Support 96b dst for gfx12. Also added disassembler tests.

Author: jwanggit86

llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp

@@ -9764,6 +9764,7 @@ unsigned AMDGPUAsmParser::validateTargetOperandClass(MCParsedAsmOperand &Op,
     // The following code enables it for SReg_64 and larger operands
     // used as source and destination. Remaining source
     // operands are handled in isInlinableImm.
+  case MCK_SReg_96:
   case MCK_SReg_128:
   case MCK_SReg_256:
   case MCK_SReg_512:

llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.cpp

@@ -1695,10 +1695,11 @@ AMDGPUDisassembler::decodeNonVGPRSrcOp(const OpWidthTy Width, unsigned Val,
   case OPW64:
   case OPWV232:
     return decodeSpecialReg64(Val);
+  case OPW96:
   case OPW128:
   case OPW256:
   case OPW512:
-    return decodeSpecialReg128(Val);
+    return decodeSpecialReg96Plus(Val);
   default:
     llvm_unreachable("unexpected immediate type");
   }
@@ -1785,7 +1786,7 @@ MCOperand AMDGPUDisassembler::decodeSpecialReg64(unsigned Val) const {
   return errOperand(Val, "unknown operand encoding " + Twine(Val));
 }
 
-MCOperand AMDGPUDisassembler::decodeSpecialReg128(unsigned Val) const {
+MCOperand AMDGPUDisassembler::decodeSpecialReg96Plus(unsigned Val) const {
   using namespace AMDGPU;
 
   switch (Val) {

llvm/lib/Target/AMDGPU/Disassembler/AMDGPUDisassembler.h

@@ -259,7 +259,7 @@ class AMDGPUDisassembler : public MCDisassembler {
   MCOperand decodeVOPDDstYOp(MCInst &Inst, unsigned Val) const;
   MCOperand decodeSpecialReg32(unsigned Val) const;
   MCOperand decodeSpecialReg64(unsigned Val) const;
-  MCOperand decodeSpecialReg128(unsigned Val) const;
+  MCOperand decodeSpecialReg96Plus(unsigned Val) const;
 
   MCOperand decodeSDWASrc(const OpWidthTy Width, unsigned Val,
                           unsigned ImmWidth,

llvm/test/MC/AMDGPU/gfx12_asm_smem.s

@@ -548,6 +548,9 @@ s_load_b32 null, s[2:3], s0 offset:0x0
 s_load_b64 null, s[2:3], s0 offset:0x0
 // GFX12: encoding: [0x01,0x3f,0x00,0xf4,0x00,0x00,0x00,0x00]
 
+s_load_b96 null, s[2:3], s0 offset:0x0
+// GFX12: encoding: [0x01,0xbf,0x00,0xf4,0x00,0x00,0x00,0x00]
+
 s_load_b128 null, s[2:3], s0 offset:0x0
 // GFX12: encoding: [0x01,0x5f,0x00,0xf4,0x00,0x00,0x00,0x00]
 

llvm/test/MC/Disassembler/AMDGPU/gfx10_smem.txt

@@ -1261,3 +1261,18 @@
 
 # GFX10: s_store_dwordx4 s[96:99], s[4:5], s0    ; encoding: [0x02,0x18,0x48,0xf4,0x00,0x00,0x00,0x00]
 0x02,0x18,0x48,0xf4,0x00,0x00,0x00,0x00
+
+# GFX10: s_load_dword null, s[2:3], s0           ; encoding: [0x41,0x1f,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x41,0x1f,0x00,0xf4,0x00,0x00,0x00,0x00
+
+# GFX10: s_load_dwordx2 null, s[2:3], s0         ; encoding: [0x41,0x1f,0x04,0xf4,0x00,0x00,0x00,0x00]
+0x41,0x1f,0x04,0xf4,0x00,0x00,0x00,0x00
+
+# GFX10: s_load_dwordx4 null, s[2:3], s0         ; encoding: [0x41,0x1f,0x08,0xf4,0x00,0x00,0x00,0x00]
+0x41,0x1f,0x08,0xf4,0x00,0x00,0x00,0x00
+
+# GFX10: s_load_dwordx8 null, s[2:3], s0         ; encoding: [0x41,0x1f,0x0c,0xf4,0x00,0x00,0x00,0x00]
+0x41,0x1f,0x0c,0xf4,0x00,0x00,0x00,0x00
+
+# GFX10: s_load_dwordx16 null, s[2:3], s0        ; encoding: [0x41,0x1f,0x10,0xf4,0x00,0x00,0x00,0x00]
+0x41,0x1f,0x10,0xf4,0x00,0x00,0x00,0x00

llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_smem.txt

@@ -471,3 +471,18 @@
 
 # GFX11: s_gl1_inv                               ; encoding: [0x00,0x00,0x80,0xf4,0x00,0x00,0x00,0x00]
 0x00,0x00,0x80,0xf4,0x00,0x00,0x00,0x00
+
+# GFX11: s_load_b32 null, s[2:3], s0             ; encoding: [0x01,0x1f,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x1f,0x00,0xf4,0x00,0x00,0x00,0x00
+
+# GFX11: s_load_b64 null, s[2:3], s0             ; encoding: [0x01,0x1f,0x04,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x1f,0x04,0xf4,0x00,0x00,0x00,0x00
+
+# GFX11: s_load_b128 null, s[2:3], s0            ; encoding: [0x01,0x1f,0x08,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x1f,0x08,0xf4,0x00,0x00,0x00,0x00
+
+# GFX11: s_load_b256 null, s[2:3], s0            ; encoding: [0x01,0x1f,0x0c,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x1f,0x0c,0xf4,0x00,0x00,0x00,0x00
+
+# GFX11: s_load_b512 null, s[2:3], s0            ; encoding: [0x01,0x1f,0x10,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x1f,0x10,0xf4,0x00,0x00,0x00,0x00

llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_smem.txt

@@ -1277,3 +1277,21 @@
 
 # GFX12: s_buffer_load_u16 s5, s[96:99], s0 offset:0x0 th:TH_LOAD_HT scope:SCOPE_SYS ; encoding:  [0x70,0x61,0x63,0xf5,0x00,0x00,0x00,0x00]
 0x70,0x61,0x63,0xf5,0x00,0x00,0x00,0x00
+
+# GFX12: s_load_b32 null, s[2:3], s0 offset:0x0  ; encoding: [0x01,0x1f,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x1f,0x00,0xf4,0x00,0x00,0x00,0x00
+
+# GFX12: s_load_b64 null, s[2:3], s0 offset:0x0  ; encoding: [0x01,0x3f,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x3f,0x00,0xf4,0x00,0x00,0x00,0x00
+
+# GFX12: s_load_b96 null, s[2:3], s0 offset:0x0  ; encoding: [0x01,0xbf,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x01,0xbf,0x00,0xf4,0x00,0x00,0x00,0x00
+
+# GFX12: s_load_b128 null, s[2:3], s0 offset:0x0  ; encoding: [0x01,0x5f,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x5f,0x00,0xf4,0x00,0x00,0x00,0x00
+
+# GFX12: s_load_b256 null, s[2:3], s0 offset:0x0  ; encoding: [0x01,0x7f,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x7f,0x00,0xf4,0x00,0x00,0x00,0x00
+
+# GFX12: s_load_b512 null, s[2:3], s0 offset:0x0  ; encoding: [0x01,0x9f,0x00,0xf4,0x00,0x00,0x00,0x00]
+0x01,0x9f,0x00,0xf4,0x00,0x00,0x00,0x00