diff options
author | Alexander Kornienko <alexfh@google.com> | 2013-04-03 14:07:16 +0000 |
---|---|---|
committer | Alexander Kornienko <alexfh@google.com> | 2013-04-03 14:07:16 +0000 |
commit | e133bc868944822bf8961f825d3aa63d6fa48fb7 (patch) | |
tree | ebbd4a8040181471467a9737d90d94dc6b58b316 /lib/Target/R600/R600Instructions.td | |
parent | 647735c781c5b37061ee03d6e9e6c7dda92218e2 (diff) | |
parent | 080e3c523e87ec68ca1ea5db4cd49816028dd8bd (diff) |
Updating branches/google/stable to r178511stable
git-svn-id: https://llvm.org/svn/llvm-project/llvm/branches/google/stable@178655 91177308-0d34-0410-b5e6-96231b3b80d8
Diffstat (limited to 'lib/Target/R600/R600Instructions.td')
-rw-r--r-- | lib/Target/R600/R600Instructions.td | 330 |
1 files changed, 301 insertions, 29 deletions
diff --git a/lib/Target/R600/R600Instructions.td b/lib/Target/R600/R600Instructions.td index c5fa3347dc..663b41a66d 100644 --- a/lib/Target/R600/R600Instructions.td +++ b/lib/Target/R600/R600Instructions.td @@ -234,6 +234,80 @@ class VTX_WORD1_GPR { let Word1{31} = SRF_MODE_ALL; } +class TEX_WORD0 { + field bits<32> Word0; + + bits<5> TEX_INST; + bits<2> INST_MOD; + bits<1> FETCH_WHOLE_QUAD; + bits<8> RESOURCE_ID; + bits<7> SRC_GPR; + bits<1> SRC_REL; + bits<1> ALT_CONST; + bits<2> RESOURCE_INDEX_MODE; + bits<2> SAMPLER_INDEX_MODE; + + let Word0{4-0} = TEX_INST; + let Word0{6-5} = INST_MOD; + let Word0{7} = FETCH_WHOLE_QUAD; + let Word0{15-8} = RESOURCE_ID; + let Word0{22-16} = SRC_GPR; + let Word0{23} = SRC_REL; + let Word0{24} = ALT_CONST; + let Word0{26-25} = RESOURCE_INDEX_MODE; + let Word0{28-27} = SAMPLER_INDEX_MODE; +} + +class TEX_WORD1 { + field bits<32> Word1; + + bits<7> DST_GPR; + bits<1> DST_REL; + bits<3> DST_SEL_X; + bits<3> DST_SEL_Y; + bits<3> DST_SEL_Z; + bits<3> DST_SEL_W; + bits<7> LOD_BIAS; + bits<1> COORD_TYPE_X; + bits<1> COORD_TYPE_Y; + bits<1> COORD_TYPE_Z; + bits<1> COORD_TYPE_W; + + let Word1{6-0} = DST_GPR; + let Word1{7} = DST_REL; + let Word1{11-9} = DST_SEL_X; + let Word1{14-12} = DST_SEL_Y; + let Word1{17-15} = DST_SEL_Z; + let Word1{20-18} = DST_SEL_W; + let Word1{27-21} = LOD_BIAS; + let Word1{28} = COORD_TYPE_X; + let Word1{29} = COORD_TYPE_Y; + let Word1{30} = COORD_TYPE_Z; + let Word1{31} = COORD_TYPE_W; +} + +class TEX_WORD2 { + field bits<32> Word2; + + bits<5> OFFSET_X; + bits<5> OFFSET_Y; + bits<5> OFFSET_Z; + bits<5> SAMPLER_ID; + bits<3> SRC_SEL_X; + bits<3> SRC_SEL_Y; + bits<3> SRC_SEL_Z; + bits<3> SRC_SEL_W; + + let Word2{4-0} = OFFSET_X; + let Word2{9-5} = OFFSET_Y; + let Word2{14-10} = OFFSET_Z; + let Word2{19-15} = SAMPLER_ID; + let Word2{22-20} = SRC_SEL_X; + let Word2{25-23} = SRC_SEL_Y; + let Word2{28-26} = SRC_SEL_Z; + let Word2{31-29} = SRC_SEL_W; +} + /* XXX: R600 subtarget uses a slightly different encoding than the other subtargets. We currently handle this in R600MCCodeEmitter, but we may @@ -277,9 +351,9 @@ class R600_1OP <bits<11> inst, string opName, list<dag> pattern, (ins WRITE:$write, OMOD:$omod, REL:$dst_rel, CLAMP:$clamp, R600_Reg32:$src0, NEG:$src0_neg, REL:$src0_rel, ABS:$src0_abs, SEL:$src0_sel, LAST:$last, R600_Pred:$pred_sel, LITERAL:$literal), - !strconcat(opName, + !strconcat(" ", opName, "$clamp $dst$write$dst_rel$omod, " - "$src0_neg$src0_abs$src0$src0_sel$src0_abs$src0_rel, " + "$src0_neg$src0_abs$src0$src0_abs$src0_rel, " "$literal $pred_sel$last"), pattern, itin>, @@ -318,10 +392,10 @@ class R600_2OP <bits<11> inst, string opName, list<dag> pattern, R600_Reg32:$src0, NEG:$src0_neg, REL:$src0_rel, ABS:$src0_abs, SEL:$src0_sel, R600_Reg32:$src1, NEG:$src1_neg, REL:$src1_rel, ABS:$src1_abs, SEL:$src1_sel, LAST:$last, R600_Pred:$pred_sel, LITERAL:$literal), - !strconcat(opName, + !strconcat(" ", opName, "$clamp $update_exec_mask$update_pred$dst$write$dst_rel$omod, " - "$src0_neg$src0_abs$src0$src0_sel$src0_abs$src0_rel, " - "$src1_neg$src1_abs$src1$src1_sel$src1_abs$src1_rel, " + "$src0_neg$src0_abs$src0$src0_abs$src0_rel, " + "$src1_neg$src1_abs$src1$src1_abs$src1_rel, " "$literal $pred_sel$last"), pattern, itin>, @@ -356,10 +430,10 @@ class R600_3OP <bits<5> inst, string opName, list<dag> pattern, R600_Reg32:$src1, NEG:$src1_neg, REL:$src1_rel, SEL:$src1_sel, R600_Reg32:$src2, NEG:$src2_neg, REL:$src2_rel, SEL:$src2_sel, LAST:$last, R600_Pred:$pred_sel, LITERAL:$literal), - !strconcat(opName, "$clamp $dst$dst_rel, " - "$src0_neg$src0$src0_sel$src0_rel, " - "$src1_neg$src1$src1_sel$src1_rel, " - "$src2_neg$src2$src2_sel$src2_rel, " + !strconcat(" ", opName, "$clamp $dst$dst_rel, " + "$src0_neg$src0$src0_rel, " + "$src1_neg$src1$src1_rel, " + "$src2_neg$src2$src2_rel, " "$literal $pred_sel$last"), pattern, itin>, @@ -386,12 +460,32 @@ class R600_REDUCTION <bits<11> inst, dag ins, string asm, list<dag> pattern, class R600_TEX <bits<11> inst, string opName, list<dag> pattern, InstrItinClass itin = AnyALU> : InstR600 <inst, - (outs R600_Reg128:$dst), - (ins R600_Reg128:$src0, i32imm:$resourceId, i32imm:$samplerId, i32imm:$textureTarget), - !strconcat(opName, "$dst, $src0, $resourceId, $samplerId, $textureTarget"), + (outs R600_Reg128:$DST_GPR), + (ins R600_Reg128:$SRC_GPR, i32imm:$RESOURCE_ID, i32imm:$SAMPLER_ID, i32imm:$textureTarget), + !strconcat(opName, "$DST_GPR, $SRC_GPR, $RESOURCE_ID, $SAMPLER_ID, $textureTarget"), pattern, - itin>{ - let Inst {10-0} = inst; + itin>, TEX_WORD0, TEX_WORD1, TEX_WORD2 { + let Inst{31-0} = Word0; + let Inst{63-32} = Word1; + + let TEX_INST = inst{4-0}; + let SRC_REL = 0; + let DST_REL = 0; + let DST_SEL_X = 0; + let DST_SEL_Y = 1; + let DST_SEL_Z = 2; + let DST_SEL_W = 3; + let LOD_BIAS = 0; + + let INST_MOD = 0; + let FETCH_WHOLE_QUAD = 0; + let ALT_CONST = 0; + let SAMPLER_INDEX_MODE = 0; + + let COORD_TYPE_X = 0; + let COORD_TYPE_Y = 0; + let COORD_TYPE_Z = 0; + let COORD_TYPE_W = 0; } } // End mayLoad = 1, mayStore = 0, hasSideEffects = 0 @@ -671,6 +765,167 @@ class ExportBufInst : InstR600ISA<( let Inst{63-32} = Word1; } +//===----------------------------------------------------------------------===// +// Control Flow Instructions +//===----------------------------------------------------------------------===// + +class CF_ALU_WORD0 { + field bits<32> Word0; + + bits<22> ADDR; + bits<4> KCACHE_BANK0; + bits<4> KCACHE_BANK1; + bits<2> KCACHE_MODE0; + + let Word0{21-0} = ADDR; + let Word0{25-22} = KCACHE_BANK0; + let Word0{29-26} = KCACHE_BANK1; + let Word0{31-30} = KCACHE_MODE0; +} + +class CF_ALU_WORD1 { + field bits<32> Word1; + + bits<2> KCACHE_MODE1; + bits<8> KCACHE_ADDR0; + bits<8> KCACHE_ADDR1; + bits<7> COUNT; + bits<1> ALT_CONST; + bits<4> CF_INST; + bits<1> WHOLE_QUAD_MODE; + bits<1> BARRIER; + + let Word1{1-0} = KCACHE_MODE1; + let Word1{9-2} = KCACHE_ADDR0; + let Word1{17-10} = KCACHE_ADDR1; + let Word1{24-18} = COUNT; + let Word1{25} = ALT_CONST; + let Word1{29-26} = CF_INST; + let Word1{30} = WHOLE_QUAD_MODE; + let Word1{31} = BARRIER; +} + +class ALU_CLAUSE<bits<4> inst, string OpName> : AMDGPUInst <(outs), +(ins i32imm:$ADDR, i32imm:$KCACHE_BANK0, i32imm:$KCACHE_BANK1, i32imm:$KCACHE_MODE0, i32imm:$KCACHE_MODE1, +i32imm:$KCACHE_ADDR0, i32imm:$KCACHE_ADDR1, i32imm:$COUNT), +!strconcat(OpName, " $COUNT, @$ADDR, " +"KC0[CB$KCACHE_BANK0:$KCACHE_ADDR0-$KCACHE_ADDR0+32]" +", KC1[CB$KCACHE_BANK1:$KCACHE_ADDR1-$KCACHE_ADDR1+32]"), +[] >, CF_ALU_WORD0, CF_ALU_WORD1 { + field bits<64> Inst; + + let CF_INST = inst; + let ALT_CONST = 0; + let WHOLE_QUAD_MODE = 0; + let BARRIER = 1; + + let Inst{31-0} = Word0; + let Inst{63-32} = Word1; +} + +class CF_WORD0 { + field bits<32> Word0; + + bits<24> ADDR; + bits<3> JUMPTABLE_SEL; + + let Word0{23-0} = ADDR; + let Word0{26-24} = JUMPTABLE_SEL; +} + +class CF_WORD1 { + field bits<32> Word1; + + bits<3> POP_COUNT; + bits<5> CF_CONST; + bits<2> COND; + bits<6> COUNT; + bits<1> VALID_PIXEL_MODE; + bits<8> CF_INST; + bits<1> BARRIER; + + let Word1{2-0} = POP_COUNT; + let Word1{7-3} = CF_CONST; + let Word1{9-8} = COND; + let Word1{15-10} = COUNT; + let Word1{20} = VALID_PIXEL_MODE; + let Word1{29-22} = CF_INST; + let Word1{31} = BARRIER; +} + +class CF_CLAUSE <bits<8> inst, dag ins, string AsmPrint> : AMDGPUInst <(outs), +ins, AsmPrint, [] >, CF_WORD0, CF_WORD1 { + field bits<64> Inst; + + let CF_INST = inst; + let BARRIER = 1; + let JUMPTABLE_SEL = 0; + let CF_CONST = 0; + let VALID_PIXEL_MODE = 0; + let COND = 0; + + let Inst{31-0} = Word0; + let Inst{63-32} = Word1; +} + +def CF_TC : CF_CLAUSE<1, (ins i32imm:$ADDR, i32imm:$COUNT), +"TEX $COUNT @$ADDR"> { + let POP_COUNT = 0; +} + +def CF_VC : CF_CLAUSE<2, (ins i32imm:$ADDR, i32imm:$COUNT), +"VTX $COUNT @$ADDR"> { + let POP_COUNT = 0; +} + +def WHILE_LOOP : CF_CLAUSE<6, (ins i32imm:$ADDR), "LOOP_START_DX10 @$ADDR"> { + let POP_COUNT = 0; + let COUNT = 0; +} + +def END_LOOP : CF_CLAUSE<5, (ins i32imm:$ADDR), "END_LOOP @$ADDR"> { + let POP_COUNT = 0; + let COUNT = 0; +} + +def LOOP_BREAK : CF_CLAUSE<9, (ins i32imm:$ADDR), "LOOP_BREAK @$ADDR"> { + let POP_COUNT = 0; + let COUNT = 0; +} + +def CF_CONTINUE : CF_CLAUSE<8, (ins i32imm:$ADDR), "CONTINUE @$ADDR"> { + let POP_COUNT = 0; + let COUNT = 0; +} + +def CF_JUMP : CF_CLAUSE<10, (ins i32imm:$ADDR, i32imm:$POP_COUNT), "JUMP @$ADDR POP:$POP_COUNT"> { + let COUNT = 0; +} + +def CF_ELSE : CF_CLAUSE<13, (ins i32imm:$ADDR, i32imm:$POP_COUNT), "ELSE @$ADDR POP:$POP_COUNT"> { + let COUNT = 0; +} + +def CF_CALL_FS : CF_CLAUSE<19, (ins), "CALL_FS"> { + let ADDR = 0; + let COUNT = 0; + let POP_COUNT = 0; +} + +def POP : CF_CLAUSE<14, (ins i32imm:$ADDR, i32imm:$POP_COUNT), "POP @$ADDR POP:$POP_COUNT"> { + let COUNT = 0; +} + +def CF_ALU : ALU_CLAUSE<8, "ALU">; +def CF_ALU_PUSH_BEFORE : ALU_CLAUSE<9, "ALU_PUSH_BEFORE">; + +def STACK_SIZE : AMDGPUInst <(outs), +(ins i32imm:$num), "nstack $num", [] > { + field bits<8> Inst; + bits<8> num; + let Inst = num; +} + let Predicates = [isR600toCayman] in { //===----------------------------------------------------------------------===// @@ -867,25 +1122,33 @@ def CNDGT_INT : R600_3OP < def TEX_LD : R600_TEX < 0x03, "TEX_LD", - [(set R600_Reg128:$dst, (int_AMDGPU_txf R600_Reg128:$src0, imm:$src1, imm:$src2, imm:$src3, imm:$resourceId, imm:$samplerId, imm:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txf R600_Reg128:$SRC_GPR, + imm:$OFFSET_X, imm:$OFFSET_Y, imm:$OFFSET_Z, imm:$RESOURCE_ID, + imm:$SAMPLER_ID, imm:$textureTarget))] > { -let AsmString = "TEX_LD $dst, $src0, $src1, $src2, $src3, $resourceId, $samplerId, $textureTarget"; -let InOperandList = (ins R600_Reg128:$src0, i32imm:$src1, i32imm:$src2, i32imm:$src3, i32imm:$resourceId, i32imm:$samplerId, i32imm:$textureTarget); +let AsmString = "TEX_LD $DST_GPR, $SRC_GPR, $OFFSET_X, $OFFSET_Y, $OFFSET_Z," + "$RESOURCE_ID, $SAMPLER_ID, $textureTarget"; +let InOperandList = (ins R600_Reg128:$SRC_GPR, i32imm:$OFFSET_X, + i32imm:$OFFSET_Y, i32imm:$OFFSET_Z, i32imm:$RESOURCE_ID, i32imm:$SAMPLER_ID, + i32imm:$textureTarget); } def TEX_GET_TEXTURE_RESINFO : R600_TEX < 0x04, "TEX_GET_TEXTURE_RESINFO", - [(set R600_Reg128:$dst, (int_AMDGPU_txq R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, imm:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txq R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))] >; def TEX_GET_GRADIENTS_H : R600_TEX < 0x07, "TEX_GET_GRADIENTS_H", - [(set R600_Reg128:$dst, (int_AMDGPU_ddx R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, imm:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_ddx R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))] >; def TEX_GET_GRADIENTS_V : R600_TEX < 0x08, "TEX_GET_GRADIENTS_V", - [(set R600_Reg128:$dst, (int_AMDGPU_ddy R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, imm:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_ddy R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))] >; def TEX_SET_GRADIENTS_H : R600_TEX < @@ -900,32 +1163,38 @@ def TEX_SET_GRADIENTS_V : R600_TEX < def TEX_SAMPLE : R600_TEX < 0x10, "TEX_SAMPLE", - [(set R600_Reg128:$dst, (int_AMDGPU_tex R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, imm:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_tex R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))] >; def TEX_SAMPLE_C : R600_TEX < 0x18, "TEX_SAMPLE_C", - [(set R600_Reg128:$dst, (int_AMDGPU_tex R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, TEX_SHADOW:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_tex R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, TEX_SHADOW:$textureTarget))] >; def TEX_SAMPLE_L : R600_TEX < 0x11, "TEX_SAMPLE_L", - [(set R600_Reg128:$dst, (int_AMDGPU_txl R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, imm:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txl R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))] >; def TEX_SAMPLE_C_L : R600_TEX < 0x19, "TEX_SAMPLE_C_L", - [(set R600_Reg128:$dst, (int_AMDGPU_txl R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, TEX_SHADOW:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txl R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, TEX_SHADOW:$textureTarget))] >; def TEX_SAMPLE_LB : R600_TEX < 0x12, "TEX_SAMPLE_LB", - [(set R600_Reg128:$dst, (int_AMDGPU_txb R600_Reg128:$src0,imm:$resourceId, imm:$samplerId, imm:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txb R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, imm:$textureTarget))] >; def TEX_SAMPLE_C_LB : R600_TEX < 0x1A, "TEX_SAMPLE_C_LB", - [(set R600_Reg128:$dst, (int_AMDGPU_txb R600_Reg128:$src0, imm:$resourceId, imm:$samplerId, TEX_SHADOW:$textureTarget))] + [(set R600_Reg128:$DST_GPR, (int_AMDGPU_txb R600_Reg128:$SRC_GPR, + imm:$RESOURCE_ID, imm:$SAMPLER_ID, TEX_SHADOW:$textureTarget))] >; def TEX_SAMPLE_G : R600_TEX < @@ -1141,6 +1410,7 @@ let Predicates = [isR600] in { def RECIP_UINT_r600 : RECIP_UINT_Common <0x78>; defm DIV_r600 : DIV_Common<RECIP_IEEE_r600>; + def : POW_Common <LOG_IEEE_r600, EXP_IEEE_r600, MUL, R600_Reg32>; def TGSI_LIT_Z_r600 : TGSI_LIT_Z_Common<MUL_LIT_r600, LOG_CLAMPED_r600, EXP_IEEE_r600>; def : Pat<(fsqrt R600_Reg32:$src), @@ -1212,6 +1482,7 @@ def RECIPSQRT_IEEE_eg : RECIPSQRT_IEEE_Common<0x89>; def SIN_eg : SIN_Common<0x8D>; def COS_eg : COS_Common<0x8E>; +def : POW_Common <LOG_IEEE_eg, EXP_IEEE_eg, MUL, R600_Reg32>; def : SIN_PAT <SIN_eg>; def : COS_PAT <COS_eg>; def : Pat<(fsqrt R600_Reg32:$src), @@ -1540,13 +1811,14 @@ def MULLO_UINT_cm : MULLO_UINT_Common<0x91>; def MULHI_UINT_cm : MULHI_UINT_Common<0x92>; def RECIPSQRT_CLAMPED_cm : RECIPSQRT_CLAMPED_Common<0x87>; def EXP_IEEE_cm : EXP_IEEE_Common<0x81>; -def LOG_IEEE_ : LOG_IEEE_Common<0x83>; +def LOG_IEEE_cm : LOG_IEEE_Common<0x83>; def RECIP_CLAMPED_cm : RECIP_CLAMPED_Common<0x84>; def RECIPSQRT_IEEE_cm : RECIPSQRT_IEEE_Common<0x89>; def SIN_cm : SIN_Common<0x8D>; def COS_cm : COS_Common<0x8E>; } // End isVector = 1 +def : POW_Common <LOG_IEEE_cm, EXP_IEEE_cm, MUL, R600_Reg32>; def : SIN_PAT <SIN_cm>; def : COS_PAT <COS_cm>; @@ -1979,8 +2251,8 @@ def : Insert_Element <i32, v4i32, R600_Reg32, R600_Reg128, 1, sub1>; def : Insert_Element <i32, v4i32, R600_Reg32, R600_Reg128, 2, sub2>; def : Insert_Element <i32, v4i32, R600_Reg32, R600_Reg128, 3, sub3>; -def : Vector_Build <v4f32, R600_Reg128, f32, R600_Reg32>; -def : Vector_Build <v4i32, R600_Reg128, i32, R600_Reg32>; +def : Vector4_Build <v4f32, R600_Reg128, f32, R600_Reg32>; +def : Vector4_Build <v4i32, R600_Reg128, i32, R600_Reg32>; // bitconvert patterns |