short off, v0, s0 offset:-4096    ; encoding: [0x00,0x50,0x68,0xdc,0x00,0x00,0x00,0x00]
+0x00,0x50,0x68,0xdc,0x00,0x00,0x00,0x00