2019-05-02 00:32:58 +08:00
|
|
|
; RUN: llc -march=amdgcn -mcpu=verde -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX6789 %s
|
|
|
|
|
; RUN: llc -march=amdgcn -mcpu=gfx900 -verify-machineinstrs < %s | FileCheck -check-prefixes=GCN,GFX6789 %s
|
|
|
|
|
; RUN: llc -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs -show-mc-encoding < %s | FileCheck -check-prefixes=GCN,GFX10 %s
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v2, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v3, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v4, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:4], v5, s[0:7], s[8:11] dmask:0xf tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:4], v5, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_tfe(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 addrspace(1)* inreg %out, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%v.vec = extractvalue {<4 x float>, i32} %v, 0
|
|
|
|
|
%v.err = extractvalue {<4 x float>, i32} %v, 1
|
|
|
|
|
store i32 %v.err, i32 addrspace(1)* %out, align 4
|
|
|
|
|
ret <4 x float> %v.vec
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe_adjust_writemask_1:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x1 tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x1 dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <2 x float> @sample_1d_tfe_adjust_writemask_1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 addrspace(1)* inreg %out, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%res.vec = extractvalue {<4 x float>,i32} %v, 0
|
|
|
|
|
%res.f = extractelement <4 x float> %res.vec, i32 0
|
|
|
|
|
%res.err = extractvalue {<4 x float>,i32} %v, 1
|
|
|
|
|
%res.errf = bitcast i32 %res.err to float
|
|
|
|
|
%res.tmp = insertelement <2 x float> undef, float %res.f, i32 0
|
|
|
|
|
%res = insertelement <2 x float> %res.tmp, float %res.errf, i32 1
|
|
|
|
|
ret <2 x float> %res
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe_adjust_writemask_2:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x2 tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x2 dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <2 x float> @sample_1d_tfe_adjust_writemask_2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%res.vec = extractvalue {<4 x float>,i32} %v, 0
|
|
|
|
|
%res.f = extractelement <4 x float> %res.vec, i32 1
|
|
|
|
|
%res.err = extractvalue {<4 x float>,i32} %v, 1
|
|
|
|
|
%res.errf = bitcast i32 %res.err to float
|
|
|
|
|
%res.tmp = insertelement <2 x float> undef, float %res.f, i32 0
|
|
|
|
|
%res = insertelement <2 x float> %res.tmp, float %res.errf, i32 1
|
|
|
|
|
ret <2 x float> %res
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe_adjust_writemask_3:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x4 tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <2 x float> @sample_1d_tfe_adjust_writemask_3(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%res.vec = extractvalue {<4 x float>,i32} %v, 0
|
|
|
|
|
%res.f = extractelement <4 x float> %res.vec, i32 2
|
|
|
|
|
%res.err = extractvalue {<4 x float>,i32} %v, 1
|
|
|
|
|
%res.errf = bitcast i32 %res.err to float
|
|
|
|
|
%res.tmp = insertelement <2 x float> undef, float %res.f, i32 0
|
|
|
|
|
%res = insertelement <2 x float> %res.tmp, float %res.errf, i32 1
|
|
|
|
|
ret <2 x float> %res
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe_adjust_writemask_4:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x8 tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:1], v2, s[0:7], s[8:11] dmask:0x8 dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <2 x float> @sample_1d_tfe_adjust_writemask_4(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%res.vec = extractvalue {<4 x float>,i32} %v, 0
|
|
|
|
|
%res.f = extractelement <4 x float> %res.vec, i32 3
|
|
|
|
|
%res.err = extractvalue {<4 x float>,i32} %v, 1
|
|
|
|
|
%res.errf = bitcast i32 %res.err to float
|
|
|
|
|
%res.tmp = insertelement <2 x float> undef, float %res.f, i32 0
|
|
|
|
|
%res = insertelement <2 x float> %res.tmp, float %res.errf, i32 1
|
|
|
|
|
ret <2 x float> %res
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe_adjust_writemask_12:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v2, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:2], v3, s[0:7], s[8:11] dmask:0x3 tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:2], v3, s[0:7], s[8:11] dmask:0x3 dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_tfe_adjust_writemask_12(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%res.vec = extractvalue {<4 x float>,i32} %v, 0
|
|
|
|
|
%res.f1 = extractelement <4 x float> %res.vec, i32 0
|
|
|
|
|
%res.f2 = extractelement <4 x float> %res.vec, i32 1
|
|
|
|
|
%res.err = extractvalue {<4 x float>,i32} %v, 1
|
|
|
|
|
%res.errf = bitcast i32 %res.err to float
|
|
|
|
|
%res.tmp1 = insertelement <4 x float> undef, float %res.f1, i32 0
|
|
|
|
|
%res.tmp2 = insertelement <4 x float> %res.tmp1, float %res.f2, i32 1
|
|
|
|
|
%res = insertelement <4 x float> %res.tmp2, float %res.errf, i32 2
|
|
|
|
|
ret <4 x float> %res
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe_adjust_writemask_24:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v2, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:2], v3, s[0:7], s[8:11] dmask:0xa tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:2], v3, s[0:7], s[8:11] dmask:0xa dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_tfe_adjust_writemask_24(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%res.vec = extractvalue {<4 x float>,i32} %v, 0
|
|
|
|
|
%res.f1 = extractelement <4 x float> %res.vec, i32 1
|
|
|
|
|
%res.f2 = extractelement <4 x float> %res.vec, i32 3
|
|
|
|
|
%res.err = extractvalue {<4 x float>,i32} %v, 1
|
|
|
|
|
%res.errf = bitcast i32 %res.err to float
|
|
|
|
|
%res.tmp1 = insertelement <4 x float> undef, float %res.f1, i32 0
|
|
|
|
|
%res.tmp2 = insertelement <4 x float> %res.tmp1, float %res.f2, i32 1
|
|
|
|
|
%res = insertelement <4 x float> %res.tmp2, float %res.errf, i32 2
|
|
|
|
|
ret <4 x float> %res
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_tfe_adjust_writemask_134:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v2, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v3, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v4, s[0:7], s[8:11] dmask:0xd tfe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v4, s[0:7], s[8:11] dmask:0xd dim:SQ_RSRC_IMG_1D tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_tfe_adjust_writemask_134(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%res.vec = extractvalue {<4 x float>,i32} %v, 0
|
|
|
|
|
%res.f1 = extractelement <4 x float> %res.vec, i32 0
|
|
|
|
|
%res.f2 = extractelement <4 x float> %res.vec, i32 2
|
|
|
|
|
%res.f3 = extractelement <4 x float> %res.vec, i32 3
|
|
|
|
|
%res.err = extractvalue {<4 x float>,i32} %v, 1
|
|
|
|
|
%res.errf = bitcast i32 %res.err to float
|
|
|
|
|
%res.tmp1 = insertelement <4 x float> undef, float %res.f1, i32 0
|
|
|
|
|
%res.tmp2 = insertelement <4 x float> %res.tmp1, float %res.f2, i32 1
|
|
|
|
|
%res.tmp3 = insertelement <4 x float> %res.tmp2, float %res.f3, i32 2
|
|
|
|
|
%res = insertelement <4 x float> %res.tmp3, float %res.errf, i32 3
|
|
|
|
|
ret <4 x float> %res
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_lwe:
|
|
|
|
|
; GCN: v_mov_b32_e32 v0, 0
|
|
|
|
|
; GCN: v_mov_b32_e32 v1, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v2, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v3, v0
|
|
|
|
|
; GCN: v_mov_b32_e32 v4, v0
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:4], v5, s[0:7], s[8:11] dmask:0xf lwe{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:4], v5, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D lwe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_lwe(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 addrspace(1)* inreg %out, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 2, i32 0)
|
|
|
|
|
%v.vec = extractvalue {<4 x float>, i32} %v, 0
|
|
|
|
|
%v.err = extractvalue {<4 x float>, i32} %v, 1
|
|
|
|
|
store i32 %v.err, i32 addrspace(1)* %out, align 4
|
|
|
|
|
ret <4 x float> %v.vec
|
|
|
|
|
}
|
|
|
|
|
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
; GCN-LABEL: {{^}}sample_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32 15, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_3d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_3D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_3d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %r) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f32(i32 15, float %s, float %t, float %r, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_cube:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf da{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_CUBE ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_cube(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %face) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.cube.v4f32.f32(i32 15, float %s, float %t, float %face, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1darray:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf da{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D_ARRAY ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %slice) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.1darray.v4f32.f32(i32 15, float %s, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_2darray:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf da{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D_ARRAY ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_2darray(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %slice) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.2darray.v4f32.f32(i32 15, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.1d.v4f32.f32(i32 15, float %zcompare, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.2d.v4f32.f32(i32 15, float %zcompare, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_cl v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.cl.1d.v4f32.f32(i32 15, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.cl.2d.v4f32.f32(i32 15, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.1d.v4f32.f32(i32 15, float %zcompare, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f32(i32 15, float %zcompare, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_b_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_b v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f32(i32 15, float %bias, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_b_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_b v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f32(i32 15, float %bias, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_b_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_b v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_b v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_b_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f32(i32 15, float %bias, float %zcompare, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_b_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_b v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_b v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_b_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f32(i32 15, float %bias, float %zcompare, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_b_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_b_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_b_cl v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f32(i32 15, float %bias, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_b_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_b_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_b_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f32(i32 15, float %bias, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_b_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_b_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_b_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_b_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f32(i32 15, float %bias, float %zcompare, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_b_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_b_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_b_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_b_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %bias, float %zcompare, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f32(i32 15, float %bias, float %zcompare, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_d_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_d v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_d v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dsdv, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f32.f32(i32 15, float %dsdh, float %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_d_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_d v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_d v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f32.f32(i32 15, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_d_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_d_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dsdv, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_d_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_d_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_d_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_d_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_d_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dsdv, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f32.f32(i32 15, float %dsdh, float %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_d_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_d_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_d_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f32.f32(i32 15, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_d_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_d_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dsdv, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_d_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_d_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_cd_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_cd v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_cd v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dsdv, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f32.f32(i32 15, float %dsdh, float %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_cd_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_cd v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_cd v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f32.f32(i32 15, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_cd_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_cd v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_cd v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_cd_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dsdv, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dsdv, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_cd_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_cd v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_cd v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_cd_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_cd_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_cd_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_cd_cl v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dsdv, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f32.f32(i32 15, float %dsdh, float %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_cd_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_cd_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_cd_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f32.f32(i32 15, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_cd_cl_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_cd_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_cd_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_cd_cl_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dsdv, float %s, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dsdv, float %s, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_cd_cl_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_cd_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_cd_cl v[0:3], v[0:7], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_cd_cl_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f32.f32(i32 15, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %clamp, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_l_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_l v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_l v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %lod) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.l.1d.v4f32.f32(i32 15, float %s, float %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_l_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_l v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_l v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t, float %lod) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.l.2d.v4f32.f32(i32 15, float %s, float %t, float %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_l_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_l v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_l v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_l_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %lod) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.l.1d.v4f32.f32(i32 15, float %zcompare, float %s, float %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_l_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_l v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_l v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_l_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %t, float %lod) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.l.2d.v4f32.f32(i32 15, float %zcompare, float %s, float %t, float %lod, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_lz_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_lz v[0:3], v0, s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_lz v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_lz_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.lz.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_lz_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.lz.2d.v4f32.f32(i32 15, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_lz_1d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_lz v[0:3], v[0:1], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_lz_1d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.lz.1d.v4f32.f32(i32 15, float %zcompare, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_lz_2d:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_lz v[0:3], v[0:3], s[0:7], s[8:11] dmask:0xf{{$}}
|
|
|
|
|
; GFX10: image_sample_c_lz v[0:3], v[0:2], s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_2D ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_lz_2d(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %zcompare, float %s, float %t) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.c.lz.2d.v4f32.f32(i32 15, float %zcompare, float %s, float %t, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_c_d_o_2darray_V1:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d_o v0, v[0:15], s[0:7], s[8:11] dmask:0x4 da{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d_o v0, v[0:15], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps float @sample_c_d_o_2darray_V1(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call float @llvm.amdgcn.image.sample.c.d.o.2darray.f32.f32.f32(i32 4, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret float %v
|
|
|
|
|
}
|
|
|
|
|
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
; GCN-LABEL: {{^}}sample_c_d_o_2darray_V1_tfe:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d_o v[9:10], v[0:15], s[0:7], s[8:11] dmask:0x4 tfe da{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d_o v[0:1], [v10, v9, v2, v3, v4, v5, v6, v7, v8], s[0:7], s[8:11] dmask:0x4 dim:SQ_RSRC_IMG_2D_ARRAY tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps float @sample_c_d_o_2darray_V1_tfe(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice, i32 addrspace(1)* inreg %out) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {float,i32} @llvm.amdgcn.image.sample.c.d.o.2darray.f32i32.f32.f32(i32 4, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%v.vec = extractvalue {float, i32} %v, 0
|
|
|
|
|
%v.err = extractvalue {float, i32} %v, 1
|
|
|
|
|
store i32 %v.err, i32 addrspace(1)* %out, align 4
|
|
|
|
|
ret float %v.vec
|
|
|
|
|
}
|
|
|
|
|
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
; GCN-LABEL: {{^}}sample_c_d_o_2darray_V2:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d_o v[0:1], v[0:15], s[0:7], s[8:11] dmask:0x6 da{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d_o v[0:1], v[0:15], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <2 x float> @sample_c_d_o_2darray_V2(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f32.f32(i32 6, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <2 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
; GCN-LABEL: {{^}}sample_c_d_o_2darray_V2_tfe:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample_c_d_o v[9:11], v[0:15], s[0:7], s[8:11] dmask:0x6 tfe da{{$}}
|
|
|
|
|
; GFX10: image_sample_c_d_o v[0:2], [v11, v10, v9, v3, v4, v5, v6, v7, v8], s[0:7], s[8:11] dmask:0x6 dim:SQ_RSRC_IMG_2D_ARRAY tfe ;
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_c_d_o_2darray_V2_tfe(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call {<2 x float>, i32} @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32i32.f32.f32(i32 6, i32 %offset, float %zcompare, float %dsdh, float %dtdh, float %dsdv, float %dtdv, float %s, float %t, float %slice, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 1, i32 0)
|
|
|
|
|
%v.vec = extractvalue {<2 x float>, i32} %v, 0
|
|
|
|
|
%v.f1 = extractelement <2 x float> %v.vec, i32 0
|
|
|
|
|
%v.f2 = extractelement <2 x float> %v.vec, i32 1
|
|
|
|
|
%v.err = extractvalue {<2 x float>, i32} %v, 1
|
|
|
|
|
%v.errf = bitcast i32 %v.err to float
|
|
|
|
|
%res.0 = insertelement <4 x float> undef, float %v.f1, i32 0
|
|
|
|
|
%res.1 = insertelement <4 x float> %res.0, float %v.f2, i32 1
|
|
|
|
|
%res.2 = insertelement <4 x float> %res.1, float %v.errf, i32 2
|
|
|
|
|
ret <4 x float> %res.2
|
|
|
|
|
}
|
|
|
|
|
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
; GCN-LABEL: {{^}}sample_1d_unorm:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf unorm{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D unorm ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_unorm(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 1, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_glc:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf glc{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D glc ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_glc(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 1)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_slc:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf slc{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D slc ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_slc(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 2)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}sample_1d_glc_slc:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GFX6789: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf glc slc{{$}}
|
|
|
|
|
; GFX10: image_sample v[0:3], v0, s[0:7], s[8:11] dmask:0xf dim:SQ_RSRC_IMG_1D glc slc ;
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
define amdgpu_ps <4 x float> @sample_1d_glc_slc(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%v = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 3)
|
|
|
|
|
ret <4 x float> %v
|
|
|
|
|
}
|
|
|
|
|
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_0:
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v0, v0, s[0:7], s[8:11] dmask:0x1
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps float @adjust_writemask_sample_0(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%elt0 = extractelement <4 x float> %r, i32 0
|
|
|
|
|
ret float %elt0
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_01
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:1], v0, s[0:7], s[8:11] dmask:0x3
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <2 x float> @adjust_writemask_sample_01(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <2 x i32> <i32 0, i32 1>
|
|
|
|
|
ret <2 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_012
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:2], v0, s[0:7], s[8:11] dmask:0x7
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <3 x float> @adjust_writemask_sample_012(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <3 x i32> <i32 0, i32 1, i32 2>
|
|
|
|
|
ret <3 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_12
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:1], v0, s[0:7], s[8:11] dmask:0x6
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <2 x float> @adjust_writemask_sample_12(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <2 x i32> <i32 1, i32 2>
|
|
|
|
|
ret <2 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_03
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:1], v0, s[0:7], s[8:11] dmask:0x9
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <2 x float> @adjust_writemask_sample_03(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <2 x i32> <i32 0, i32 3>
|
|
|
|
|
ret <2 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_13
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:1], v0, s[0:7], s[8:11] dmask:0xa
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <2 x float> @adjust_writemask_sample_13(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <2 x i32> <i32 1, i32 3>
|
|
|
|
|
ret <2 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_123
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:2], v0, s[0:7], s[8:11] dmask:0xe
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <3 x float> @adjust_writemask_sample_123(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 15, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <3 x i32> <i32 1, i32 2, i32 3>
|
|
|
|
|
ret <3 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_none_enabled
|
|
|
|
|
; GCN-NOT: image
|
|
|
|
|
define amdgpu_ps <4 x float> @adjust_writemask_sample_none_enabled(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 0, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
ret <4 x float> %r
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_123_to_12
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:1], v0, s[0:7], s[8:11] dmask:0x6
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <2 x float> @adjust_writemask_sample_123_to_12(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 14, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <2 x i32> <i32 0, i32 1>
|
|
|
|
|
ret <2 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
|
|
|
|
; GCN-LABEL: {{^}}adjust_writemask_sample_013_to_13
|
2019-05-02 00:32:58 +08:00
|
|
|
; GCN: image_sample v[0:1], v0, s[0:7], s[8:11] dmask:0xa
|
AMDGPU: Convert test cases to the dimension-aware intrinsics
Summary:
Also explicitly port over some tests in llvm.amdgcn.image.* that were
missing. Some tests are removed because they no longer apply (i.e.
explicitly testing building an address vector via insertelement).
This is in preparation for the eventual removal of the old-style
intrinsics.
Some additional notes:
- constant-address-space-32bit.ll: change some GCN-NEXT to GCN because
the instruction schedule was subtly altered
- insert_vector_elt.ll: the old test didn't actually test anything,
because %tmp1 was not used; remove the load, because it doesn't work
(Because of the amdgpu_ps calling convention? In any case, it's
orthogonal to what the test claims to be testing.)
Change-Id: Idfa99b6512ad139e755e82b8b89548ab08f0afcf
Reviewers: arsenm, rampitec
Subscribers: MatzeB, qcolombet, kzhuravl, wdng, yaxunl, dstuttard, tpr, t-tye, javed.absar, llvm-commits
Differential Revision: https://reviews.llvm.org/D48018
llvm-svn: 335229
2018-06-21 21:37:19 +08:00
|
|
|
define amdgpu_ps <2 x float> @adjust_writemask_sample_013_to_13(<8 x i32> inreg %rsrc, <4 x i32> inreg %samp, float %s) {
|
|
|
|
|
main_body:
|
|
|
|
|
%r = call <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32 11, float %s, <8 x i32> %rsrc, <4 x i32> %samp, i1 0, i32 0, i32 0)
|
|
|
|
|
%out = shufflevector <4 x float> %r, <4 x float> undef, <2 x i32> <i32 1, i32 2>
|
|
|
|
|
ret <2 x float> %out
|
|
|
|
|
}
|
|
|
|
|
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.1d.v4f32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
declare {<4 x float>,i32} @llvm.amdgcn.image.sample.1d.v4f32i32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.3d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.cube.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.1darray.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.2darray.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.1d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.2d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.cl.1d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.cl.2d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.cl.1d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.cl.2d.v4f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.b.1d.v4f32.f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.b.2d.v4f32.f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.b.1d.v4f32.f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.b.2d.v4f32.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.b.cl.1d.v4f32.f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.b.cl.2d.v4f32.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.1d.v4f32.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.b.cl.2d.v4f32.f32.f32(i32, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.d.1d.v4f32.f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.d.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.d.1d.v4f32.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.d.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.d.cl.1d.v4f32.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.d.cl.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.1d.v4f32.f32.f32(i32, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.d.cl.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.cd.1d.v4f32.f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.cd.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.cd.1d.v4f32.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.cd.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.1d.v4f32.f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.cd.cl.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.1d.v4f32.f32.f32(i32, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.cd.cl.2d.v4f32.f32.f32(i32, float, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.l.1d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.l.2d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.l.1d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.l.2d.v4f32.f32(i32, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.lz.1d.v4f32.f32(i32, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.lz.2d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.lz.1d.v4f32.f32(i32, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
declare <4 x float> @llvm.amdgcn.image.sample.c.lz.2d.v4f32.f32(i32, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
|
|
|
|
|
|
|
|
|
declare float @llvm.amdgcn.image.sample.c.d.o.2darray.f32.f32.f32(i32, i32, float, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
declare {float, i32} @llvm.amdgcn.image.sample.c.d.o.2darray.f32i32.f32.f32(i32, i32, float, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
declare <2 x float> @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32.f32.f32(i32, i32, float, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
[AMDGPU] Add support for TFE/LWE in image intrinsics. 2nd try
TFE and LWE support requires extra result registers that are written in the
event of a failure in order to detect that failure case.
The specific use-case that initiated these changes is sparse texture support.
This means that if image intrinsics are used with either option turned on, the
programmer must ensure that the return type can contain all of the expected
results. This can result in redundant registers since the vector size must be a
power-of-2.
This change takes roughly 6 parts:
1. Modify the instruction defs in tablegen to add new instruction variants that
can accomodate the extra return values.
2. Updates to lowerImage in SIISelLowering.cpp to accomodate setting TFE or LWE
(where the bulk of the work for these instruction types is now done)
3. Extra verification code to catch cases where intrinsics have been used but
insufficient return registers are used.
4. Modification to the adjustWritemask optimisation to account for TFE/LWE being
enabled (requires extra registers to be maintained for error return value).
5. An extra pass to zero initialize the error value return - this is because if
the error does not occur, the register is not written and thus must be zeroed
before use. Also added a new (on by default) option to ensure ALL return values
are zero-initialized that is required for sparse texture support.
6. Disable the inst_combine optimization in the presence of tfe/lwe (later TODO
for this to re-enable and handle correctly).
There's an additional fix now to avoid a dmask=0
For an image intrinsic with tfe where all result channels except tfe
were unused, I was getting an image instruction with dmask=0 and only a
single vgpr result for tfe. That is incorrect because the hardware
assumes there is at least one vgpr result, plus the one for tfe.
Fixed by forcing dmask to 1, which gives the desired two vgpr result
with tfe in the second one.
The TFE or LWE result is returned from the intrinsics using an aggregate
type. Look in the test code provided to see how this works, but in essence IR
code to invoke the intrinsic looks as follows:
%v = call {<4 x float>,i32} @llvm.amdgcn.image.load.1d.v4f32i32.i32(i32 15,
i32 %s, <8 x i32> %rsrc, i32 1, i32 0)
%v.vec = extractvalue {<4 x float>, i32} %v, 0
%v.err = extractvalue {<4 x float>, i32} %v, 1
This re-submit of the change also includes a slight modification in
SIISelLowering.cpp to work-around a compiler bug for the powerpc_le
platform that caused a buildbot failure on a previous submission.
Differential revision: https://reviews.llvm.org/D48826
Change-Id: If222bc03642e76cf98059a6bef5d5bffeda38dda
Work around for ppcle compiler bug
Change-Id: Ie284cf24b2271215be1b9dc95b485fd15000e32b
llvm-svn: 351054
2019-01-14 19:55:24 +08:00
|
|
|
declare {<2 x float>, i32} @llvm.amdgcn.image.sample.c.d.o.2darray.v2f32i32.f32.f32(i32, i32, float, float, float, float, float, float, float, float, <8 x i32>, <4 x i32>, i1, i32, i32) #1
|
AMDGPU: Dimension-aware image intrinsics
Summary:
These new image intrinsics contain the texture type as part of
their name and have each component of the address/coordinate as
individual parameters.
This is a preparatory step for implementing the A16 feature, where
coordinates are passed as half-floats or -ints, but the Z compare
value and texel offsets are still full dwords, making it difficult
or impossible to distinguish between A16 on or off in the old-style
intrinsics.
Additionally, these intrinsics pass the 'texfailpolicy' and
'cachectrl' as i32 bit fields to reduce operand clutter and allow
for future extensibility.
v2:
- gather4 supports 2darray images
- fix a bug with 1D images on SI
Change-Id: I099f309e0a394082a5901ea196c3967afb867f04
Reviewers: arsenm, rampitec, b-sumner
Subscribers: kzhuravl, wdng, yaxunl, dstuttard, tpr, llvm-commits, t-tye
Differential Revision: https://reviews.llvm.org/D44939
llvm-svn: 329166
2018-04-04 18:58:54 +08:00
|
|
|
|
|
|
|
|
attributes #0 = { nounwind }
|
|
|
|
|
attributes #1 = { nounwind readonly }
|
|
|
|
|
attributes #2 = { nounwind readnone }
|