[AMDGPU][MC][VI][GFX9] Added s_atc_probe* instructions

[dp]

Summary

See bug 36839: https://bugs.llvm.org/show_bug.cgi?id=36839

[artem.tamazov]

Great, thanks.

lib/Target/AMDGPU/SMInstructions.td

class SM_Real <SM_Pseudo ps>
 
   // encoding
   bits<7>  sbase;
   bits<7>  sdst;
   bits<32> offset;
   bits<1> imm = !if(ps.has_offset, ps.offset_is_imm, 0);
 }
 
+class SM_Probe_Pseudo <string opName, dag ins, bit isImm>
+  : SM_Pseudo<opName, (outs), ins, " $sdata, $sbase, $offset"> {
+  let mayLoad = 0;
+  let mayStore = 0;
+  let has_glc = 0;
+  let LGKM_CNT = 0;
+  let ScalarStore = 0;
+  let hasSideEffects = 1;
+  let offset_is_imm = isImm;
+  let PseudoInstr = opName # !if(isImm, "_IMM", "_SGPR");
+}
+
 class SM_Load_Pseudo <string opName, dag outs, dag ins, string asmOps, list<dag> pattern=[]>
   : SM_Pseudo<opName, outs, ins, asmOps, pattern> {
   RegisterClass BaseClass;
   let mayLoad = 1;
   let mayStore = 0;
   let has_glc = 1;
 }
 
class SM_Inval_Pseudo <string opName, SDPatternOperator node> : SM_Pseudo<
   opName, (outs), (ins), "", [(node)]> {
   let hasSideEffects = 1;
   let mayStore = 1;
   let has_sdst = 0;
   let has_sbase = 0;
   let has_offset = 0;
 }
 
+multiclass SM_Pseudo_Probe<string opName, RegisterClass baseClass> {
+  def _IMM  : SM_Probe_Pseudo <opName, (ins i8imm:$sdata, baseClass:$sbase, smrd_offset_20:$offset), 1>;
+  def _SGPR : SM_Probe_Pseudo <opName, (ins i8imm:$sdata, baseClass:$sbase, SReg_32:$offset), 0>;
+}
+
 //===----------------------------------------------------------------------===//
 // Scalar Atomic Memory Classes
 //===----------------------------------------------------------------------===//
 
 class SM_Atomic_Pseudo <string opName,
                         dag outs, dag ins, string asmOps, bit isRet>
   : SM_Pseudo<opName, outs, ins, asmOps, []> {
 

 let SubtargetPredicate = isCIVI in {
 def S_DCACHE_INV_VOL : SM_Inval_Pseudo <"s_dcache_inv_vol", int_amdgcn_s_dcache_inv_vol>;
 } // let SubtargetPredicate = isCIVI
 
 let SubtargetPredicate = isVI in {
 def S_DCACHE_WB     : SM_Inval_Pseudo <"s_dcache_wb", int_amdgcn_s_dcache_wb>;
 def S_DCACHE_WB_VOL : SM_Inval_Pseudo <"s_dcache_wb_vol", int_amdgcn_s_dcache_wb_vol>;
 def S_MEMREALTIME   : SM_Time_Pseudo <"s_memrealtime", int_amdgcn_s_memrealtime>;
+
+defm S_ATC_PROBE        : SM_Pseudo_Probe <"s_atc_probe", SReg_64>;
+defm S_ATC_PROBE_BUFFER : SM_Pseudo_Probe <"s_atc_probe_buffer", SReg_128>;
 } // SubtargetPredicate = isVI
 
 let SubtargetPredicate = HasFlatScratchInsts, Uses = [FLAT_SCR] in {
 defm S_SCRATCH_LOAD_DWORD    : SM_Pseudo_Loads <"s_scratch_load_dword",   SReg_64, SReg_32_XM0_XEXEC>;
 defm S_SCRATCH_LOAD_DWORDX2  : SM_Pseudo_Loads <"s_scratch_load_dwordx2", SReg_64, SReg_64_XEXEC>;
 defm S_SCRATCH_LOAD_DWORDX4  : SM_Pseudo_Loads <"s_scratch_load_dwordx4", SReg_64, SReg_128>;
 
 defm S_SCRATCH_STORE_DWORD   : SM_Pseudo_Stores <"s_scratch_store_dword",   SReg_64, SReg_32_XM0_XEXEC>;
def _IMM_vi : SMEM_Real_Store_vi <op, immPs> {
     let InOperandList = (ins immPs.SrcClass:$sdata, immPs.BaseClass:$sbase, smrd_offset_20:$offset, GLC:$glc);
   }
 
   def _SGPR_vi : SMEM_Real_Store_vi <op, sgprPs> {
     let InOperandList = (ins sgprPs.SrcClass:$sdata, sgprPs.BaseClass:$sbase, SReg_32:$offset, GLC:$glc);
   }
 }
 
+multiclass SM_Real_Probe_vi<bits<8> op, string ps> {
+  def _IMM_vi  : SMEM_Real_Store_vi <op, !cast<SM_Probe_Pseudo>(ps#_IMM)>;
+  def _SGPR_vi : SMEM_Real_Store_vi <op, !cast<SM_Probe_Pseudo>(ps#_SGPR)>;
+}
+
 defm S_LOAD_DWORD           : SM_Real_Loads_vi <0x00, "S_LOAD_DWORD">;
 defm S_LOAD_DWORDX2         : SM_Real_Loads_vi <0x01, "S_LOAD_DWORDX2">;
 defm S_LOAD_DWORDX4         : SM_Real_Loads_vi <0x02, "S_LOAD_DWORDX4">;
 defm S_LOAD_DWORDX8         : SM_Real_Loads_vi <0x03, "S_LOAD_DWORDX8">;
 defm S_LOAD_DWORDX16        : SM_Real_Loads_vi <0x04, "S_LOAD_DWORDX16">;
 defm S_BUFFER_LOAD_DWORD    : SM_Real_Loads_vi <0x08, "S_BUFFER_LOAD_DWORD">;
 defm S_BUFFER_LOAD_DWORDX2  : SM_Real_Loads_vi <0x09, "S_BUFFER_LOAD_DWORDX2">;
 defm S_BUFFER_LOAD_DWORDX4  : SM_Real_Loads_vi <0x0a, "S_BUFFER_LOAD_DWORDX4">;

 defm S_SCRATCH_LOAD_DWORD    : SM_Real_Loads_vi <0x05, "S_SCRATCH_LOAD_DWORD">;
 defm S_SCRATCH_LOAD_DWORDX2  : SM_Real_Loads_vi <0x06, "S_SCRATCH_LOAD_DWORDX2">;
 defm S_SCRATCH_LOAD_DWORDX4  : SM_Real_Loads_vi <0x07, "S_SCRATCH_LOAD_DWORDX4">;
 
 defm S_SCRATCH_STORE_DWORD   : SM_Real_Stores_vi <0x15, "S_SCRATCH_STORE_DWORD">;
 defm S_SCRATCH_STORE_DWORDX2 : SM_Real_Stores_vi <0x16, "S_SCRATCH_STORE_DWORDX2">;
 defm S_SCRATCH_STORE_DWORDX4 : SM_Real_Stores_vi <0x17, "S_SCRATCH_STORE_DWORDX4">;
 
+defm S_ATC_PROBE        : SM_Real_Probe_vi <0x26, "S_ATC_PROBE">;
+defm S_ATC_PROBE_BUFFER : SM_Real_Probe_vi <0x27, "S_ATC_PROBE_BUFFER">;
+
 //===----------------------------------------------------------------------===//
 // GFX9
 //===----------------------------------------------------------------------===//
 
 class SMEM_Atomic_Real_vi <bits<8> op, SM_Atomic_Pseudo ps>
   : SMEM_Real_vi <op, ps> {
 
   bits<7> sdata;

test/MC/AMDGPU/smem.s

 s_dcache_wb
 // GFX89: s_dcache_wb  ; encoding: [0x00,0x00,0x84,0xc0,0x00,0x00,0x00,0x00]
 // NOSICI: error: instruction not supported on this GPU
 
 s_dcache_wb_vol
 // GFX89: s_dcache_wb_vol ; encoding: [0x00,0x00,0x8c,0xc0,0x00,0x00,0x00,0x00]
 // NOSICI: error: instruction not supported on this GPU
 
+s_atc_probe 0x7, s[4:5], s0
+// GFX89:  s_atc_probe 7, s[4:5], s0 ; encoding: [0xc2,0x01,0x98,0xc0,0x00,0x00,0x00,0x00]
+// NOSICI: error: instruction not supported on this GPU
+
+s_atc_probe 0x0, s[4:5], 0x0
+// GFX89:  s_atc_probe 0, s[4:5], 0x0 ; encoding: [0x02,0x00,0x9a,0xc0,0x00,0x00,0x00,0x00]
+// NOSICI: error: instruction not supported on this GPU
+
+s_atc_probe_buffer 0x1, s[8:11], s0
+// GFX89:  s_atc_probe_buffer 1, s[8:11], s0 ; encoding: [0x44,0x00,0x9c,0xc0,0x00,0x00,0x00,0x00]
+// NOSICI: error: instruction not supported on this GPU
+
+s_atc_probe_buffer 0x0, s[8:11], s101
+// GFX89:  s_atc_probe_buffer 0, s[8:11], s101 ; encoding: [0x04,0x00,0x9c,0xc0,0x65,0x00,0x00,0x00]
+// NOSICI: error: instruction not supported on this GPU
+
 s_memrealtime s[4:5]
 // GFX89: s_memrealtime s[4:5] ; encoding: [0x00,0x01,0x94,0xc0,0x00,0x00,0x00,0x00]
 // NOSICI: error: instruction not supported on this GPU
 
 s_memrealtime tba
 // VI: s_memrealtime tba ; encoding: [0x00,0x1b,0x94,0xc0,0x00,0x00,0x00,0x00]
 // NOSICI: error: instruction not supported on this GPU
 // NOGFX9: error: not a valid operand.

test/MC/Disassembler/AMDGPU/smem_vi.txt

 # VI: s_buffer_load_dword s10, s[92:95], m0 ; encoding: [0xae,0x02,0x20,0xc0,0x7c,0x00,0x00,0x00]
 0xae 0x02 0x20 0xc0 0x7c 0x00 0x00 0x00
 
 # VI: s_buffer_load_dwordx2 s[10:11], s[92:95], m0 ; encoding: [0xae,0x02,0x24,0xc0,0x7c,0x00,0x00,0x00]
 0xae 0x02 0x24 0xc0 0x7c 0x00 0x00 0x00
 
 # VI: s_buffer_load_dwordx4 s[8:11], s[92:95], m0 glc ; encoding: [0x2e,0x02,0x29,0xc0,0x7c,0x00,0x00,0x00]
 0x2e 0x02 0x29 0xc0 0x7c 0x00 0x00 0x00
+
+# VI: s_atc_probe 7, s[4:5], s0 ; encoding: [0xc2,0x01,0x98,0xc0,0x00,0x00,0x00,0x00]
+0xc2,0x01,0x98,0xc0,0x00,0x00,0x00,0x00
+
+# VI: s_atc_probe 0, s[4:5], 0x0 ; encoding: [0x02,0x00,0x9a,0xc0,0x00,0x00,0x00,0x00]
+0x02,0x00,0x9a,0xc0,0x00,0x00,0x00,0x00
+
+# VI: s_atc_probe_buffer 1, s[8:11], s0 ; encoding: [0x44,0x00,0x9c,0xc0,0x00,0x00,0x00,0x00]
+0x44,0x00,0x9c,0xc0,0x00,0x00,0x00,0x00
+
+# VI: s_atc_probe_buffer 0, s[8:11], s101 ; encoding: [0x04,0x00,0x9c,0xc0,0x65,0x00,0x00,0x00]
+0x04,0x00,0x9c,0xc0,0x65,0x00,0x00,0x00