[llvm] a4a3ac1 - [AMDGPU] Remove extract_subvector patterns
Jay Foad via llvm-commits
llvm-commits at lists.llvm.org
Tue Jun 6 06:08:20 PDT 2023
Author: Jay Foad
Date: 2023-06-06T14:04:50+01:00
New Revision: a4a3ac10cb1a40ccebed4e81cd7e94f1eb71602d
URL: https://github.com/llvm/llvm-project/commit/a4a3ac10cb1a40ccebed4e81cd7e94f1eb71602d
DIFF: https://github.com/llvm/llvm-project/commit/a4a3ac10cb1a40ccebed4e81cd7e94f1eb71602d.diff
LOG: [AMDGPU] Remove extract_subvector patterns
Removing them seems to slightly increase code quality as well as
simplifying both the tablegen and C++ parts of the code.
Differential Revision: https://reviews.llvm.org/D149853
Added:
Modified:
llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
llvm/lib/Target/AMDGPU/SIInstructions.td
llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
llvm/test/CodeGen/AMDGPU/greedy-reverse-local-assignment.ll
llvm/test/CodeGen/AMDGPU/sra.ll
Removed:
################################################################################
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
index b1c461fcf9088..3def0c47661d7 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUISelLowering.cpp
@@ -1441,11 +1441,6 @@ SDValue AMDGPUTargetLowering::LowerEXTRACT_SUBVECTOR(SDValue Op,
unsigned NumSrcElt = SrcVT.getVectorNumElements();
assert(NumElt % 2 == 0 && NumSrcElt % 2 == 0 && "expect legal types");
- // We have some TableGen patterns for when the extracted vector is exactly
- // the low or high half of the operand.
- if ((NumSrcElt == 2 * NumElt) && (Start == 0 || Start == NumElt))
- return Op;
-
// Extract 32-bit registers at a time.
EVT NewSrcVT = EVT::getVectorVT(*DAG.getContext(), MVT::i32, NumSrcElt / 2);
EVT NewVT = NumElt == 2
diff --git a/llvm/lib/Target/AMDGPU/SIInstructions.td b/llvm/lib/Target/AMDGPU/SIInstructions.td
index 17fbce3e2edcb..f9eba91d22b79 100644
--- a/llvm/lib/Target/AMDGPU/SIInstructions.td
+++ b/llvm/lib/Target/AMDGPU/SIInstructions.td
@@ -1356,66 +1356,6 @@ foreach Index = 0-15 in {
}
-def : Pat <
- (extract_subvector v4i16:$vec, (i32 0)),
- (v2i16 (EXTRACT_SUBREG v4i16:$vec, sub0))
->;
-
-def : Pat <
- (extract_subvector v4i16:$vec, (i32 2)),
- (v2i16 (EXTRACT_SUBREG v4i16:$vec, sub1))
->;
-
-def : Pat <
- (extract_subvector v4f16:$vec, (i32 0)),
- (v2f16 (EXTRACT_SUBREG v4f16:$vec, sub0))
->;
-
-def : Pat <
- (extract_subvector v4f16:$vec, (i32 2)),
- (v2f16 (EXTRACT_SUBREG v4f16:$vec, sub1))
->;
-
-def : Pat <
- (extract_subvector v8i16:$vec, (i32 0)),
- (v4i16 (EXTRACT_SUBREG v8i16:$vec, sub0_sub1))
->;
-
-def : Pat <
- (extract_subvector v8i16:$vec, (i32 4)),
- (v4i16 (EXTRACT_SUBREG v8i16:$vec, sub2_sub3))
->;
-
-def : Pat <
- (extract_subvector v8f16:$vec, (i32 0)),
- (v4f16 (EXTRACT_SUBREG v8f16:$vec, sub0_sub1))
->;
-
-def : Pat <
- (extract_subvector v8f16:$vec, (i32 4)),
- (v4f16 (EXTRACT_SUBREG v8f16:$vec, sub2_sub3))
->;
-
-def : Pat <
- (extract_subvector v16i16:$vec, (i32 0)),
- (v8i16 (EXTRACT_SUBREG v16i16:$vec, sub0_sub1_sub2_sub3))
->;
-
-def : Pat <
- (extract_subvector v16i16:$vec, (i32 8)),
- (v8i16 (EXTRACT_SUBREG v16i16:$vec, sub4_sub5_sub6_sub7))
->;
-
-def : Pat <
- (extract_subvector v16f16:$vec, (i32 0)),
- (v8f16 (EXTRACT_SUBREG v16f16:$vec, sub0_sub1_sub2_sub3))
->;
-
-def : Pat <
- (extract_subvector v16f16:$vec, (i32 8)),
- (v8f16 (EXTRACT_SUBREG v16f16:$vec, sub4_sub5_sub6_sub7))
->;
-
foreach Index = 0-31 in {
def Extract_Element_v32i32_#Index : Extract_Element <
i32, v32i32, Index, !cast<SubRegIndex>(sub#Index)
diff --git a/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll b/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
index 6c564b244012a..0b2ec25f38c7d 100644
--- a/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
+++ b/llvm/test/CodeGen/AMDGPU/extract-subvector-16bit.ll
@@ -366,22 +366,20 @@ define <4 x half> @vec_8xf16_extract_4xf16(ptr addrspace(1) %p0, ptr addrspace(1
; GFX9-NEXT: global_load_dwordx4 v[2:5], v[0:1], off glc
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: .LBB2_4: ; %exit
-; GFX9-NEXT: s_mov_b32 s4, 0x5040100
+; GFX9-NEXT: v_mov_b32_e32 v0, 0x3900
+; GFX9-NEXT: v_mov_b32_e32 v1, 0x3d00
; GFX9-NEXT: s_waitcnt vmcnt(0)
-; GFX9-NEXT: v_perm_b32 v0, v3, v3, s4
-; GFX9-NEXT: v_mov_b32_e32 v1, 0x3800
-; GFX9-NEXT: v_mov_b32_e32 v3, 0x3900
-; GFX9-NEXT: v_mov_b32_e32 v4, 0x3d00
-; GFX9-NEXT: v_cmp_ge_f16_e32 vcc, 0.5, v0
-; GFX9-NEXT: v_cndmask_b32_e32 v5, v3, v4, vcc
-; GFX9-NEXT: v_cmp_nle_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT: v_cndmask_b32_e32 v6, v4, v3, vcc
; GFX9-NEXT: v_cmp_ge_f16_e32 vcc, 0.5, v2
-; GFX9-NEXT: v_cndmask_b32_e32 v0, v3, v4, vcc
-; GFX9-NEXT: v_cmp_le_f16_sdwa vcc, v2, v1 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT: v_cndmask_b32_e32 v1, v3, v4, vcc
-; GFX9-NEXT: v_pack_b32_f16 v0, v0, v1
-; GFX9-NEXT: v_pack_b32_f16 v1, v5, v6
+; GFX9-NEXT: v_mov_b32_e32 v5, 0x3800
+; GFX9-NEXT: v_cndmask_b32_e32 v4, v0, v1, vcc
+; GFX9-NEXT: v_cmp_le_f16_sdwa vcc, v2, v5 src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v0, v1, vcc
+; GFX9-NEXT: v_cmp_nge_f16_e32 vcc, 0.5, v3
+; GFX9-NEXT: v_cndmask_b32_e32 v5, v1, v0, vcc
+; GFX9-NEXT: v_cmp_ge_f16_e32 vcc, 0.5, v3
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT: v_pack_b32_f16 v1, v0, v5
+; GFX9-NEXT: v_pack_b32_f16 v0, v4, v2
; GFX9-NEXT: s_setpc_b64 s[30:31]
br i1 undef, label %T, label %F
@@ -882,22 +880,20 @@ define <4 x half> @vec_16xf16_extract_4xf16(ptr addrspace(1) %p0, ptr addrspace(
; GFX9-NEXT: s_waitcnt vmcnt(0)
; GFX9-NEXT: ; kill: killed $vgpr0 killed $vgpr1
; GFX9-NEXT: .LBB5_4: ; %exit
-; GFX9-NEXT: s_mov_b32 s4, 0x5040100
+; GFX9-NEXT: v_mov_b32_e32 v0, 0x3900
+; GFX9-NEXT: v_mov_b32_e32 v1, 0x3d00
; GFX9-NEXT: s_waitcnt vmcnt(0)
-; GFX9-NEXT: v_perm_b32 v0, v5, v5, s4
-; GFX9-NEXT: v_mov_b32_e32 v1, 0x3800
-; GFX9-NEXT: v_mov_b32_e32 v2, 0x3900
-; GFX9-NEXT: v_mov_b32_e32 v3, 0x3d00
-; GFX9-NEXT: v_cmp_ge_f16_e32 vcc, 0.5, v0
-; GFX9-NEXT: v_cndmask_b32_e32 v5, v2, v3, vcc
-; GFX9-NEXT: v_cmp_nle_f16_sdwa vcc, v0, v1 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT: v_cndmask_b32_e32 v6, v3, v2, vcc
; GFX9-NEXT: v_cmp_ge_f16_e32 vcc, 0.5, v4
-; GFX9-NEXT: v_cndmask_b32_e32 v0, v2, v3, vcc
-; GFX9-NEXT: v_cmp_le_f16_sdwa vcc, v4, v1 src0_sel:WORD_1 src1_sel:DWORD
-; GFX9-NEXT: v_cndmask_b32_e32 v1, v2, v3, vcc
-; GFX9-NEXT: v_pack_b32_f16 v0, v0, v1
-; GFX9-NEXT: v_pack_b32_f16 v1, v5, v6
+; GFX9-NEXT: v_mov_b32_e32 v3, 0x3800
+; GFX9-NEXT: v_cndmask_b32_e32 v2, v0, v1, vcc
+; GFX9-NEXT: v_cmp_le_f16_sdwa vcc, v4, v3 src0_sel:WORD_1 src1_sel:DWORD
+; GFX9-NEXT: v_cndmask_b32_e32 v3, v0, v1, vcc
+; GFX9-NEXT: v_cmp_nge_f16_e32 vcc, 0.5, v5
+; GFX9-NEXT: v_cndmask_b32_e32 v4, v1, v0, vcc
+; GFX9-NEXT: v_cmp_ge_f16_e32 vcc, 0.5, v5
+; GFX9-NEXT: v_cndmask_b32_e32 v0, v0, v1, vcc
+; GFX9-NEXT: v_pack_b32_f16 v1, v0, v4
+; GFX9-NEXT: v_pack_b32_f16 v0, v2, v3
; GFX9-NEXT: s_setpc_b64 s[30:31]
br i1 undef, label %T, label %F
diff --git a/llvm/test/CodeGen/AMDGPU/greedy-reverse-local-assignment.ll b/llvm/test/CodeGen/AMDGPU/greedy-reverse-local-assignment.ll
index 6c921441c972d..e1d253db9fce1 100644
--- a/llvm/test/CodeGen/AMDGPU/greedy-reverse-local-assignment.ll
+++ b/llvm/test/CodeGen/AMDGPU/greedy-reverse-local-assignment.ll
@@ -30,12 +30,12 @@ define <4 x half> @shuffle_v4f16_234u(ptr addrspace(1) %arg0, ptr addrspace(1) %
; REVERSEXNACK-LABEL: shuffle_v4f16_234u:
; REVERSEXNACK: ; %bb.0:
; REVERSEXNACK-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; REVERSEXNACK-NEXT: v_mov_b32_e32 v6, v1
-; REVERSEXNACK-NEXT: v_mov_b32_e32 v5, v0
-; REVERSEXNACK-NEXT: v_mov_b32_e32 v4, v3
-; REVERSEXNACK-NEXT: v_mov_b32_e32 v3, v2
-; REVERSEXNACK-NEXT: global_load_dword v0, v[5:6], off offset:4
-; REVERSEXNACK-NEXT: global_load_dwordx2 v[1:2], v[3:4], off
+; REVERSEXNACK-NEXT: v_mov_b32_e32 v6, v3
+; REVERSEXNACK-NEXT: v_mov_b32_e32 v4, v1
+; REVERSEXNACK-NEXT: v_mov_b32_e32 v3, v0
+; REVERSEXNACK-NEXT: v_mov_b32_e32 v5, v2
+; REVERSEXNACK-NEXT: global_load_dword v0, v[3:4], off offset:4
+; REVERSEXNACK-NEXT: global_load_dwordx2 v[1:2], v[5:6], off
; REVERSEXNACK-NEXT: s_waitcnt vmcnt(0)
; REVERSEXNACK-NEXT: s_setpc_b64 s[30:31]
;
diff --git a/llvm/test/CodeGen/AMDGPU/sra.ll b/llvm/test/CodeGen/AMDGPU/sra.ll
index 8ebdb29706983..f0f498dd264cc 100644
--- a/llvm/test/CodeGen/AMDGPU/sra.ll
+++ b/llvm/test/CodeGen/AMDGPU/sra.ll
@@ -187,13 +187,13 @@ define amdgpu_kernel void @ashr_v2i16(ptr addrspace(1) %out, ptr addrspace(1) %i
; VI-NEXT: s_mov_b32 s4, s0
; VI-NEXT: s_mov_b32 s5, s1
; VI-NEXT: s_waitcnt vmcnt(0)
-; VI-NEXT: v_readfirstlane_b32 s0, v0
-; VI-NEXT: v_readfirstlane_b32 s1, v1
-; VI-NEXT: s_ashr_i32 s2, s0, 16
-; VI-NEXT: s_sext_i32_i16 s0, s0
-; VI-NEXT: s_ashr_i32 s3, s1, 16
+; VI-NEXT: v_readfirstlane_b32 s0, v1
+; VI-NEXT: v_readfirstlane_b32 s1, v0
+; VI-NEXT: s_ashr_i32 s2, s1, 16
; VI-NEXT: s_sext_i32_i16 s1, s1
-; VI-NEXT: s_ashr_i32 s0, s0, s1
+; VI-NEXT: s_ashr_i32 s3, s0, 16
+; VI-NEXT: s_sext_i32_i16 s0, s0
+; VI-NEXT: s_ashr_i32 s0, s1, s0
; VI-NEXT: s_ashr_i32 s1, s2, s3
; VI-NEXT: s_lshl_b32 s1, s1, 16
; VI-NEXT: s_and_b32 s0, s0, 0xffff
More information about the llvm-commits
mailing list