X-Git-Url: http://demsky.eecs.uci.edu/git/?a=blobdiff_plain;f=lib%2FTarget%2FR600%2FR600InstrFormats.td;h=ae3046d62e1e3d3a436017f30f25a818f959223e;hb=d08a9303614355cfdcac5f2c27c09ce809565423;hp=7cc48f454c974eac0685c879aaeb5830c847035a;hpb=2def95fc1e5537e5c156bd12027d41212e0b2fc2;p=oota-llvm.git

diff --git a/lib/Target/R600/R600InstrFormats.td b/lib/Target/R600/R600InstrFormats.td
index 7cc48f454c9..ae3046d62e1 100644
--- a/lib/Target/R600/R600InstrFormats.td
+++ b/lib/Target/R600/R600InstrFormats.td
@@ -23,9 +23,14 @@ class InstR600 <dag outs, dag ins, string asm, list<dag> pattern,
   bits<2> FlagOperandIdx = 0;
   bit Op1 = 0;
   bit Op2 = 0;
+  bit LDS_1A = 0;
+  bit LDS_1A1D = 0;
   bit HasNativeOperands = 0;
   bit VTXInst = 0;
   bit TEXInst = 0;
+  bit ALUInst = 0;
+  bit IsExport = 0;
+  bit LDS_1A2D = 0;
 
   let Namespace = "AMDGPU";
   let OutOperandList = outs;
@@ -47,21 +52,24 @@ class InstR600 <dag outs, dag ins, string asm, list<dag> pattern,
   let TSFlags{11} = Op2;
   let TSFlags{12} = VTXInst;
   let TSFlags{13} = TEXInst;
+  let TSFlags{14} = ALUInst;
+  let TSFlags{15} = LDS_1A;
+  let TSFlags{16} = LDS_1A1D;
+  let TSFlags{17} = IsExport;
+  let TSFlags{18} = LDS_1A2D;
 }
 
 //===----------------------------------------------------------------------===//
 // ALU instructions
 //===----------------------------------------------------------------------===//
 
-class R600ALU_Word0 {
+class R600_ALU_LDS_Word0 {
   field bits<32> Word0;
 
   bits<11> src0;
-  bits<1>  src0_neg;
   bits<1>  src0_rel;
   bits<11> src1;
   bits<1>  src1_rel;
-  bits<1>  src1_neg;
   bits<3>  index_mode = 0;
   bits<2>  pred_sel;
   bits<1>  last;
@@ -74,16 +82,23 @@ class R600ALU_Word0 {
   let Word0{8-0}   = src0_sel;
   let Word0{9}     = src0_rel;
   let Word0{11-10} = src0_chan;
-  let Word0{12}    = src0_neg;
   let Word0{21-13} = src1_sel;
   let Word0{22}    = src1_rel;
   let Word0{24-23} = src1_chan;
-  let Word0{25}    = src1_neg;
   let Word0{28-26} = index_mode;
   let Word0{30-29} = pred_sel;
   let Word0{31}    = last;
 }
 
+class R600ALU_Word0 : R600_ALU_LDS_Word0 {
+
+  bits<1>  src0_neg;
+  bits<1>  src1_neg;
+
+  let Word0{12}    = src0_neg;
+  let Word0{25}    = src1_neg;
+}
+
 class R600ALU_Word1 {
   field bits<32> Word1;
 
@@ -136,6 +151,30 @@ class R600ALU_Word1_OP3 <bits<5> alu_inst> : R600ALU_Word1{
   let Word1{17-13} = alu_inst;
 }
 
+class R600LDS_Word1 {
+  field bits<32> Word1;
+
+  bits<11> src2;
+  bits<9>  src2_sel  = src2{8-0};
+  bits<2>  src2_chan = src2{10-9};
+  bits<1>  src2_rel;
+  // offset specifies the stride offset to the second set of data to be read
+  // from.  This is a dword offset.
+  bits<5>  alu_inst = 17; // OP3_INST_LDS_IDX_OP
+  bits<3>  bank_swizzle;
+  bits<6>  lds_op;
+  bits<2>  dst_chan = 0;
+
+  let Word1{8-0}   = src2_sel;
+  let Word1{9}     = src2_rel;
+  let Word1{11-10} = src2_chan;
+  let Word1{17-13} = alu_inst;
+  let Word1{20-18} = bank_swizzle;
+  let Word1{26-21} = lds_op;
+  let Word1{30-29} = dst_chan;
+}
+
+
 /*
 XXX: R600 subtarget uses a slightly different encoding than the other
 subtargets.  We currently handle this in R600MCCodeEmitter, but we may
@@ -166,28 +205,46 @@ class R600ALU_Word1_OP2_r700 : R600ALU_Word1_OP2 {
 
 class VTX_WORD0 {
   field bits<32> Word0;
-  bits<7> SRC_GPR;
+  bits<7> src_gpr;
   bits<5> VC_INST;
   bits<2> FETCH_TYPE;
   bits<1> FETCH_WHOLE_QUAD;
   bits<8> BUFFER_ID;
   bits<1> SRC_REL;
   bits<2> SRC_SEL_X;
-  bits<6> MEGA_FETCH_COUNT;
 
   let Word0{4-0}   = VC_INST;
   let Word0{6-5}   = FETCH_TYPE;
   let Word0{7}     = FETCH_WHOLE_QUAD;
   let Word0{15-8}  = BUFFER_ID;
-  let Word0{22-16} = SRC_GPR;
+  let Word0{22-16} = src_gpr;
   let Word0{23}    = SRC_REL;
   let Word0{25-24} = SRC_SEL_X;
+}
+
+class VTX_WORD0_eg : VTX_WORD0 {
+
+  bits<6> MEGA_FETCH_COUNT;
+
   let Word0{31-26} = MEGA_FETCH_COUNT;
 }
 
+class VTX_WORD0_cm : VTX_WORD0 {
+
+  bits<2> SRC_SEL_Y;
+  bits<2> STRUCTURED_READ;
+  bits<1> LDS_REQ;
+  bits<1> COALESCED_READ;
+
+  let Word0{27-26} = SRC_SEL_Y;
+  let Word0{29-28} = STRUCTURED_READ;
+  let Word0{30}    = LDS_REQ;
+  let Word0{31}    = COALESCED_READ;
+}
+
 class VTX_WORD1_GPR {
   field bits<32> Word1;
-  bits<7> DST_GPR;
+  bits<7> dst_gpr;
   bits<1> DST_REL;
   bits<3> DST_SEL_X;
   bits<3> DST_SEL_Y;
@@ -199,7 +256,7 @@ class VTX_WORD1_GPR {
   bits<1> FORMAT_COMP_ALL;
   bits<1> SRF_MODE_ALL;
 
-  let Word1{6-0} = DST_GPR;
+  let Word1{6-0} = dst_gpr;
   let Word1{7}    = DST_REL;
   let Word1{8}    = 0; // Reserved
   let Word1{11-9} = DST_SEL_X;
@@ -390,3 +447,48 @@ class CF_ALU_WORD1 {
   let Word1{30} = WHOLE_QUAD_MODE;
   let Word1{31} = BARRIER;
 }
+
+class CF_ALLOC_EXPORT_WORD0_RAT {
+  field bits<32> Word0;
+
+  bits<4> rat_id;
+  bits<6> rat_inst;
+  bits<2> rim;
+  bits<2> type;
+  bits<7> rw_gpr;
+  bits<1> rw_rel;
+  bits<7> index_gpr;
+  bits<2> elem_size;
+
+  let Word0{3-0}   = rat_id;
+  let Word0{9-4}   = rat_inst;
+  let Word0{10}    = 0; // Reserved
+  let Word0{12-11} = rim;
+  let Word0{14-13} = type;
+  let Word0{21-15} = rw_gpr;
+  let Word0{22}    = rw_rel;
+  let Word0{29-23} = index_gpr;
+  let Word0{31-30} = elem_size;
+}
+
+class CF_ALLOC_EXPORT_WORD1_BUF {
+  field bits<32> Word1;
+
+  bits<12> array_size;
+  bits<4>  comp_mask;
+  bits<4>  burst_count;
+  bits<1>  vpm;
+  bits<1>  eop;
+  bits<8>  cf_inst;
+  bits<1>  mark;
+  bits<1>  barrier;
+
+  let Word1{11-0} = array_size;
+  let Word1{15-12} = comp_mask;
+  let Word1{19-16} = burst_count;
+  let Word1{20}    = vpm;
+  let Word1{21}    = eop;
+  let Word1{29-22} = cf_inst;
+  let Word1{30}    = mark;
+  let Word1{31}    = barrier;
+}