[llvm] [AMDGPU] Update permlane test for GFX12 (PR #75572)
via llvm-commits
llvm-commits at lists.llvm.org
Fri Dec 15 00:44:30 PST 2023
llvmbot wrote:
<!--LLVM PR SUMMARY COMMENT-->
@llvm/pr-subscribers-backend-amdgpu
Author: Mariusz Sikora (mariusz-sikora-at-amd)
<details>
<summary>Changes</summary>
---
Patch is 36.17 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/75572.diff
2 Files Affected:
- (modified) llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane.ll (+552)
- (modified) llvm/test/CodeGen/AMDGPU/vcmpx-permlane-hazard.mir (+1)
``````````diff
diff --git a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane.ll b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane.ll
index a31a5c9a30f359..f202326b873b1f 100644
--- a/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane.ll
+++ b/llvm/test/CodeGen/AMDGPU/llvm.amdgcn.permlane.ll
@@ -3,6 +3,8 @@
; RUN: llc -global-isel=1 -amdgpu-load-store-vectorizer=0 -march=amdgcn -mcpu=gfx1010 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX10,GFX10-GISEL %s
; RUN: llc -global-isel=0 -amdgpu-load-store-vectorizer=0 -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-SDAG %s
; RUN: llc -global-isel=1 -amdgpu-load-store-vectorizer=0 -march=amdgcn -mcpu=gfx1100 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX11,GFX11-GISEL %s
+; RUN: llc -global-isel=0 -amdgpu-load-store-vectorizer=0 -march=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-SDAG %s
+; RUN: llc -global-isel=1 -amdgpu-load-store-vectorizer=0 -march=amdgcn -mcpu=gfx1200 -verify-machineinstrs < %s | FileCheck -check-prefixes=GFX12,GFX12-GISEL %s
declare i32 @llvm.amdgcn.permlane16(i32, i32, i32, i32, i1, i1)
declare i32 @llvm.amdgcn.permlanex16(i32, i32, i32, i32, i1, i1)
@@ -35,6 +37,20 @@ define amdgpu_kernel void @v_permlane16_b32_vss(ptr addrspace(1) %out, i32 %src0
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlane16_b32_vss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b128 s[4:7], s[0:1], 0x24
+; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x34
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s6
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_permlane16_b32 v0, v0, s7, s0
+; GFX12-NEXT: global_store_b32 v1, v0, s[4:5]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -66,6 +82,20 @@ define amdgpu_kernel void @v_permlane16_b32_vii(ptr addrspace(1) %out, i32 %src0
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlane16_b32_vii:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_permlane16_b32 v0, v0, 1, 2
+; GFX12-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 1, i32 2, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -100,6 +130,21 @@ define amdgpu_kernel void @v_permlane16_b32_vll(ptr addrspace(1) %out, i32 %src0
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlane16_b32_vll:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-NEXT: s_movk_i32 s2, 0x1234
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX12-NEXT: v_permlane16_b32 v0, v0, s2, 0xc1d1
+; GFX12-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 4660, i32 49617, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -159,6 +204,45 @@ define amdgpu_kernel void @v_permlane16_b32_vvv(ptr addrspace(1) %out, i32 %src0
; GFX11-GISEL-NEXT: s_nop 0
; GFX11-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-GISEL-NEXT: s_endpgm
+;
+; GFX12-SDAG-LABEL: v_permlane16_b32_vvv:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_clause 0x1
+; GFX12-SDAG-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-SDAG-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-SDAG-NEXT: v_and_b32_e32 v1, 0x3ff, v0
+; GFX12-SDAG-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3)
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s3, v1
+; GFX12-SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, s2
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s2, v0
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_permlane16_b32 v1, v1, s3, s2
+; GFX12-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-SDAG-NEXT: s_nop 0
+; GFX12-SDAG-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-SDAG-NEXT: s_endpgm
+;
+; GFX12-GISEL-LABEL: v_permlane16_b32_vvv:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_clause 0x1
+; GFX12-GISEL-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-GISEL-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-GISEL-NEXT: v_and_b32_e32 v1, 0x3ff, v0
+; GFX12-GISEL-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_2)
+; GFX12-GISEL-NEXT: v_readfirstlane_b32 s4, v0
+; GFX12-GISEL-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX12-GISEL-NEXT: v_readfirstlane_b32 s3, v1
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 0
+; GFX12-GISEL-NEXT: v_permlane16_b32 v0, v0, s3, s4
+; GFX12-GISEL-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-GISEL-NEXT: s_nop 0
+; GFX12-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-GISEL-NEXT: s_endpgm
%tidx = call i32 @llvm.amdgcn.workitem.id.x()
%tidy = call i32 @llvm.amdgcn.workitem.id.y()
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
@@ -215,6 +299,33 @@ define amdgpu_kernel void @v_permlane16_b32_vvs(ptr addrspace(1) %out, i32 %src0
; GFX11-GISEL-NEXT: s_nop 0
; GFX11-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-GISEL-NEXT: s_endpgm
+;
+; GFX12-SDAG-LABEL: v_permlane16_b32_vvs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, s2
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s2, v0
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_permlane16_b32 v1, v1, s2, s3
+; GFX12-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-SDAG-NEXT: s_nop 0
+; GFX12-SDAG-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-SDAG-NEXT: s_endpgm
+;
+; GFX12-GISEL-LABEL: v_permlane16_b32_vvs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-GISEL-NEXT: v_readfirstlane_b32 s4, v0
+; GFX12-GISEL-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-GISEL-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-GISEL-NEXT: v_permlane16_b32 v0, v0, s4, s3
+; GFX12-GISEL-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-GISEL-NEXT: s_nop 0
+; GFX12-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-GISEL-NEXT: s_endpgm
%tidx = call i32 @llvm.amdgcn.workitem.id.x()
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
@@ -273,6 +384,36 @@ define amdgpu_kernel void @v_permlane16_b32_vsv(ptr addrspace(1) %out, i32 %src0
; GFX11-GISEL-NEXT: s_nop 0
; GFX11-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-GISEL-NEXT: s_endpgm
+;
+; GFX12-SDAG-LABEL: v_permlane16_b32_vsv:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-SDAG-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, s2
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s2, v0
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-SDAG-NEXT: v_permlane16_b32 v1, v1, s3, s2
+; GFX12-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-SDAG-NEXT: s_nop 0
+; GFX12-SDAG-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-SDAG-NEXT: s_endpgm
+;
+; GFX12-GISEL-LABEL: v_permlane16_b32_vsv:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-GISEL-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 0
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_1)
+; GFX12-GISEL-NEXT: v_readfirstlane_b32 s4, v0
+; GFX12-GISEL-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX12-GISEL-NEXT: v_permlane16_b32 v0, v0, s3, s4
+; GFX12-GISEL-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-GISEL-NEXT: s_nop 0
+; GFX12-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-GISEL-NEXT: s_endpgm
%tidy = call i32 @llvm.amdgcn.workitem.id.y()
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %src1, i32 %tidy, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
@@ -305,6 +446,20 @@ define amdgpu_kernel void @v_permlane16_b32_vss_fi(ptr addrspace(1) %out, i32 %s
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlane16_b32_vss_fi:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b128 s[4:7], s[0:1], 0x24
+; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x34
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s6
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_permlane16_b32 v0, v0, s7, s0 op_sel:[1,0]
+; GFX12-NEXT: global_store_b32 v1, v0, s[4:5]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 true, i1 false)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -336,6 +491,20 @@ define amdgpu_kernel void @v_permlane16_b32_vss_bc(ptr addrspace(1) %out, i32 %s
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlane16_b32_vss_bc:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b128 s[4:7], s[0:1], 0x24
+; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x34
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s6
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_permlane16_b32 v0, v0, s7, s0 op_sel:[0,1]
+; GFX12-NEXT: global_store_b32 v1, v0, s[4:5]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 true)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -367,6 +536,20 @@ define amdgpu_kernel void @v_permlane16_b32_vss_fi_bc(ptr addrspace(1) %out, i32
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlane16_b32_vss_fi_bc:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b128 s[4:7], s[0:1], 0x24
+; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x34
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s6
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_permlane16_b32 v0, v0, s7, s0 op_sel:[1,1]
+; GFX12-NEXT: global_store_b32 v1, v0, s[4:5]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlane16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 true, i1 true)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -398,6 +581,20 @@ define amdgpu_kernel void @v_permlanex16_b32_vss(ptr addrspace(1) %out, i32 %src
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlanex16_b32_vss:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b128 s[4:7], s[0:1], 0x24
+; GFX12-NEXT: s_load_b32 s0, s[0:1], 0x34
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s6
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_permlanex16_b32 v0, v0, s7, s0
+; GFX12-NEXT: global_store_b32 v1, v0, s[4:5]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlanex16(i32 %src0, i32 %src0, i32 %src1, i32 %src2, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -429,6 +626,20 @@ define amdgpu_kernel void @v_permlanex16_b32_vii(ptr addrspace(1) %out, i32 %src
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlanex16_b32_vii:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-NEXT: v_permlanex16_b32 v0, v0, 1, 2
+; GFX12-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlanex16(i32 %src0, i32 %src0, i32 1, i32 2, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -463,6 +674,21 @@ define amdgpu_kernel void @v_permlanex16_b32_vll(ptr addrspace(1) %out, i32 %src
; GFX11-NEXT: s_nop 0
; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-NEXT: s_endpgm
+;
+; GFX12-LABEL: v_permlanex16_b32_vll:
+; GFX12: ; %bb.0:
+; GFX12-NEXT: s_clause 0x1
+; GFX12-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-NEXT: s_movk_i32 s2, 0x1234
+; GFX12-NEXT: s_delay_alu instid0(VALU_DEP_1) | instid1(SALU_CYCLE_1)
+; GFX12-NEXT: v_permlanex16_b32 v0, v0, s2, 0xc1d1
+; GFX12-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-NEXT: s_nop 0
+; GFX12-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-NEXT: s_endpgm
%v = call i32 @llvm.amdgcn.permlanex16(i32 %src0, i32 %src0, i32 4660, i32 49617, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
ret void
@@ -522,6 +748,45 @@ define amdgpu_kernel void @v_permlanex16_b32_vvv(ptr addrspace(1) %out, i32 %src
; GFX11-GISEL-NEXT: s_nop 0
; GFX11-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-GISEL-NEXT: s_endpgm
+;
+; GFX12-SDAG-LABEL: v_permlanex16_b32_vvv:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_clause 0x1
+; GFX12-SDAG-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-SDAG-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-SDAG-NEXT: v_and_b32_e32 v1, 0x3ff, v0
+; GFX12-SDAG-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) | instid1(VALU_DEP_3)
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s3, v1
+; GFX12-SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, s2
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s2, v0
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_permlanex16_b32 v1, v1, s3, s2
+; GFX12-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-SDAG-NEXT: s_nop 0
+; GFX12-SDAG-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-SDAG-NEXT: s_endpgm
+;
+; GFX12-GISEL-LABEL: v_permlanex16_b32_vvv:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_clause 0x1
+; GFX12-GISEL-NEXT: s_load_b32 s2, s[0:1], 0x2c
+; GFX12-GISEL-NEXT: s_load_b64 s[0:1], s[0:1], 0x24
+; GFX12-GISEL-NEXT: v_and_b32_e32 v1, 0x3ff, v0
+; GFX12-GISEL-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1) | instskip(SKIP_4) | instid1(VALU_DEP_2)
+; GFX12-GISEL-NEXT: v_readfirstlane_b32 s4, v0
+; GFX12-GISEL-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v0, s2
+; GFX12-GISEL-NEXT: v_readfirstlane_b32 s3, v1
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 0
+; GFX12-GISEL-NEXT: v_permlanex16_b32 v0, v0, s3, s4
+; GFX12-GISEL-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-GISEL-NEXT: s_nop 0
+; GFX12-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-GISEL-NEXT: s_endpgm
%tidx = call i32 @llvm.amdgcn.workitem.id.x()
%tidy = call i32 @llvm.amdgcn.workitem.id.y()
%v = call i32 @llvm.amdgcn.permlanex16(i32 %src0, i32 %src0, i32 %tidx, i32 %tidy, i1 false, i1 false)
@@ -578,6 +843,33 @@ define amdgpu_kernel void @v_permlanex16_b32_vvs(ptr addrspace(1) %out, i32 %src
; GFX11-GISEL-NEXT: s_nop 0
; GFX11-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-GISEL-NEXT: s_endpgm
+;
+; GFX12-SDAG-LABEL: v_permlanex16_b32_vvs:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, s2
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s2, v0
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_permlanex16_b32 v1, v1, s2, s3
+; GFX12-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-SDAG-NEXT: s_nop 0
+; GFX12-SDAG-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-SDAG-NEXT: s_endpgm
+;
+; GFX12-GISEL-LABEL: v_permlanex16_b32_vvs:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-GISEL-NEXT: v_readfirstlane_b32 s4, v0
+; GFX12-GISEL-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-GISEL-NEXT: v_dual_mov_b32 v1, 0 :: v_dual_mov_b32 v0, s2
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_1)
+; GFX12-GISEL-NEXT: v_permlanex16_b32 v0, v0, s4, s3
+; GFX12-GISEL-NEXT: global_store_b32 v1, v0, s[0:1]
+; GFX12-GISEL-NEXT: s_nop 0
+; GFX12-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-GISEL-NEXT: s_endpgm
%tidx = call i32 @llvm.amdgcn.workitem.id.x()
%v = call i32 @llvm.amdgcn.permlanex16(i32 %src0, i32 %src0, i32 %tidx, i32 %src2, i1 false, i1 false)
store i32 %v, ptr addrspace(1) %out
@@ -636,6 +928,36 @@ define amdgpu_kernel void @v_permlanex16_b32_vsv(ptr addrspace(1) %out, i32 %src
; GFX11-GISEL-NEXT: s_nop 0
; GFX11-GISEL-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
; GFX11-GISEL-NEXT: s_endpgm
+;
+; GFX12-SDAG-LABEL: v_permlanex16_b32_vsv:
+; GFX12-SDAG: ; %bb.0:
+; GFX12-SDAG-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-SDAG-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-SDAG-NEXT: s_waitcnt lgkmcnt(0)
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v1, s2
+; GFX12-SDAG-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_1) | instid1(VALU_DEP_2)
+; GFX12-SDAG-NEXT: v_readfirstlane_b32 s2, v0
+; GFX12-SDAG-NEXT: v_mov_b32_e32 v0, 0
+; GFX12-SDAG-NEXT: v_permlanex16_b32 v1, v1, s3, s2
+; GFX12-SDAG-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX12-SDAG-NEXT: s_nop 0
+; GFX12-SDAG-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX12-SDAG-NEXT: s_endpgm
+;
+; GFX12-GISEL-LABEL: v_permlanex16_b32_vsv:
+; GFX12-GISEL: ; %bb.0:
+; GFX12-GISEL-NEXT: s_load_b128 s[0:3], s[0:1], 0x24
+; GFX12-GISEL-NEXT: v_bfe_u32 v0, v0, 10, 10
+; GFX12-GISEL-NEXT: v_mov_b32_e32 v1, 0
+; GFX12-GISEL-NEXT: s_delay_alu instid0(VALU_DEP_2) | instskip(SKIP_2) |...
[truncated]
``````````
</details>
https://github.com/llvm/llvm-project/pull/75572
More information about the llvm-commits
mailing list