[AMDGPU][MC] Added support for several VI-specific opcodes (s_wakeup, etc)

[dp]

Summary

Added support for VI:

• s_endpgm_saved
• s_wakeup
• s_rfe_restore_b64
• v_perm_b32

Enabled for VI:

• v_mov_fed_b32
• v_mov_fed_b32_e64

See bug 32593: https://bugs.llvm.org//show_bug.cgi?id=32593

[artem.tamazov]

LGTM

lib/Target/AMDGPU/SOPInstructions.td

 > {
   let has_sdst = 0;
 }
 
 let Defs = [SCC] in {
 def S_ABSDIFF_I32 : SOP2_32 <"s_absdiff_i32">;
 } // End Defs = [SCC]
 
+let SubtargetPredicate = isVI in {
+  def S_RFE_RESTORE_B64 : SOP2_Pseudo <
+    "s_rfe_restore_b64", (outs),
+    (ins SSrc_b64:$src0, SSrc_b32:$src1),
+    "$src0, $src1"
+  > {
+    let hasSideEffects = 1;
+    let has_sdst = 0;
+  }
+}
+
 let SubtargetPredicate = isGFX9 in {
   def S_PACK_LL_B32_B16 : SOP2_32<"s_pack_ll_b32_b16">;
   def S_PACK_LH_B32_B16 : SOP2_32<"s_pack_lh_b32_b16">;
   def S_PACK_HH_B32_B16 : SOP2_32<"s_pack_hh_b32_b16">;
 }
 
 //===----------------------------------------------------------------------===//
 // SOPK Instructions

 
 def S_ENDPGM : SOPP <0x00000001, (ins), "s_endpgm",
   [(AMDGPUendpgm)]> {
   let simm16 = 0;
   let isBarrier = 1;
   let isReturn = 1;
 }
 
+let SubtargetPredicate = isVI in {
+def S_ENDPGM_SAVED : SOPP <0x0000001B, (ins), "s_endpgm_saved"> {
+  let simm16 = 0;
+  let isBarrier = 1;
+  let isReturn = 1;
+}
+}
+
 let isBranch = 1, SchedRW = [WriteBranch] in {
 def S_BRANCH : SOPP <
   0x00000002, (ins sopp_brtarget:$simm16), "s_branch $simm16",
   [(br bb:$simm16)]> {
   let isBarrier = 1;
 }
 
 let Uses = [SCC] in {
def S_BARRIER : SOPP <0x0000000a, (ins), "s_barrier",
   [(int_amdgcn_s_barrier)]> {
   let SchedRW = [WriteBarrier];
   let simm16 = 0;
   let mayLoad = 1;
   let mayStore = 1;
   let isConvergent = 1;
 }
 
+let SubtargetPredicate = isVI in {
+def S_WAKEUP : SOPP <0x00000003, (ins), "s_wakeup"> {
+  let simm16 = 0;
+  let mayLoad = 1;
+  let mayStore = 1;
+}
+}
+
 let mayLoad = 1, mayStore = 1, hasSideEffects = 1 in
 def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "s_waitcnt $simm16">;
 def S_SETHALT : SOPP <0x0000000d, (ins i16imm:$simm16), "s_sethalt $simm16">;
 def S_SETKILL : SOPP <0x0000000b, (ins i16imm:$simm16), "s_setkill $simm16">;
 
 // On SI the documentation says sleep for approximately 64 * low 2
 // bits, consistent with the reported maximum of 448. On VI the
 // maximum reported is 960 cycles, so 960 / 64 = 15 max, so is the

 def S_BFE_I32_vi           : SOP2_Real_vi <0x26, S_BFE_I32>;
 def S_BFE_U64_vi           : SOP2_Real_vi <0x27, S_BFE_U64>;
 def S_BFE_I64_vi           : SOP2_Real_vi <0x28, S_BFE_I64>;
 def S_CBRANCH_G_FORK_vi    : SOP2_Real_vi <0x29, S_CBRANCH_G_FORK>;
 def S_ABSDIFF_I32_vi       : SOP2_Real_vi <0x2a, S_ABSDIFF_I32>;
 def S_PACK_LL_B32_B16_vi   : SOP2_Real_vi <0x32, S_PACK_LL_B32_B16>;
 def S_PACK_LH_B32_B16_vi   : SOP2_Real_vi <0x33, S_PACK_LH_B32_B16>;
 def S_PACK_HH_B32_B16_vi   : SOP2_Real_vi <0x34, S_PACK_HH_B32_B16>;
+def S_RFE_RESTORE_B64_vi   : SOP2_Real_vi <0x2b, S_RFE_RESTORE_B64>;
 
 def S_MOVK_I32_vi          : SOPK_Real_vi <0x00, S_MOVK_I32>;
 def S_CMOVK_I32_vi         : SOPK_Real_vi <0x01, S_CMOVK_I32>;
 def S_CMPK_EQ_I32_vi       : SOPK_Real_vi <0x02, S_CMPK_EQ_I32>;
 def S_CMPK_LG_I32_vi       : SOPK_Real_vi <0x03, S_CMPK_LG_I32>;
 def S_CMPK_GT_I32_vi       : SOPK_Real_vi <0x04, S_CMPK_GT_I32>;
 def S_CMPK_GE_I32_vi       : SOPK_Real_vi <0x05, S_CMPK_GE_I32>;
 def S_CMPK_LT_I32_vi       : SOPK_Real_vi <0x06, S_CMPK_LT_I32>;

lib/Target/AMDGPU/VOP1Instructions.td

  // indexing from. Tablegen seems to not let you define an implicit
  // virtual register output for the super register being written into,
  // so this must have an implicit def of the register added to it.
 defm V_MOVRELD_B32 : VOP1Inst <"v_movreld_b32", VOP_MOVRELD>;
 defm V_MOVRELS_B32 : VOP1Inst <"v_movrels_b32", VOP_I32_VI32_NO_EXT>;
 defm V_MOVRELSD_B32 : VOP1Inst <"v_movrelsd_b32", VOP_NO_EXT<VOP_I32_I32>>;
 } // End Uses = [M0, EXEC]
 
+let SchedRW = [WriteQuarterRate32] in {
+defm V_MOV_FED_B32 : VOP1Inst <"v_mov_fed_b32", VOP_I32_I32>;
+}
+
 // These instruction only exist on SI and CI
 let SubtargetPredicate = isSICI in {
 
 let SchedRW = [WriteQuarterRate32] in {
-defm V_MOV_FED_B32 : VOP1Inst <"v_mov_fed_b32", VOP_I32_I32>;
 defm V_LOG_CLAMP_F32 : VOP1Inst <"v_log_clamp_f32", VOP_F32_F32, int_amdgcn_log_clamp>;
 defm V_RCP_CLAMP_F32 : VOP1Inst <"v_rcp_clamp_f32", VOP_F32_F32>;
 defm V_RCP_LEGACY_F32 : VOP1Inst <"v_rcp_legacy_f32", VOP_F32_F32, AMDGPUrcp_legacy>;
 defm V_RSQ_CLAMP_F32 : VOP1Inst <"v_rsq_clamp_f32", VOP_F32_F32, AMDGPUrsq_clamp>;
 defm V_RSQ_LEGACY_F32 : VOP1Inst <"v_rsq_legacy_f32", VOP_F32_F32, AMDGPUrsq_legacy>;
 } // End SchedRW = [WriteQuarterRate32]
 
 let SchedRW = [WriteDouble] in {

 defm V_NOP               : VOP1_Real_vi <0x0>;
 defm V_MOV_B32           : VOP1_Real_vi <0x1>;
 defm V_CVT_I32_F64       : VOP1_Real_vi <0x3>;
 defm V_CVT_F64_I32       : VOP1_Real_vi <0x4>;
 defm V_CVT_F32_I32       : VOP1_Real_vi <0x5>;
 defm V_CVT_F32_U32       : VOP1_Real_vi <0x6>;
 defm V_CVT_U32_F32       : VOP1_Real_vi <0x7>;
 defm V_CVT_I32_F32       : VOP1_Real_vi <0x8>;
+defm V_MOV_FED_B32       : VOP1_Real_vi <0x9>;
 defm V_CVT_F16_F32       : VOP1_Real_vi <0xa>;
 defm V_CVT_F32_F16       : VOP1_Real_vi <0xb>;
 defm V_CVT_RPI_I32_F32   : VOP1_Real_vi <0xc>;
 defm V_CVT_FLR_I32_F32   : VOP1_Real_vi <0xd>;
 defm V_CVT_OFF_F32_I4    : VOP1_Real_vi <0xe>;
 defm V_CVT_F32_F64       : VOP1_Real_vi <0xf>;
 defm V_CVT_F64_F32       : VOP1_Real_vi <0x10>;
 defm V_CVT_F32_UBYTE0    : VOP1_Real_vi <0x11>;

lib/Target/AMDGPU/VOP3Instructions.td

 def V_INTERP_P2_F16   : VOP3Inst <"v_interp_p2_f16", VOP3_Profile<VOP_F16_F32_F16_F32>>;
 def V_MAD_F16         : VOP3Inst <"v_mad_f16", VOP3_Profile<VOP_F16_F16_F16_F16>, fmad>;
 
 def V_MAD_U16 : VOP3Inst <"v_mad_u16", VOP3_Profile<VOP_I16_I16_I16_I16>>;
 def V_MAD_I16 : VOP3Inst <"v_mad_i16", VOP3_Profile<VOP_I16_I16_I16_I16>>;
 
 }  // End isCommutable = 1
 
+def V_PERM_B32 : VOP3Inst <"v_perm_b32", VOP3_Profile<VOP_I32_I32_I32_I32>>;
+
 } // End SubtargetPredicate = isVI
 
 let Predicates = [isVI] in {
 
 multiclass Ternary_i16_Pats <SDPatternOperator op1, SDPatternOperator op2,
                              Instruction inst, SDPatternOperator op3> {
 def : Pat<
   (op2 (op1 i16:$src0, i16:$src1), i16:$src2),

 defm V_QSAD_PK_U16_U8   : VOP3_Real_vi <0x1e5>;
 defm V_MQSAD_PK_U16_U8  : VOP3_Real_vi <0x1e6>;
 defm V_MQSAD_U32_U8     : VOP3_Real_vi <0x1e7>;
 
 defm V_MAD_F16          : VOP3_Real_vi <0x1ea>;
 defm V_MAD_U16          : VOP3_Real_vi <0x1eb>;
 defm V_MAD_I16          : VOP3_Real_vi <0x1ec>;
 
+defm V_PERM_B32         : VOP3_Real_vi <0x1ed>;
+
 defm V_FMA_F16          : VOP3_Real_vi <0x1ee>;
 defm V_DIV_FIXUP_F16    : VOP3_Real_vi <0x1ef>;
 
 defm V_INTERP_P1LL_F16  : VOP3_Real_vi <0x274>;
 defm V_INTERP_P1LV_F16  : VOP3_Real_vi <0x275>;
 defm V_INTERP_P2_F16    : VOP3_Real_vi <0x276>;
 defm V_ADD_F64          : VOP3_Real_vi <0x280>;
 defm V_MUL_F64          : VOP3_Real_vi <0x281>;

test/MC/AMDGPU/gfx8_asm_all.s

 // CHECK: [0xf9,0x04,0xbe,0x7d,0x01,0x16,0x06,0x04]
 
 v_cmpx_t_u32_sdwa vcc, v1, v2 src0_sel:DWORD src1_sel:WORD_1
 // CHECK: [0xf9,0x04,0xbe,0x7d,0x01,0x16,0x06,0x05]
 
 v_cmpx_t_u32_sdwa vcc, v1, sext(v2) src0_sel:DWORD src1_sel:DWORD
 // CHECK: [0xf9,0x04,0xbe,0x7d,0x01,0x16,0x06,0x0e]
 
+s_rfe_restore_b64 s[4:5], s2
+// CHECK: [0x04,0x02,0x80,0x95]
+
+v_mov_fed_b32_dpp v5, v1 quad_perm:[0,1,2,3] row_mask:0x0 bank_mask:0x0
+// CHECK: [0xfa,0x12,0x0a,0x7e,0x01,0xe4,0x00,0x00]
+
+v_mov_fed_b32_e64 v5, s1
+// CHECK: [0x05,0x00,0x49,0xd1,0x01,0x00,0x00,0x00]
+
+v_mov_fed_b32_sdwa v5, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD
+// CHECK: [0xf9,0x12,0x0a,0x7e,0x01,0x06,0x06,0x06]
+
+v_perm_b32 v5, s1, 0, v255
+// CHECK: [0x05,0x00,0xed,0xd1,0x01,0x00,0xfd,0x07]

test/MC/AMDGPU/sopp.s

 
 s_set_gpr_idx_mode 0
 // VI: s_set_gpr_idx_mode 0 ; encoding: [0x00,0x00,0x9d,0xbf]
 // NOSICI: error: instruction not supported on this GPU
 
 s_set_gpr_idx_mode 15
 // VI: s_set_gpr_idx_mode dst src0 src1 src2 ; encoding: [0x0f,0x00,0x9d,0xbf]
 // NOSICI: error: instruction not supported on this GPU
+
+s_endpgm_saved
+// VI: s_endpgm_saved ; encoding: [0x00,0x00,0x9b,0xbf]
+// NOSICI: error: instruction not supported on this GPU
+
+s_wakeup
+// VI: s_wakeup ; encoding: [0x00,0x00,0x83,0xbf]
+// NOSICI: error: instruction not supported on this GPU

test/MC/AMDGPU/vop1.s

 
 // GCN: v_cvt_u32_f32_e32 v1, v2 ; encoding: [0x02,0x0f,0x02,0x7e
 v_cvt_u32_f32_e32 v1, v2
 
 // GCN: v_cvt_i32_f32_e32 v1, v2 ; encoding: [0x02,0x11,0x02,0x7e]
 v_cvt_i32_f32_e32 v1, v2
 
 // SICI: v_mov_fed_b32_e32 v1, v2 ; encoding: [0x02,0x13,0x02,0x7e]
-// NOVI: error: instruction not supported on this GPU
+// VI:   v_mov_fed_b32_e32 v1, v2 ; encoding: [0x02,0x13,0x02,0x7e]
 v_mov_fed_b32_e32 v1, v2
 
 // GCN: v_cvt_f16_f32_e32 v1, v2 ; encoding: [0x02,0x15,0x02,0x7e]
 v_cvt_f16_f32_e32 v1, v2
 
 // GCN: v_cvt_f32_f16_e32 v1, v2 ; encoding: [0x02,0x17,0x02,0x7e]
 v_cvt_f32_f16_e32 v1, v2