AMDGPU: Fix asserts on invalid register ranges

If the requested SGPR was not actually aligned, it was
accepted and rounded down instead of rejected.

Also fix an assert if the range is an invalid size.

git-svn-id: https://llvm.org/svn/llvm-project/llvm/trunk@252009 91177308-0d34-0410-b5e6-96231b3b80d8

diff --git a/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp b/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
index a2420ad20700315a42c8b00ee05313d9d731f13c..3e3ceaa06b7008d51909afe87b60f716d68d47dc 100644 (file)
--- a/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
+++ b/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
@@ -447,10 +447,10 @@ struct OptionalOperand {
 
 }
 
 
 }
 

-static unsigned getRegClass(bool IsVgpr, unsigned RegWidth) {
+static int getRegClass(bool IsVgpr, unsigned RegWidth) {

   if (IsVgpr) {
     switch (RegWidth) {
   if (IsVgpr) {
     switch (RegWidth) {

-      default: llvm_unreachable("Unknown register width");
+      default: return -1;

       case 1: return AMDGPU::VGPR_32RegClassID;
       case 2: return AMDGPU::VReg_64RegClassID;
       case 3: return AMDGPU::VReg_96RegClassID;
       case 1: return AMDGPU::VGPR_32RegClassID;
       case 2: return AMDGPU::VReg_64RegClassID;
       case 3: return AMDGPU::VReg_96RegClassID;


@@ -461,7 +461,7 @@ static unsigned getRegClass(bool IsVgpr, unsigned RegWidth) {
   }
 
   switch (RegWidth) {
   }
 
   switch (RegWidth) {

-    default: llvm_unreachable("Unknown register width");
+    default: return -1;

     case 1: return AMDGPU::SGPR_32RegClassID;
     case 2: return AMDGPU::SGPR_64RegClassID;
     case 4: return AMDGPU::SReg_128RegClassID;
     case 1: return AMDGPU::SGPR_32RegClassID;
     case 2: return AMDGPU::SGPR_64RegClassID;
     case 4: return AMDGPU::SReg_128RegClassID;


@@ -541,12 +541,20 @@ bool AMDGPUAsmParser::ParseRegister(unsigned &RegNo, SMLoc &StartLoc, SMLoc &End
       RegIndexInClass = RegLo;
     } else {
       // SGPR registers are aligned.  Max alignment is 4 dwords.
       RegIndexInClass = RegLo;
     } else {
       // SGPR registers are aligned.  Max alignment is 4 dwords.

-      RegIndexInClass = RegLo / std::min(RegWidth, 4u);
+      unsigned Size = std::min(RegWidth, 4u);
+      if (RegLo % Size != 0)
+        return true;
+
+      RegIndexInClass = RegLo / Size;

     }
   }
 
   const MCRegisterInfo *TRI = getContext().getRegisterInfo();
     }
   }
 
   const MCRegisterInfo *TRI = getContext().getRegisterInfo();

-  const MCRegisterClass RC = TRI->getRegClass(getRegClass(IsVgpr, RegWidth));
+  int RCID = getRegClass(IsVgpr, RegWidth);
+  if (RCID == -1)
+    return true;
+
+  const MCRegisterClass RC = TRI->getRegClass(RCID);

   if (RegIndexInClass >= RC.getNumRegs())
     return true;
 
   if (RegIndexInClass >= RC.getNumRegs())
     return true;
 

diff --git a/test/MC/AMDGPU/out-of-range-registers.s b/test/MC/AMDGPU/out-of-range-registers.s
index 312c78e73efb65ee2e4cc7f0572b6529f72edf72..947c64d3e642eda1ca44ababfb0d88ebc519c50c 100644 (file)
--- a/test/MC/AMDGPU/out-of-range-registers.s
+++ b/test/MC/AMDGPU/out-of-range-registers.s
@@ -12,3 +12,51 @@ v_add_i32 v256, v0, v1
 
 v_add_i32 v257, v0, v1
 // CHECK: error: invalid operand for instruction
 
 v_add_i32 v257, v0, v1
 // CHECK: error: invalid operand for instruction

+
+s_mov_b64 s[0:17], -1
+// CHECK: error: invalid operand for instruction
+
+s_mov_b64 s[103:104], -1
+// CHECK: error: invalid operand for instruction
+
+s_mov_b64 s[104:105], -1
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx4 s[102:105], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx4 s[104:108], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx4 s[108:112], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx4 s[1:4], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx4 s[1:4], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx8 s[104:111], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx8 s[100:107], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx8 s[108:115], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx16 s[92:107], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx16 s[96:111], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx16 s[100:115], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx16 s[104:119], s[2:3], s4
+// CHECK: error: invalid operand for instruction
+
+s_load_dwordx16 s[108:123], s[2:3], s4
+// CHECK: error: invalid operand for instruction

diff --git a/test/MC/AMDGPU/smrd.s b/test/MC/AMDGPU/smrd.s
index 9b8471b19d29a4d66d88225bf50bb487b2d1bdf4..56841914c6f0f1a38c9e8761da1e678d06590ddf 100644 (file)
--- a/test/MC/AMDGPU/smrd.s
+++ b/test/MC/AMDGPU/smrd.s
@@ -40,18 +40,27 @@ s_load_dwordx4 s[4:7], s[2:3], 1
 s_load_dwordx4 s[4:7], s[2:3], s4
 // GCN: s_load_dwordx4 s[4:7], s[2:3], s4 ; encoding: [0x04,0x02,0x82,0xc0]
 
 s_load_dwordx4 s[4:7], s[2:3], s4
 // GCN: s_load_dwordx4 s[4:7], s[2:3], s4 ; encoding: [0x04,0x02,0x82,0xc0]
 

+s_load_dwordx4 s[100:103], s[2:3], s4
+// GCN: s_load_dwordx4 s[100:103], s[2:3], s4 ; encoding: [0x04,0x02,0xb2,0xc0]
+

 s_load_dwordx8 s[8:15], s[2:3], 1
 // GCN: s_load_dwordx8 s[8:15], s[2:3], 0x1 ; encoding: [0x01,0x03,0xc4,0xc0]
 
 s_load_dwordx8 s[8:15], s[2:3], s4
 // GCN: s_load_dwordx8 s[8:15], s[2:3], s4 ; encoding: [0x04,0x02,0xc4,0xc0]
 
 s_load_dwordx8 s[8:15], s[2:3], 1
 // GCN: s_load_dwordx8 s[8:15], s[2:3], 0x1 ; encoding: [0x01,0x03,0xc4,0xc0]
 
 s_load_dwordx8 s[8:15], s[2:3], s4
 // GCN: s_load_dwordx8 s[8:15], s[2:3], s4 ; encoding: [0x04,0x02,0xc4,0xc0]
 

+s_load_dwordx8 s[96:103], s[2:3], s4
+// GCN: s_load_dwordx8 s[96:103], s[2:3], s4 ; encoding: [0x04,0x02,0xf0,0xc0]
+

 s_load_dwordx16 s[16:31], s[2:3], 1
 // GCN: s_load_dwordx16 s[16:31], s[2:3], 0x1 ; encoding: [0x01,0x03,0x08,0xc1]
 
 s_load_dwordx16 s[16:31], s[2:3], s4
 // GCN: s_load_dwordx16 s[16:31], s[2:3], s4 ; encoding: [0x04,0x02,0x08,0xc1]
 
 s_load_dwordx16 s[16:31], s[2:3], 1
 // GCN: s_load_dwordx16 s[16:31], s[2:3], 0x1 ; encoding: [0x01,0x03,0x08,0xc1]
 
 s_load_dwordx16 s[16:31], s[2:3], s4
 // GCN: s_load_dwordx16 s[16:31], s[2:3], s4 ; encoding: [0x04,0x02,0x08,0xc1]
 

+s_load_dwordx16 s[88:103], s[2:3], s4
+// GCN: s_load_dwordx16 s[88:103], s[2:3], s4 ; encoding: [0x04,0x02,0x2c,0xc1]
+

 s_dcache_inv
 // GCN: s_dcache_inv ; encoding: [0x00,0x00,0xc0,0xc7]
 
 s_dcache_inv
 // GCN: s_dcache_inv ; encoding: [0x00,0x00,0xc0,0xc7]
 

diff --git a/test/MC/AMDGPU/sop1.s b/test/MC/AMDGPU/sop1.s
index a99bc9ab1cd642c51b71dc4f0d47b6f8c3e3b17c..5f63f9930747dc61e3d2edc89bf781ccd8cdbe81 100644 (file)
--- a/test/MC/AMDGPU/sop1.s
+++ b/test/MC/AMDGPU/sop1.s
@@ -30,6 +30,9 @@ s_mov_b64 s[2:3], 0xffffffff
 s_mov_b64 s[0:1], 0x80000000
 // CHECK: s_mov_b64 s[0:1], 0x80000000 ; encoding: [0xff,0x04,0x80,0xbe,0x00,0x00,0x00,0x80]
 
 s_mov_b64 s[0:1], 0x80000000
 // CHECK: s_mov_b64 s[0:1], 0x80000000 ; encoding: [0xff,0x04,0x80,0xbe,0x00,0x00,0x00,0x80]
 

+s_mov_b64 s[102:103], -1
+// CHECK: s_mov_b64 s[102:103], -1 ; encoding: [0xc1,0x04,0xe6,0xbe]
+

 s_cmov_b32 s1, 200
 // CHECK: s_cmov_b32 s1, 0xc8 ; encoding: [0xff,0x05,0x81,0xbe,0xc8,0x00,0x00,0x00]
 
 s_cmov_b32 s1, 200
 // CHECK: s_cmov_b32 s1, 0xc8 ; encoding: [0xff,0x05,0x81,0xbe,0xc8,0x00,0x00,0x00]