diff options
Diffstat (limited to 'lib/Target/R600/SIInstructions.td')
| -rw-r--r-- | lib/Target/R600/SIInstructions.td | 108 |
1 files changed, 94 insertions, 14 deletions
diff --git a/lib/Target/R600/SIInstructions.td b/lib/Target/R600/SIInstructions.td index 76f05eb..2ca6a95 100644 --- a/lib/Target/R600/SIInstructions.td +++ b/lib/Target/R600/SIInstructions.td @@ -22,6 +22,8 @@ def InterpSlot : Operand<i32> { let PrintMethod = "printInterpSlot"; } +def SendMsgImm : Operand<i32>; + def isSI : Predicate<"Subtarget.getGeneration() " ">= AMDGPUSubtarget::SOUTHERN_ISLANDS">; @@ -826,17 +828,25 @@ def S_BARRIER : SOPP <0x0000000a, (ins), "S_BARRIER", def S_WAITCNT : SOPP <0x0000000c, (ins WAIT_FLAG:$simm16), "S_WAITCNT $simm16", [] >; -} // End hasSideEffects //def S_SETHALT : SOPP_ <0x0000000d, "S_SETHALT", []>; //def S_SLEEP : SOPP_ <0x0000000e, "S_SLEEP", []>; //def S_SETPRIO : SOPP_ <0x0000000f, "S_SETPRIO", []>; -//def S_SENDMSG : SOPP_ <0x00000010, "S_SENDMSG", []>; + +let Uses = [EXEC] in { + def S_SENDMSG : SOPP <0x00000010, (ins SendMsgImm:$simm16, M0Reg:$m0), "S_SENDMSG $simm16", + [(int_SI_sendmsg imm:$simm16, M0Reg:$m0)] + > { + let DisableEncoding = "$m0"; + } +} // End Uses = [EXEC] + //def S_SENDMSGHALT : SOPP_ <0x00000011, "S_SENDMSGHALT", []>; //def S_TRAP : SOPP_ <0x00000012, "S_TRAP", []>; //def S_ICACHE_INV : SOPP_ <0x00000013, "S_ICACHE_INV", []>; //def S_INCPERFLEVEL : SOPP_ <0x00000014, "S_INCPERFLEVEL", []>; //def S_DECPERFLEVEL : SOPP_ <0x00000015, "S_DECPERFLEVEL", []>; //def S_TTRACEDATA : SOPP_ <0x00000016, "S_TTRACEDATA", []>; +} // End hasSideEffects def V_CNDMASK_B32_e32 : VOP2 <0x00000000, (outs VReg_32:$dst), (ins VSrc_32:$src0, VReg_32:$src1, VCCReg:$vcc), @@ -979,14 +989,16 @@ defm V_MBCNT_HI_U32_B32 : VOP2_32 <0x00000024, "V_MBCNT_HI_U32_B32", []>; let isCommutable = 1, Defs = [VCC] in { // Carry-out goes to VCC // No patterns so that the scalar instructions are always selected. // The scalar versions will be replaced with vector when needed later. -defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32", []>; -defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32", []>; -defm V_SUBREV_I32 : VOP2b_32 <0x00000027, "V_SUBREV_I32", [], "V_SUB_I32">; +defm V_ADD_I32 : VOP2b_32 <0x00000025, "V_ADD_I32", [], VSrc_32>; +defm V_SUB_I32 : VOP2b_32 <0x00000026, "V_SUB_I32", [], VSrc_32>; +defm V_SUBREV_I32 : VOP2b_32 <0x00000027, "V_SUBREV_I32", [], VSrc_32, + "V_SUB_I32">; let Uses = [VCC] in { // Carry-in comes from VCC -defm V_ADDC_U32 : VOP2b_32 <0x00000028, "V_ADDC_U32", []>; -defm V_SUBB_U32 : VOP2b_32 <0x00000029, "V_SUBB_U32", []>; -defm V_SUBBREV_U32 : VOP2b_32 <0x0000002a, "V_SUBBREV_U32", [], "V_SUBB_U32">; +defm V_ADDC_U32 : VOP2b_32 <0x00000028, "V_ADDC_U32", [], VReg_32>; +defm V_SUBB_U32 : VOP2b_32 <0x00000029, "V_SUBB_U32", [], VReg_32>; +defm V_SUBBREV_U32 : VOP2b_32 <0x0000002a, "V_SUBBREV_U32", [], VReg_32, + "V_SUBB_U32">; } // End Uses = [VCC] } // End isCommutable = 1, Defs = [VCC] @@ -1403,7 +1415,7 @@ def : Pat < /* int_SI_vs_load_input */ def : Pat< (SIload_input i128:$tlst, IMM12bit:$attr_offset, i32:$buf_idx_vgpr), - (BUFFER_LOAD_FORMAT_XYZW_IDXEN $tlst, $buf_idx_vgpr, imm:$attr_offset) + (BUFFER_LOAD_FORMAT_XYZW_IDXEN $tlst, $buf_idx_vgpr, imm:$attr_offset, 0, 0, 0, 0) >; /* int_SI_export */ @@ -1658,16 +1670,30 @@ def : Pat < 0 /* ABS */, 1 /* CLAMP */, 0 /* OMOD */, 0 /* NEG */) >; +/********** ================================ **********/ +/********** Floating point absolute/negative **********/ +/********** ================================ **********/ + +// Manipulate the sign bit directly, as e.g. using the source negation modifier +// in V_ADD_F32_e64 $src, 0, [...] does not result in -0.0 for $src == +0.0, +// breaking the piglit *s-floatBitsToInt-neg* tests + +// TODO: Look into not implementing isFNegFree/isFAbsFree for SI, and possibly +// removing these patterns + +def : Pat < + (fneg (fabs f32:$src)), + (V_OR_B32_e32 $src, (V_MOV_B32_e32 0x80000000)) /* Set sign bit */ +>; + def : Pat < (fabs f32:$src), - (V_ADD_F32_e64 $src, (i32 0 /* SRC1 */), - 1 /* ABS */, 0 /* CLAMP */, 0 /* OMOD */, 0 /* NEG */) + (V_AND_B32_e32 $src, (V_MOV_B32_e32 0x7fffffff)) /* Clear sign bit */ >; def : Pat < (fneg f32:$src), - (V_ADD_F32_e64 $src, (i32 0 /* SRC1 */), - 0 /* ABS */, 0 /* CLAMP */, 0 /* OMOD */, 1 /* NEG */) + (V_XOR_B32_e32 $src, (V_MOV_B32_e32 0x80000000)) /* Toggle sign bit */ >; /********** ================== **********/ @@ -1794,6 +1820,11 @@ def : Pat < (V_CNDMASK_B32_e64 (i32 0), (i32 -1), $src0) >; +def : Pat < + (i32 (zext i1:$src0)), + (V_CNDMASK_B32_e64 (i32 0), (i32 1), $src0) +>; + // 1. Offset as 8bit DWORD immediate def : Pat < (SIload_constant i128:$sbase, IMM8bitDWORD:$offset), @@ -1809,7 +1840,7 @@ def : Pat < // 3. Offset in an 32Bit VGPR def : Pat < (SIload_constant i128:$sbase, i32:$voff), - (BUFFER_LOAD_DWORD_OFFEN $sbase, $voff) + (BUFFER_LOAD_DWORD_OFFEN $sbase, $voff, 0, 0, 0, 0) >; // The multiplication scales from [0,1] to the unsigned integer range @@ -1970,6 +2001,50 @@ defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, i64, global_store>; defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX2, v2i32, global_store>; defm : MUBUFStore_Pattern <BUFFER_STORE_DWORDX4, v4i32, global_store>; +// BUFFER_LOAD_DWORD*, addr64=0 +multiclass MUBUF_Load_Dword <ValueType vt, MUBUF offset, MUBUF offen, MUBUF idxen, + MUBUF bothen> { + + def : Pat < + (vt (int_SI_buffer_load_dword i128:$rsrc, i32:$vaddr, i32:$soffset, + imm:$offset, 0, 0, imm:$glc, imm:$slc, + imm:$tfe)), + (offset $rsrc, $vaddr, (as_i16imm $offset), $soffset, (as_i1imm $glc), + (as_i1imm $slc), (as_i1imm $tfe)) + >; + + def : Pat < + (vt (int_SI_buffer_load_dword i128:$rsrc, i32:$vaddr, i32:$soffset, + imm, 1, 0, imm:$glc, imm:$slc, + imm:$tfe)), + (offen $rsrc, $vaddr, $soffset, (as_i1imm $glc), (as_i1imm $slc), + (as_i1imm $tfe)) + >; + + def : Pat < + (vt (int_SI_buffer_load_dword i128:$rsrc, i32:$vaddr, i32:$soffset, + imm:$offset, 0, 1, imm:$glc, imm:$slc, + imm:$tfe)), + (idxen $rsrc, $vaddr, (as_i16imm $offset), $soffset, (as_i1imm $glc), + (as_i1imm $slc), (as_i1imm $tfe)) + >; + + def : Pat < + (vt (int_SI_buffer_load_dword i128:$rsrc, v2i32:$vaddr, i32:$soffset, + imm, 1, 1, imm:$glc, imm:$slc, + imm:$tfe)), + (bothen $rsrc, $vaddr, $soffset, (as_i1imm $glc), (as_i1imm $slc), + (as_i1imm $tfe)) + >; +} + +defm : MUBUF_Load_Dword <i32, BUFFER_LOAD_DWORD_OFFSET, BUFFER_LOAD_DWORD_OFFEN, + BUFFER_LOAD_DWORD_IDXEN, BUFFER_LOAD_DWORD_BOTHEN>; +defm : MUBUF_Load_Dword <v2i32, BUFFER_LOAD_DWORDX2_OFFSET, BUFFER_LOAD_DWORDX2_OFFEN, + BUFFER_LOAD_DWORDX2_IDXEN, BUFFER_LOAD_DWORDX2_BOTHEN>; +defm : MUBUF_Load_Dword <v4i32, BUFFER_LOAD_DWORDX4_OFFSET, BUFFER_LOAD_DWORDX4_OFFEN, + BUFFER_LOAD_DWORDX4_IDXEN, BUFFER_LOAD_DWORDX4_BOTHEN>; + //===----------------------------------------------------------------------===// // MTBUF Patterns //===----------------------------------------------------------------------===// @@ -2057,6 +2132,11 @@ def : Pat < (EXTRACT_SUBREG $a, sub0) >; +def : Pat < + (i1 (trunc i32:$a)), + (V_CMP_EQ_I32_e64 (V_AND_B32_e32 (i32 1), $a), 1) +>; + // V_ADD_I32_e32/S_ADD_I32 produces carry in VCC/SCC. For the vector // case, the sgpr-copies pass will fix this to use the vector version. def : Pat < |
