maxjhandsome
/
llvm-project
forked from OSchip/llvm-project
1
0
Fork 0
Code Issues Pull Requests Packages Releases Wiki Activity
llvm-project/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.raw.buffer.atom...

Ignoring revisions in .git-blame-ignore-revs. Click here to bypass and see the normal blame view.

124 lines
6.9 KiB
LLVM
Raw Normal View History

[AMDGPU] Add an AMDGPU specific atomic optimizer. This commit adds a new IR level pass to the AMDGPU backend to perform atomic optimizations. It works by: - Running through a function and finding atomicrmw add/sub or uses of the atomic buffer intrinsics for add/sub. - If all arguments except the value to be added/subtracted are uniform, record the value to be optimized. - Run through the atomic operations we can optimize and, depending on whether the value is uniform/divergent use wavefront wide operations (DPP in the divergent case) to calculate the total amount to be atomically added/subtracted. - Then let only a single lane of each wavefront perform the atomic operation, reducing the total number of atomic operations in flight. - Lastly we recombine the result from the single lane to each lane of the wavefront, and calculate our individual lanes offset into the final result. Differential Revision: https://reviews.llvm.org/D51969 llvm-svn: 343973
2018-10-08 23:49:19 +08:00
;RUN: llc < %s -march=amdgcn -mcpu=verde -amdgpu-atomic-optimizations=false -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=SICI
;RUN: llc < %s -march=amdgcn -mcpu=tonga -amdgpu-atomic-optimizations=false -verify-machineinstrs | FileCheck %s -check-prefix=CHECK -check-prefix=VI
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
;CHECK-LABEL: {{^}}test1:
;CHECK-NOT: s_waitcnt
;CHECK: buffer_atomic_swap v0, off, s[0:3], 0 glc
;CHECK: s_movk_i32 [[SOFS:s[0-9]+]], 0x1ffc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_swap v0, v1, s[0:3], 0 offen glc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_swap v0, v1, s[0:3], 0 offen offset:42 glc
;CHECK-DAG: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_swap v0, off, s[0:3], [[SOFS]] offset:4 glc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_swap v0, off, s[0:3], 0{{$}}
define amdgpu_ps float @test1(<4 x i32> inreg %rsrc, i32 %data, i32 %voffset) {
main_body:
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
%o1 = call i32 @llvm.amdgcn.raw.buffer.atomic.swap.i32(i32 %data, <4 x i32> %rsrc, i32 0, i32 0, i32 0)
%o3 = call i32 @llvm.amdgcn.raw.buffer.atomic.swap.i32(i32 %o1, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
%off5 = add i32 %voffset, 42
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
%o5 = call i32 @llvm.amdgcn.raw.buffer.atomic.swap.i32(i32 %o3, <4 x i32> %rsrc, i32 %off5, i32 0, i32 0)
%o6 = call i32 @llvm.amdgcn.raw.buffer.atomic.swap.i32(i32 %o5, <4 x i32> %rsrc, i32 4, i32 8188, i32 0)
%unused = call i32 @llvm.amdgcn.raw.buffer.atomic.swap.i32(i32 %o6, <4 x i32> %rsrc, i32 0, i32 0, i32 0)
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
%out = bitcast i32 %o6 to float
ret float %out
}
;CHECK-LABEL: {{^}}test2:
;CHECK-NOT: s_waitcnt
;CHECK: buffer_atomic_add v0, v1, s[0:3], 0 offen glc{{$}}
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_sub v0, v1, s[0:3], 0 offen glc slc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_smin v0, v1, s[0:3], 0 offen glc{{$}}
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_umin v0, v1, s[0:3], 0 offen glc slc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_smax v0, v1, s[0:3], 0 offen glc{{$}}
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_umax v0, v1, s[0:3], 0 offen glc slc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_and v0, v1, s[0:3], 0 offen glc{{$}}
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_or v0, v1, s[0:3], 0 offen glc slc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_xor v0, v1, s[0:3], 0 offen glc
AMDGPU: add missing llvm.amdgcn.{raw,struct}.buffer.atomic.{inc,dec} Summary: Wrapping increment/decrement. These aren't exposed by many APIs... Change-Id: I1df25c7889de5a5ba76468ad8e8a2597efa9af6c Reviewers: arsenm, tpr, dstuttard Subscribers: kzhuravl, jvesely, wdng, yaxunl, t-tye, jfb, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65283 llvm-svn: 367821
2019-08-05 17:36:06 +08:00
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_inc v0, v1, s[0:3], 0 offen glc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_dec v0, v1, s[0:3], 0 offen glc
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
define amdgpu_ps float @test2(<4 x i32> inreg %rsrc, i32 %data, i32 %voffset) {
main_body:
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
%t1 = call i32 @llvm.amdgcn.raw.buffer.atomic.add.i32(i32 %data, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
%t2 = call i32 @llvm.amdgcn.raw.buffer.atomic.sub.i32(i32 %t1, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 2)
%t3 = call i32 @llvm.amdgcn.raw.buffer.atomic.smin.i32(i32 %t2, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
%t4 = call i32 @llvm.amdgcn.raw.buffer.atomic.umin.i32(i32 %t3, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 2)
%t5 = call i32 @llvm.amdgcn.raw.buffer.atomic.smax.i32(i32 %t4, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
%t6 = call i32 @llvm.amdgcn.raw.buffer.atomic.umax.i32(i32 %t5, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 2)
%t7 = call i32 @llvm.amdgcn.raw.buffer.atomic.and.i32(i32 %t6, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
%t8 = call i32 @llvm.amdgcn.raw.buffer.atomic.or.i32(i32 %t7, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 2)
%t9 = call i32 @llvm.amdgcn.raw.buffer.atomic.xor.i32(i32 %t8, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
AMDGPU: add missing llvm.amdgcn.{raw,struct}.buffer.atomic.{inc,dec} Summary: Wrapping increment/decrement. These aren't exposed by many APIs... Change-Id: I1df25c7889de5a5ba76468ad8e8a2597efa9af6c Reviewers: arsenm, tpr, dstuttard Subscribers: kzhuravl, jvesely, wdng, yaxunl, t-tye, jfb, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65283 llvm-svn: 367821
2019-08-05 17:36:06 +08:00
%t10 = call i32 @llvm.amdgcn.raw.buffer.atomic.inc.i32(i32 %t9, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
%t11 = call i32 @llvm.amdgcn.raw.buffer.atomic.dec.i32(i32 %t10, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
%out = bitcast i32 %t11 to float
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
ret float %out
}
; Ideally, we would teach tablegen & friends that cmpswap only modifies the
; first vgpr. Since we don't do that yet, the register allocator will have to
; create copies which we don't bother to track here.
;
;CHECK-LABEL: {{^}}test3:
;CHECK-NOT: s_waitcnt
;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, off, s[0:3], 0 glc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: s_movk_i32 [[SOFS:s[0-9]+]], 0x1ffc
;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, v2, s[0:3], 0 offen glc
;CHECK: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, v2, s[0:3], 0 offen offset:44 glc
;CHECK-DAG: s_waitcnt vmcnt(0)
;CHECK: buffer_atomic_cmpswap {{v\[[0-9]+:[0-9]+\]}}, off, s[0:3], [[SOFS]] offset:4 glc
define amdgpu_ps float @test3(<4 x i32> inreg %rsrc, i32 %data, i32 %cmp, i32 %vindex, i32 %voffset) {
main_body:
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
%o1 = call i32 @llvm.amdgcn.raw.buffer.atomic.cmpswap.i32(i32 %data, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 0, i32 0)
%o3 = call i32 @llvm.amdgcn.raw.buffer.atomic.cmpswap.i32(i32 %o1, i32 %cmp, <4 x i32> %rsrc, i32 %voffset, i32 0, i32 0)
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
%ofs.5 = add i32 %voffset, 44
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
%o5 = call i32 @llvm.amdgcn.raw.buffer.atomic.cmpswap.i32(i32 %o3, i32 %cmp, <4 x i32> %rsrc, i32 %ofs.5, i32 0, i32 0)
%o6 = call i32 @llvm.amdgcn.raw.buffer.atomic.cmpswap.i32(i32 %o5, i32 %cmp, <4 x i32> %rsrc, i32 4, i32 8188, i32 0)
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
; Detecting the no-return variant doesn't work right now because of how the
; intrinsic is replaced by an instruction that feeds into an EXTRACT_SUBREG.
; Since there probably isn't a reasonable use-case of cmpswap that discards
; the return value, that seems okay.
;
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
; %unused = call i32 @llvm.amdgcn.raw.buffer.atomic.cmpswap.i32(i32 %o6, i32 %cmp, <4 x i32> %rsrc, i32 0, i32 0, i32 0)
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
%out = bitcast i32 %o6 to float
ret float %out
}
;CHECK-LABEL: {{^}}test4:
;CHECK: buffer_atomic_add v0,
define amdgpu_ps float @test4() {
main_body:
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
%v = call i32 @llvm.amdgcn.raw.buffer.atomic.add.i32(i32 1, <4 x i32> undef, i32 4, i32 0, i32 0)
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
%v.float = bitcast i32 %v to float
ret float %v.float
}
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
declare i32 @llvm.amdgcn.raw.buffer.atomic.swap.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.add.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.sub.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.smin.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.umin.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.smax.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.umax.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.and.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.or.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.xor.i32(i32, <4 x i32>, i32, i32, i32) #0
AMDGPU: add missing llvm.amdgcn.{raw,struct}.buffer.atomic.{inc,dec} Summary: Wrapping increment/decrement. These aren't exposed by many APIs... Change-Id: I1df25c7889de5a5ba76468ad8e8a2597efa9af6c Reviewers: arsenm, tpr, dstuttard Subscribers: kzhuravl, jvesely, wdng, yaxunl, t-tye, jfb, llvm-commits Tags: #llvm Differential Revision: https://reviews.llvm.org/D65283 llvm-svn: 367821
2019-08-05 17:36:06 +08:00
declare i32 @llvm.amdgcn.raw.buffer.atomic.inc.i32(i32, <4 x i32>, i32, i32, i32) #0
declare i32 @llvm.amdgcn.raw.buffer.atomic.dec.i32(i32, <4 x i32>, i32, i32, i32) #0
AMDGPU: Future-proof {raw,struct}.buffer.atomic intrinsics Summary: The ISA is really supposed to support 64-bit atomics as well, so the data type should be an overload. Mesa doesn't use these atomics yet, in fact I noticed this issue while trying to use the atomics from Mesa. Change-Id: I77f58317a085a0d3eb933cc7e99308c48a19f83e Reviewers: tpr Subscribers: kzhuravl, jvesely, wdng, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D52291 llvm-svn: 343978
2018-10-09 00:53:48 +08:00
declare i32 @llvm.amdgcn.raw.buffer.atomic.cmpswap.i32(i32, i32, <4 x i32>, i32, i32, i32) #0
[AMDGPU] New buffer intrinsics Summary: This commit adds new intrinsics llvm.amdgcn.raw.buffer.load llvm.amdgcn.raw.buffer.load.format llvm.amdgcn.raw.buffer.load.format.d16 llvm.amdgcn.struct.buffer.load llvm.amdgcn.struct.buffer.load.format llvm.amdgcn.struct.buffer.load.format.d16 llvm.amdgcn.raw.buffer.store llvm.amdgcn.raw.buffer.store.format llvm.amdgcn.raw.buffer.store.format.d16 llvm.amdgcn.struct.buffer.store llvm.amdgcn.struct.buffer.store.format llvm.amdgcn.struct.buffer.store.format.d16 llvm.amdgcn.raw.buffer.atomic.* llvm.amdgcn.struct.buffer.atomic.* with the following changes from the llvm.amdgcn.buffer.* intrinsics: * there are separate raw and struct versions: raw does not have an index arg and sets idxen=0 in the instruction, and struct always sets idxen=1 in the instruction even if the index is 0, to allow for the fact that gfx9 does bounds checking differently depending on whether idxen is set; * there is a combined cachepolicy arg (glc+slc) * there are now only two offset args: one for the offset that is included in bounds checking and swizzling, to be split between the instruction's voffset and immoffset fields, and one for the offset that is excluded from bounds checking and swizzling, to go into the instruction's soffset field. The AMDISD::BUFFER_* SD nodes always have an index operand, all three offset operands, combined cachepolicy operand, and an extra idxen operand. The obsolescent llvm.amdgcn.buffer.* intrinsics continue to work. Subscribers: arsenm, kzhuravl, wdng, nhaehnle, yaxunl, dstuttard, t-tye, jfb, llvm-commits Differential Revision: https://reviews.llvm.org/D50306 Change-Id: If897ea7dc34fcbf4d5496e98cc99a934f62fc205 llvm-svn: 340269
2018-08-21 19:07:10 +08:00
attributes #0 = { nounwind }