[llvm] 0e1d6e2 - [AMDGPU] Auto-generating lit test patterns (NFC) (#93837)
via llvm-commits
llvm-commits at lists.llvm.org
Thu Jun 6 20:32:21 PDT 2024
Author: Christudasan Devadasan
Date: 2024-06-07T09:02:18+05:30
New Revision: 0e1d6e2fc48accbfdbb349dc460a4e91edf1d884
URL: https://github.com/llvm/llvm-project/commit/0e1d6e2fc48accbfdbb349dc460a4e91edf1d884
DIFF: https://github.com/llvm/llvm-project/commit/0e1d6e2fc48accbfdbb349dc460a4e91edf1d884.diff
LOG: [AMDGPU] Auto-generating lit test patterns (NFC) (#93837)
Test CodeGen/AMDGPU/build_vector.ll has the lit patterns partially
hand-written and the rest auto-generated. It doesn't look good when
changes are required with future patches. Auto-generating the entire
pattern. Moved out the R600 test into build_vector-r600.ll.
Added:
llvm/test/CodeGen/AMDGPU/build_vector-r600.ll
Modified:
llvm/test/CodeGen/AMDGPU/build_vector.ll
Removed:
################################################################################
diff --git a/llvm/test/CodeGen/AMDGPU/build_vector-r600.ll b/llvm/test/CodeGen/AMDGPU/build_vector-r600.ll
new file mode 100644
index 0000000000000..2abcbbcdd1bc6
--- /dev/null
+++ b/llvm/test/CodeGen/AMDGPU/build_vector-r600.ll
@@ -0,0 +1,107 @@
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=r600-- -mcpu=redwood | FileCheck %s --check-prefixes=R600
+
+define amdgpu_kernel void @build_vector2 (ptr addrspace(1) %out) {
+; R600-LABEL: build_vector2:
+; R600: ; %bb.0: ; %entry
+; R600-NEXT: ALU 4, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
+; R600-NEXT: CF_END
+; R600-NEXT: PAD
+; R600-NEXT: ALU clause starting at 4:
+; R600-NEXT: MOV * T0.Y, literal.x,
+; R600-NEXT: 6(8.407791e-45), 0(0.000000e+00)
+; R600-NEXT: MOV T0.X, literal.x,
+; R600-NEXT: LSHR * T1.X, KC0[2].Y, literal.y,
+; R600-NEXT: 5(7.006492e-45), 2(2.802597e-45)
+entry:
+ store <2 x i32> <i32 5, i32 6>, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @build_vector4 (ptr addrspace(1) %out) {
+; R600-LABEL: build_vector4:
+; R600: ; %bb.0: ; %entry
+; R600-NEXT: ALU 8, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XYZW, T1.X, 1
+; R600-NEXT: CF_END
+; R600-NEXT: PAD
+; R600-NEXT: ALU clause starting at 4:
+; R600-NEXT: MOV * T0.W, literal.x,
+; R600-NEXT: 8(1.121039e-44), 0(0.000000e+00)
+; R600-NEXT: MOV * T0.Z, literal.x,
+; R600-NEXT: 7(9.809089e-45), 0(0.000000e+00)
+; R600-NEXT: MOV * T0.Y, literal.x,
+; R600-NEXT: 6(8.407791e-45), 0(0.000000e+00)
+; R600-NEXT: MOV T0.X, literal.x,
+; R600-NEXT: LSHR * T1.X, KC0[2].Y, literal.y,
+; R600-NEXT: 5(7.006492e-45), 2(2.802597e-45)
+entry:
+ store <4 x i32> <i32 5, i32 6, i32 7, i32 8>, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @build_vector_v2i16 (ptr addrspace(1) %out) {
+; R600-LABEL: build_vector_v2i16:
+; R600: ; %bb.0: ; %entry
+; R600-NEXT: ALU 2, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T4.X, T5.X, 1
+; R600-NEXT: CF_END
+; R600-NEXT: PAD
+; R600-NEXT: ALU clause starting at 4:
+; R600-NEXT: MOV T4.X, literal.x,
+; R600-NEXT: LSHR * T5.X, KC0[2].Y, literal.y,
+; R600-NEXT: 393221(5.510200e-40), 2(2.802597e-45)
+entry:
+ store <2 x i16> <i16 5, i16 6>, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @build_vector_v2i16_trunc (ptr addrspace(1) %out, i32 %a) {
+; R600-LABEL: build_vector_v2i16_trunc:
+; R600: ; %bb.0:
+; R600-NEXT: ALU 4, @4, KC0[CB0:0-32], KC1[]
+; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T4.X, T5.X, 1
+; R600-NEXT: CF_END
+; R600-NEXT: PAD
+; R600-NEXT: ALU clause starting at 4:
+; R600-NEXT: LSHR * T0.W, KC0[2].Z, literal.x,
+; R600-NEXT: 16(2.242078e-44), 0(0.000000e+00)
+; R600-NEXT: OR_INT T4.X, PV.W, literal.x,
+; R600-NEXT: LSHR * T5.X, KC0[2].Y, literal.y,
+; R600-NEXT: 327680(4.591775e-40), 2(2.802597e-45)
+ %srl = lshr i32 %a, 16
+ %trunc = trunc i32 %srl to i16
+ %ins.0 = insertelement <2 x i16> undef, i16 %trunc, i32 0
+ %ins.1 = insertelement <2 x i16> %ins.0, i16 5, i32 1
+ store <2 x i16> %ins.1, ptr addrspace(1) %out
+ ret void
+}
+
+define amdgpu_kernel void @build_v2i32_from_v4i16_shuffle(ptr addrspace(1) %out, <4 x i16> %in) {
+; R600-LABEL: build_v2i32_from_v4i16_shuffle:
+; R600: ; %bb.0: ; %entry
+; R600-NEXT: ALU 0, @10, KC0[], KC1[]
+; R600-NEXT: TEX 1 @6
+; R600-NEXT: ALU 4, @11, KC0[CB0:0-32], KC1[]
+; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
+; R600-NEXT: CF_END
+; R600-NEXT: PAD
+; R600-NEXT: Fetch clause starting at 6:
+; R600-NEXT: VTX_READ_16 T1.X, T0.X, 48, #3
+; R600-NEXT: VTX_READ_16 T0.X, T0.X, 44, #3
+; R600-NEXT: ALU clause starting at 10:
+; R600-NEXT: MOV * T0.X, 0.0,
+; R600-NEXT: ALU clause starting at 11:
+; R600-NEXT: LSHL * T0.Y, T1.X, literal.x,
+; R600-NEXT: 16(2.242078e-44), 0(0.000000e+00)
+; R600-NEXT: LSHL T0.X, T0.X, literal.x,
+; R600-NEXT: LSHR * T1.X, KC0[2].Y, literal.y,
+; R600-NEXT: 16(2.242078e-44), 2(2.802597e-45)
+entry:
+ %shuf = shufflevector <4 x i16> %in, <4 x i16> zeroinitializer, <2 x i32> <i32 0, i32 2>
+ %zextended = zext <2 x i16> %shuf to <2 x i32>
+ %shifted = shl <2 x i32> %zextended, <i32 16, i32 16>
+ store <2 x i32> %shifted, ptr addrspace(1) %out
+ ret void
+}
diff --git a/llvm/test/CodeGen/AMDGPU/build_vector.ll b/llvm/test/CodeGen/AMDGPU/build_vector.ll
index 99755133f36d6..b26d15ed3a1c8 100644
--- a/llvm/test/CodeGen/AMDGPU/build_vector.ll
+++ b/llvm/test/CodeGen/AMDGPU/build_vector.ll
@@ -1,95 +1,256 @@
-; RUN: llc < %s -mtriple=r600 -mcpu=redwood | FileCheck %s --check-prefixes=R600,ALL
-; RUN: llc < %s -mtriple=amdgcn -verify-machineinstrs | FileCheck %s --check-prefixes=GFX6,GFX678,ALL
-; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga -mattr=-flat-for-global -verify-machineinstrs | FileCheck %s --check-prefixes=GFX8,GFX678,ALL
-; RUN: llc < %s -mtriple=amdgcn-amd-amdpal -mcpu=gfx1030 -verify-machineinstrs | FileCheck %s --check-prefixes=GFX10,GFX1011,ALL
-; RUN: llc < %s -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -amdgpu-enable-vopd=0 -verify-machineinstrs | FileCheck %s --check-prefixes=GFX11,GFX1011,ALL
-; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx940 | FileCheck %s --check-prefixes=GFX940,ALL
+; NOTE: Assertions have been autogenerated by utils/update_llc_test_checks.py UTC_ARGS: --version 5
+; RUN: llc < %s -mtriple=amdgcn -mcpu=tahiti | FileCheck %s --check-prefixes=GFX6,GCN
+; RUN: llc < %s -mtriple=amdgcn -mcpu=tonga | FileCheck %s --check-prefixes=GFX8,GCN
+; RUN: llc < %s -mtriple=amdgcn-amd-amdpal -mcpu=gfx1030 | FileCheck %s --check-prefixes=GFX10,GCN
+; RUN: llc < %s -mtriple=amdgcn-amd-amdpal -mcpu=gfx1100 -amdgpu-enable-vopd=0 | FileCheck %s --check-prefixes=GFX11,GCN
+; RUN: llc < %s -mtriple=amdgcn -mcpu=gfx940 | FileCheck %s --check-prefixes=GFX940,GCN
-; ALL-LABEL: {{^}}build_vector2:
-; R600: MOV
-; R600: MOV
-; R600-NOT: MOV
-; GFX678-DAG: v_mov_b32_e32 v[[X:[0-9]]], 5
-; GFX678-DAG: v_mov_b32_e32 v[[Y:[0-9]]], 6
-; GFX1011-DAG: v_mov_b32_e32 v[[X:[0-9]]], 5
-; GFX1011-DAG: v_mov_b32_e32 v[[Y:[0-9]]], 6
-; GFX678: buffer_store_dwordx2 v[[[X]]:[[Y]]]
-; GFX10: global_store_dwordx2 v2, v[0:1], s[0:1]
-; GFX11: global_store_b64 v2, v[0:1], s[0:1]
define amdgpu_kernel void @build_vector2 (ptr addrspace(1) %out) {
+; GFX6-LABEL: build_vector2:
+; GFX6: ; %bb.0: ; %entry
+; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT: s_mov_b32 s3, 0xf000
+; GFX6-NEXT: s_mov_b32 s2, -1
+; GFX6-NEXT: v_mov_b32_e32 v0, 5
+; GFX6-NEXT: v_mov_b32_e32 v1, 6
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[0:3], 0
+; GFX6-NEXT: s_endpgm
+;
+; GFX8-LABEL: build_vector2:
+; GFX8: ; %bb.0: ; %entry
+; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT: v_mov_b32_e32 v0, 5
+; GFX8-NEXT: v_mov_b32_e32 v1, 6
+; GFX8-NEXT: s_waitcnt lgkmcnt(0)
+; GFX8-NEXT: v_mov_b32_e32 v3, s1
+; GFX8-NEXT: v_mov_b32_e32 v2, s0
+; GFX8-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
+; GFX8-NEXT: s_endpgm
+;
+; GFX10-LABEL: build_vector2:
+; GFX10: ; %bb.0: ; %entry
+; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX10-NEXT: v_mov_b32_e32 v2, 0
+; GFX10-NEXT: v_mov_b32_e32 v0, 5
+; GFX10-NEXT: v_mov_b32_e32 v1, 6
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: build_vector2:
+; GFX11: ; %bb.0: ; %entry
+; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT: v_mov_b32_e32 v2, 0
+; GFX11-NEXT: v_mov_b32_e32 v0, 5
+; GFX11-NEXT: v_mov_b32_e32 v1, 6
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: global_store_b64 v2, v[0:1], s[0:1]
+; GFX11-NEXT: s_nop 0
+; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT: s_endpgm
+;
+; GFX940-LABEL: build_vector2:
+; GFX940: ; %bb.0: ; %entry
+; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: v_mov_b32_e32 v2, 0
+; GFX940-NEXT: v_mov_b32_e32 v0, 5
+; GFX940-NEXT: v_mov_b32_e32 v1, 6
+; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] sc0 sc1
+; GFX940-NEXT: s_endpgm
entry:
store <2 x i32> <i32 5, i32 6>, ptr addrspace(1) %out
ret void
}
-; ALL-LABEL: {{^}}build_vector4:
-; R600: MOV
-; R600: MOV
-; R600: MOV
-; R600: MOV
-; R600-NOT: MOV
-; GFX678-DAG: v_mov_b32_e32 v[[X:[0-9]]], 5
-; GFX678-DAG: v_mov_b32_e32 v[[Y:[0-9]]], 6
-; GFX678-DAG: v_mov_b32_e32 v[[Z:[0-9]]], 7
-; GFX678-DAG: v_mov_b32_e32 v[[W:[0-9]]], 8
-; GFX1011-DAG: v_mov_b32_e32 v[[X:[0-9]]], 5
-; GFX1011-DAG: v_mov_b32_e32 v[[Y:[0-9]]], 6
-; GFX1011-DAG: v_mov_b32_e32 v[[Z:[0-9]]], 7
-; GFX1011-DAG: v_mov_b32_e32 v[[W:[0-9]]], 8
-; GFX678: buffer_store_dwordx4 v[[[X]]:[[W]]]
-; GFX10: global_store_dwordx4 v4, v[0:3], s[0:1]
-; GFX11: global_store_b128 v4, v[0:3], s[0:1]
define amdgpu_kernel void @build_vector4 (ptr addrspace(1) %out) {
+; GFX6-LABEL: build_vector4:
+; GFX6: ; %bb.0: ; %entry
+; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT: s_mov_b32 s3, 0xf000
+; GFX6-NEXT: s_mov_b32 s2, -1
+; GFX6-NEXT: v_mov_b32_e32 v0, 5
+; GFX6-NEXT: v_mov_b32_e32 v1, 6
+; GFX6-NEXT: v_mov_b32_e32 v2, 7
+; GFX6-NEXT: v_mov_b32_e32 v3, 8
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: buffer_store_dwordx4 v[0:3], off, s[0:3], 0
+; GFX6-NEXT: s_endpgm
+;
+; GFX8-LABEL: build_vector4:
+; GFX8: ; %bb.0: ; %entry
+; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT: v_mov_b32_e32 v0, 5
+; GFX8-NEXT: v_mov_b32_e32 v1, 6
+; GFX8-NEXT: v_mov_b32_e32 v2, 7
+; GFX8-NEXT: v_mov_b32_e32 v3, 8
+; GFX8-NEXT: s_waitcnt lgkmcnt(0)
+; GFX8-NEXT: v_mov_b32_e32 v5, s1
+; GFX8-NEXT: v_mov_b32_e32 v4, s0
+; GFX8-NEXT: flat_store_dwordx4 v[4:5], v[0:3]
+; GFX8-NEXT: s_endpgm
+;
+; GFX10-LABEL: build_vector4:
+; GFX10: ; %bb.0: ; %entry
+; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX10-NEXT: v_mov_b32_e32 v4, 0
+; GFX10-NEXT: v_mov_b32_e32 v0, 5
+; GFX10-NEXT: v_mov_b32_e32 v1, 6
+; GFX10-NEXT: v_mov_b32_e32 v2, 7
+; GFX10-NEXT: v_mov_b32_e32 v3, 8
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: build_vector4:
+; GFX11: ; %bb.0: ; %entry
+; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT: v_mov_b32_e32 v4, 0
+; GFX11-NEXT: v_mov_b32_e32 v0, 5
+; GFX11-NEXT: v_mov_b32_e32 v1, 6
+; GFX11-NEXT: v_mov_b32_e32 v2, 7
+; GFX11-NEXT: v_mov_b32_e32 v3, 8
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: global_store_b128 v4, v[0:3], s[0:1]
+; GFX11-NEXT: s_nop 0
+; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT: s_endpgm
+;
+; GFX940-LABEL: build_vector4:
+; GFX940: ; %bb.0: ; %entry
+; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: v_mov_b32_e32 v4, 0
+; GFX940-NEXT: v_mov_b32_e32 v0, 5
+; GFX940-NEXT: v_mov_b32_e32 v1, 6
+; GFX940-NEXT: v_mov_b32_e32 v2, 7
+; GFX940-NEXT: v_mov_b32_e32 v3, 8
+; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: global_store_dwordx4 v4, v[0:3], s[0:1] sc0 sc1
+; GFX940-NEXT: s_endpgm
entry:
store <4 x i32> <i32 5, i32 6, i32 7, i32 8>, ptr addrspace(1) %out
ret void
}
-
-; ALL-LABEL: {{^}}build_vector_v2i16:
-; R600: MOV
-; R600-NOT: MOV
-; GFX678: s_mov_b32 s3, 0xf000
-; GFX678: s_mov_b32 s2, -1
-; GFX678: v_mov_b32_e32 v0, 0x60005
-; GFX678: s_waitcnt lgkmcnt(0)
-; GFX678: buffer_store_dword v0, off, s[0:3], 0
-; GFX1011: v_mov_b32_e32 v0, 0
-; GFX1011: v_mov_b32_e32 v1, 0x60005
-; GFX1011: s_waitcnt lgkmcnt(0)
-; GFX10: global_store_dword v0, v1, s[0:1]
-; GFX11: global_store_b32 v0, v1, s[0:1]
define amdgpu_kernel void @build_vector_v2i16 (ptr addrspace(1) %out) {
+; GFX6-LABEL: build_vector_v2i16:
+; GFX6: ; %bb.0: ; %entry
+; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT: s_mov_b32 s3, 0xf000
+; GFX6-NEXT: s_mov_b32 s2, -1
+; GFX6-NEXT: v_mov_b32_e32 v0, 0x60005
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT: s_endpgm
+;
+; GFX8-LABEL: build_vector_v2i16:
+; GFX8: ; %bb.0: ; %entry
+; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT: v_mov_b32_e32 v2, 0x60005
+; GFX8-NEXT: s_waitcnt lgkmcnt(0)
+; GFX8-NEXT: v_mov_b32_e32 v0, s0
+; GFX8-NEXT: v_mov_b32_e32 v1, s1
+; GFX8-NEXT: flat_store_dword v[0:1], v2
+; GFX8-NEXT: s_endpgm
+;
+; GFX10-LABEL: build_vector_v2i16:
+; GFX10: ; %bb.0: ; %entry
+; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX10-NEXT: v_mov_b32_e32 v0, 0
+; GFX10-NEXT: v_mov_b32_e32 v1, 0x60005
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: build_vector_v2i16:
+; GFX11: ; %bb.0: ; %entry
+; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT: v_mov_b32_e32 v0, 0
+; GFX11-NEXT: v_mov_b32_e32 v1, 0x60005
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT: s_nop 0
+; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT: s_endpgm
+;
+; GFX940-LABEL: build_vector_v2i16:
+; GFX940: ; %bb.0: ; %entry
+; GFX940-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX940-NEXT: v_mov_b32_e32 v0, 0
+; GFX940-NEXT: v_mov_b32_e32 v1, 0x60005
+; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: global_store_dword v0, v1, s[0:1] sc0 sc1
+; GFX940-NEXT: s_endpgm
entry:
store <2 x i16> <i16 5, i16 6>, ptr addrspace(1) %out
ret void
}
-; ALL-LABEL: {{^}}build_vector_v2i16_trunc:
-; R600: LSHR
-; R600: OR_INT
-; R600: LSHR
-; R600-NOT: MOV
-; GFX6: s_mov_b32 s3, 0xf000
-; GFX6: s_waitcnt lgkmcnt(0)
-; GFX6: v_alignbit_b32 v0, 5, s4, 16
-; GFX6: buffer_store_dword v0, off, s[0:3], 0
-; GFX8: s_mov_b32 s3, 0xf000
-; GFX8: s_mov_b32 s2, -1
-; GFX8: s_waitcnt lgkmcnt(0)
-; GFX8: s_lshr_b32 s4, s4, 16
-; GFX8: s_or_b32 s4, s4, 0x50000
-; GFX8: v_mov_b32_e32 v0, s4
-; GFX8: buffer_store_dword v0, off, s[0:3], 0
-; GFX1011: v_mov_b32_e32 v0, 0
-; GFX1011: s_waitcnt lgkmcnt(0)
-; GFX10: s_lshr_b32 s2, s2, 16
-; GFX10: s_pack_ll_b32_b16 s2, s2, 5
-; GFX11: s_pack_hl_b32_b16 s2, s2, 5
-; GFX1011: v_mov_b32_e32 v1, s2
-; GFX10: global_store_dword v0, v1, s[0:1]
-; GFX11: global_store_b32 v0, v1, s[0:1]
define amdgpu_kernel void @build_vector_v2i16_trunc (ptr addrspace(1) %out, i32 %a) {
+; GFX6-LABEL: build_vector_v2i16_trunc:
+; GFX6: ; %bb.0:
+; GFX6-NEXT: s_load_dword s4, s[0:1], 0xb
+; GFX6-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x9
+; GFX6-NEXT: s_mov_b32 s3, 0xf000
+; GFX6-NEXT: s_mov_b32 s2, -1
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
+; GFX6-NEXT: v_alignbit_b32 v0, 5, s4, 16
+; GFX6-NEXT: buffer_store_dword v0, off, s[0:3], 0
+; GFX6-NEXT: s_endpgm
+;
+; GFX8-LABEL: build_vector_v2i16_trunc:
+; GFX8: ; %bb.0:
+; GFX8-NEXT: s_load_dword s2, s[0:1], 0x2c
+; GFX8-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x24
+; GFX8-NEXT: s_waitcnt lgkmcnt(0)
+; GFX8-NEXT: s_lshr_b32 s2, s2, 16
+; GFX8-NEXT: s_or_b32 s2, s2, 0x50000
+; GFX8-NEXT: v_mov_b32_e32 v0, s0
+; GFX8-NEXT: v_mov_b32_e32 v1, s1
+; GFX8-NEXT: v_mov_b32_e32 v2, s2
+; GFX8-NEXT: flat_store_dword v[0:1], v2
+; GFX8-NEXT: s_endpgm
+;
+; GFX10-LABEL: build_vector_v2i16_trunc:
+; GFX10: ; %bb.0:
+; GFX10-NEXT: s_clause 0x1
+; GFX10-NEXT: s_load_dword s2, s[0:1], 0x8
+; GFX10-NEXT: s_load_dwordx2 s[0:1], s[0:1], 0x0
+; GFX10-NEXT: v_mov_b32_e32 v0, 0
+; GFX10-NEXT: s_waitcnt lgkmcnt(0)
+; GFX10-NEXT: s_lshr_b32 s2, s2, 16
+; GFX10-NEXT: s_pack_ll_b32_b16 s2, s2, 5
+; GFX10-NEXT: v_mov_b32_e32 v1, s2
+; GFX10-NEXT: global_store_dword v0, v1, s[0:1]
+; GFX10-NEXT: s_endpgm
+;
+; GFX11-LABEL: build_vector_v2i16_trunc:
+; GFX11: ; %bb.0:
+; GFX11-NEXT: s_clause 0x1
+; GFX11-NEXT: s_load_b32 s2, s[0:1], 0x8
+; GFX11-NEXT: s_load_b64 s[0:1], s[0:1], 0x0
+; GFX11-NEXT: v_mov_b32_e32 v0, 0
+; GFX11-NEXT: s_waitcnt lgkmcnt(0)
+; GFX11-NEXT: s_pack_hl_b32_b16 s2, s2, 5
+; GFX11-NEXT: s_delay_alu instid0(SALU_CYCLE_1)
+; GFX11-NEXT: v_mov_b32_e32 v1, s2
+; GFX11-NEXT: global_store_b32 v0, v1, s[0:1]
+; GFX11-NEXT: s_nop 0
+; GFX11-NEXT: s_sendmsg sendmsg(MSG_DEALLOC_VGPRS)
+; GFX11-NEXT: s_endpgm
+;
+; GFX940-LABEL: build_vector_v2i16_trunc:
+; GFX940: ; %bb.0:
+; GFX940-NEXT: s_load_dword s4, s[0:1], 0x2c
+; GFX940-NEXT: s_load_dwordx2 s[2:3], s[0:1], 0x24
+; GFX940-NEXT: v_mov_b32_e32 v0, 0
+; GFX940-NEXT: s_waitcnt lgkmcnt(0)
+; GFX940-NEXT: s_lshr_b32 s0, s4, 16
+; GFX940-NEXT: s_pack_ll_b32_b16 s0, s0, 5
+; GFX940-NEXT: v_mov_b32_e32 v1, s0
+; GFX940-NEXT: global_store_dword v0, v1, s[2:3] sc0 sc1
+; GFX940-NEXT: s_endpgm
%srl = lshr i32 %a, 16
%trunc = trunc i32 %srl to i16
%ins.0 = insertelement <2 x i16> undef, i16 %trunc, i32 0
@@ -98,54 +259,33 @@ define amdgpu_kernel void @build_vector_v2i16_trunc (ptr addrspace(1) %out, i32
ret void
}
-; R600-LABEL: build_v2i32_from_v4i16_shuffle:
-; R600: ; %bb.0: ; %entry
-; R600-NEXT: ALU 0, @10, KC0[], KC1[]
-; R600-NEXT: TEX 1 @6
-; R600-NEXT: ALU 4, @11, KC0[CB0:0-32], KC1[]
-; R600-NEXT: MEM_RAT_CACHELESS STORE_RAW T0.XY, T1.X, 1
-; R600-NEXT: CF_END
-; R600-NEXT: PAD
-; R600-NEXT: Fetch clause starting at 6:
-; R600-NEXT: VTX_READ_16 T1.X, T0.X, 48, #3
-; R600-NEXT: VTX_READ_16 T0.X, T0.X, 44, #3
-; R600-NEXT: ALU clause starting at 10:
-; R600-NEXT: MOV * T0.X, 0.0,
-; R600-NEXT: ALU clause starting at 11:
-; R600-NEXT: LSHL * T0.Y, T1.X, literal.x,
-; R600-NEXT: 16(2.242078e-44), 0(0.000000e+00)
-; R600-NEXT: LSHL T0.X, T0.X, literal.x,
-; R600-NEXT: LSHR * T1.X, KC0[2].Y, literal.y,
-; R600-NEXT: 16(2.242078e-44), 2(2.802597e-45)
-;
+define amdgpu_kernel void @build_v2i32_from_v4i16_shuffle(ptr addrspace(1) %out, <4 x i16> %in) {
; GFX6-LABEL: build_v2i32_from_v4i16_shuffle:
; GFX6: ; %bb.0: ; %entry
; GFX6-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x9
; GFX6-NEXT: s_mov_b32 s7, 0xf000
-; GFX6-NEXT: s_waitcnt lgkmcnt(0)
-; GFX6-NEXT: s_lshl_b32 s3, s3, 16
-; GFX6-NEXT: s_lshl_b32 s2, s2, 16
; GFX6-NEXT: s_mov_b32 s6, -1
+; GFX6-NEXT: s_waitcnt lgkmcnt(0)
; GFX6-NEXT: s_mov_b32 s4, s0
; GFX6-NEXT: s_mov_b32 s5, s1
-; GFX6-NEXT: v_mov_b32_e32 v0, s2
-; GFX6-NEXT: v_mov_b32_e32 v1, s3
+; GFX6-NEXT: s_lshl_b32 s0, s3, 16
+; GFX6-NEXT: s_lshl_b32 s1, s2, 16
+; GFX6-NEXT: v_mov_b32_e32 v0, s1
+; GFX6-NEXT: v_mov_b32_e32 v1, s0
; GFX6-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
; GFX6-NEXT: s_endpgm
;
; GFX8-LABEL: build_v2i32_from_v4i16_shuffle:
; GFX8: ; %bb.0: ; %entry
; GFX8-NEXT: s_load_dwordx4 s[0:3], s[0:1], 0x24
-; GFX8-NEXT: s_mov_b32 s7, 0xf000
-; GFX8-NEXT: s_mov_b32 s6, -1
; GFX8-NEXT: s_waitcnt lgkmcnt(0)
-; GFX8-NEXT: s_mov_b32 s4, s0
-; GFX8-NEXT: s_mov_b32 s5, s1
-; GFX8-NEXT: s_lshl_b32 s0, s3, 16
-; GFX8-NEXT: s_lshl_b32 s1, s2, 16
-; GFX8-NEXT: v_mov_b32_e32 v0, s1
-; GFX8-NEXT: v_mov_b32_e32 v1, s0
-; GFX8-NEXT: buffer_store_dwordx2 v[0:1], off, s[4:7], 0
+; GFX8-NEXT: s_lshl_b32 s3, s3, 16
+; GFX8-NEXT: s_lshl_b32 s2, s2, 16
+; GFX8-NEXT: v_mov_b32_e32 v3, s1
+; GFX8-NEXT: v_mov_b32_e32 v0, s2
+; GFX8-NEXT: v_mov_b32_e32 v1, s3
+; GFX8-NEXT: v_mov_b32_e32 v2, s0
+; GFX8-NEXT: flat_store_dwordx2 v[2:3], v[0:1]
; GFX8-NEXT: s_endpgm
;
; GFX10-LABEL: build_v2i32_from_v4i16_shuffle:
@@ -185,7 +325,6 @@ define amdgpu_kernel void @build_vector_v2i16_trunc (ptr addrspace(1) %out, i32
; GFX940-NEXT: v_mov_b32_e32 v1, s3
; GFX940-NEXT: global_store_dwordx2 v2, v[0:1], s[0:1] sc0 sc1
; GFX940-NEXT: s_endpgm
-define amdgpu_kernel void @build_v2i32_from_v4i16_shuffle(ptr addrspace(1) %out, <4 x i16> %in) {
entry:
%shuf = shufflevector <4 x i16> %in, <4 x i16> zeroinitializer, <2 x i32> <i32 0, i32 2>
%zextended = zext <2 x i16> %shuf to <2 x i32>
@@ -193,3 +332,5 @@ entry:
store <2 x i32> %shifted, ptr addrspace(1) %out
ret void
}
+;; NOTE: These prefixes are unused and the list is autogenerated. Do not add tests below this line:
+; GCN: {{.*}}
More information about the llvm-commits
mailing list