2019-05-02 00:32:58 +08:00
|
|
|
;RUN: llc < %s -march=amdgcn -mcpu=verde -verify-machineinstrs | FileCheck -check-prefixes=GCN,PREGFX10 %s
|
|
|
|
|
;RUN: llc < %s -march=amdgcn -mcpu=tonga -verify-machineinstrs | FileCheck -check-prefixes=GCN,PREGFX10 %s
|
|
|
|
|
;RUN: llc < %s -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs | FileCheck -check-prefixes=GCN,GFX10 %s
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}tbuffer_load:
|
|
|
|
|
; GCN: v_mov_b32_e32 [[ZEROREG:v[0-9]+]], 0
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:14, nfmt:4, 0 idxen
|
|
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:15, nfmt:3, 0 idxen glc
|
|
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:6, nfmt:1, 0 idxen slc
|
|
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:6, nfmt:1, 0 idxen glc
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:78, 0 idxen
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:63, 0 idxen glc
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:22, 0 idxen slc
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:22, 0 idxen glc dlc
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
; GCN: s_waitcnt
|
|
|
|
|
define amdgpu_vs {<4 x float>, <4 x float>, <4 x float>, <4 x float>} @tbuffer_load(<4 x i32> inreg) {
|
|
|
|
|
main_body:
|
|
|
|
|
%vdata = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 78, i32 0)
|
|
|
|
|
%vdata_glc = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 63, i32 1)
|
|
|
|
|
%vdata_slc = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 22, i32 2)
|
2019-05-02 00:32:58 +08:00
|
|
|
%vdata_f32 = call <4 x float> @llvm.amdgcn.struct.tbuffer.load.v4f32(<4 x i32> %0, i32 0, i32 0, i32 0, i32 22, i32 5)
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
%vdata.f = bitcast <4 x i32> %vdata to <4 x float>
|
|
|
|
|
%vdata_glc.f = bitcast <4 x i32> %vdata_glc to <4 x float>
|
|
|
|
|
%vdata_slc.f = bitcast <4 x i32> %vdata_slc to <4 x float>
|
|
|
|
|
%r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %vdata.f, 0
|
|
|
|
|
%r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %vdata_glc.f, 1
|
|
|
|
|
%r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %vdata_slc.f, 2
|
|
|
|
|
%r3 = insertvalue {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r2, <4 x float> %vdata_f32, 3
|
|
|
|
|
ret {<4 x float>, <4 x float>, <4 x float>, <4 x float>} %r3
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}tbuffer_load_immoffs:
|
|
|
|
|
; GCN: v_mov_b32_e32 [[ZEROREG:v[0-9]+]], 0
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:14, nfmt:4, 0 idxen offset:42
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:78, 0 idxen offset:42
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
define amdgpu_vs <4 x float> @tbuffer_load_immoffs(<4 x i32> inreg) {
|
|
|
|
|
main_body:
|
|
|
|
|
%vdata = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 42, i32 0, i32 78, i32 0)
|
|
|
|
|
%vdata.f = bitcast <4 x i32> %vdata to <4 x float>
|
|
|
|
|
ret <4 x float> %vdata.f
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}tbuffer_load_immoffs_large
|
|
|
|
|
; GCN: v_mov_b32_e32 [[ZEROREG:v[0-9]+]], 0
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:15, nfmt:2, 61 idxen offset:4095
|
|
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:14, nfmt:3, {{s[0-9]+}} idxen offset:73
|
|
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, dfmt:13, nfmt:4, {{s[0-9]+}} idxen offset:1
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:47, 61 idxen offset:4095
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:62, {{s[0-9]+}} idxen offset:73
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, [[ZEROREG]], {{s\[[0-9]+:[0-9]+\]}}, format:77, {{s[0-9]+}} idxen offset:1
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
; GCN: s_waitcnt
|
|
|
|
|
define amdgpu_vs {<4 x float>, <4 x float>, <4 x float>} @tbuffer_load_immoffs_large(<4 x i32> inreg, i32 inreg %soffs) {
|
|
|
|
|
%vdata = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 4095, i32 61, i32 47, i32 0)
|
|
|
|
|
%vdata_glc = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 73, i32 %soffs, i32 62, i32 0)
|
|
|
|
|
%vdata_slc = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 1, i32 %soffs, i32 77, i32 0)
|
|
|
|
|
%vdata.f = bitcast <4 x i32> %vdata to <4 x float>
|
|
|
|
|
%vdata_glc.f = bitcast <4 x i32> %vdata_glc to <4 x float>
|
|
|
|
|
%vdata_slc.f = bitcast <4 x i32> %vdata_slc to <4 x float>
|
|
|
|
|
%r0 = insertvalue {<4 x float>, <4 x float>, <4 x float>} undef, <4 x float> %vdata.f, 0
|
|
|
|
|
%r1 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r0, <4 x float> %vdata_glc.f, 1
|
|
|
|
|
%r2 = insertvalue {<4 x float>, <4 x float>, <4 x float>} %r1, <4 x float> %vdata_slc.f, 2
|
|
|
|
|
ret {<4 x float>, <4 x float>, <4 x float>} %r2
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}tbuffer_load_idx:
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, dfmt:14, nfmt:4, 0 idxen
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, format:78, 0 idxen
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
define amdgpu_vs <4 x float> @tbuffer_load_idx(<4 x i32> inreg, i32 %vindex) {
|
|
|
|
|
main_body:
|
|
|
|
|
%vdata = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 %vindex, i32 0, i32 0, i32 78, i32 0)
|
|
|
|
|
%vdata.f = bitcast <4 x i32> %vdata to <4 x float>
|
|
|
|
|
ret <4 x float> %vdata.f
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}tbuffer_load_ofs:
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, dfmt:14, nfmt:4, 0 idxen offen
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, format:78, 0 idxen offen
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
define amdgpu_vs <4 x float> @tbuffer_load_ofs(<4 x i32> inreg, i32 %voffs) {
|
|
|
|
|
main_body:
|
|
|
|
|
%vdata = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 %voffs, i32 0, i32 78, i32 0)
|
|
|
|
|
%vdata.f = bitcast <4 x i32> %vdata to <4 x float>
|
|
|
|
|
ret <4 x float> %vdata.f
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}tbuffer_load_ofs_imm:
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, dfmt:14, nfmt:4, 0 idxen offen offset:52
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, format:78, 0 idxen offen offset:52
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
define amdgpu_vs <4 x float> @tbuffer_load_ofs_imm(<4 x i32> inreg, i32 %voffs) {
|
|
|
|
|
main_body:
|
|
|
|
|
%ofs = add i32 %voffs, 52
|
|
|
|
|
%vdata = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 0, i32 %ofs, i32 0, i32 78, i32 0)
|
|
|
|
|
%vdata.f = bitcast <4 x i32> %vdata to <4 x float>
|
|
|
|
|
ret <4 x float> %vdata.f
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}tbuffer_load_both:
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, dfmt:14, nfmt:4, 0 idxen offen
|
|
|
|
|
; GFX10: tbuffer_load_format_xyzw {{v\[[0-9]+:[0-9]+\]}}, {{v\[[0-9]+:[0-9]+\]}}, {{s\[[0-9]+:[0-9]+\]}}, format:78, 0 idxen offen
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
define amdgpu_vs <4 x float> @tbuffer_load_both(<4 x i32> inreg, i32 %vindex, i32 %voffs) {
|
|
|
|
|
main_body:
|
|
|
|
|
%vdata = call <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32> %0, i32 %vindex, i32 %voffs, i32 0, i32 78, i32 0)
|
|
|
|
|
%vdata.f = bitcast <4 x i32> %vdata to <4 x float>
|
|
|
|
|
ret <4 x float> %vdata.f
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}buffer_load_xy:
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_xy {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, dfmt:13, nfmt:4, 0 idxen
|
|
|
|
|
; GFX10: tbuffer_load_format_xy {{v\[[0-9]+:[0-9]+\]}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, format:77, 0 idxen
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
define amdgpu_vs <2 x float> @buffer_load_xy(<4 x i32> inreg %rsrc) {
|
|
|
|
|
%vdata = call <2 x i32> @llvm.amdgcn.struct.tbuffer.load.v2i32(<4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 77, i32 0)
|
|
|
|
|
%vdata.f = bitcast <2 x i32> %vdata to <2 x float>
|
|
|
|
|
ret <2 x float> %vdata.f
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}buffer_load_x:
|
2019-05-02 00:32:58 +08:00
|
|
|
; PREGFX10: tbuffer_load_format_x {{v[0-9]+}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, dfmt:13, nfmt:4, 0 idxen
|
|
|
|
|
; GFX10: tbuffer_load_format_x {{v[0-9]+}}, {{v[0-9]+}}, {{s\[[0-9]+:[0-9]+\]}}, format:77, 0 idxen
|
[AMDGPU] New tbuffer intrinsics
Summary:
This commit adds new intrinsics
llvm.amdgcn.raw.tbuffer.load
llvm.amdgcn.struct.tbuffer.load
llvm.amdgcn.raw.tbuffer.store
llvm.amdgcn.struct.tbuffer.store
with the following changes from the llvm.amdgcn.tbuffer.* intrinsics:
* there are separate raw and struct versions: raw does not have an index
arg and sets idxen=0 in the instruction, and struct always sets
idxen=1 in the instruction even if the index is 0, to allow for the
fact that gfx9 does bounds checking differently depending on whether
idxen is set;
* there is a combined format arg (dfmt+nfmt)
* there is a combined cachepolicy arg (glc+slc)
* there are now only two offset args: one for the offset that is
included in bounds checking and swizzling, to be split between the
instruction's voffset and immoffset fields, and one for the offset
that is excluded from bounds checking and swizzling, to go into the
instruction's soffset field.
The AMDISD::TBUFFER_* SD nodes always have an index operand, all three
offset operands, combined format operand, combined cachepolicy operand,
and an extra idxen operand.
The tbuffer pseudo- and real instructions now also have a combined
format operand.
The obsolescent llvm.amdgcn.tbuffer.* and llvm.SI.tbuffer.store
intrinsics continue to work.
V2: Separate raw and struct intrinsics.
V3: Moved extract_glc and extract_slc defs to a more sensible place.
V4: Rebased on D49995.
V5: Only two separate offset args instead of three.
V6: Pseudo- and real instructions have joint format operand.
V7: Restored optionality of dfmt and nfmt in assembler.
V8: Addressed minor review comments.
Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, llvm-commits
Differential Revision: https://reviews.llvm.org/D49026
Change-Id: If22ad77e349fac3a5d2f72dda53c010377d470d4
llvm-svn: 340268
2018-08-21 19:06:05 +08:00
|
|
|
define amdgpu_vs float @buffer_load_x(<4 x i32> inreg %rsrc) {
|
|
|
|
|
%vdata = call i32 @llvm.amdgcn.struct.tbuffer.load.i32(<4 x i32> %rsrc, i32 0, i32 0, i32 0, i32 77, i32 0)
|
|
|
|
|
%vdata.f = bitcast i32 %vdata to float
|
|
|
|
|
ret float %vdata.f
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
declare i32 @llvm.amdgcn.struct.tbuffer.load.i32(<4 x i32>, i32, i32, i32, i32, i32)
|
|
|
|
|
declare <2 x i32> @llvm.amdgcn.struct.tbuffer.load.v2i32(<4 x i32>, i32, i32, i32, i32, i32)
|
|
|
|
|
declare <4 x i32> @llvm.amdgcn.struct.tbuffer.load.v4i32(<4 x i32>, i32, i32, i32, i32, i32)
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.struct.tbuffer.load.v4f32(<4 x i32>, i32, i32, i32, i32, i32)
|
|
|
|
|
|