[AMDGPU] Enable GCNRewritePartialRegUses pass by default. #72975
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
vpykhtin
requested review from
jayfoad,
arsenm,
perlfu,
rampitec and
petar-avramovic
|
@llvm/pr-subscribers-llvm-transforms @llvm/pr-subscribers-llvm-globalisel Author: Valery Pykhtin (vpykhtin) ChangesLet's try once again after #69957 has landed. Most of the tests has changed their register numbers, but GlobalISel/extractelement.i128.ll got additional s_set_gpr_idx_on/s_set_gpr_idx_off in extractelement_vgpr_v4i128_sgpr_idx function. I had to disable the pass for some tests involving subregisters, I'm not sure if this also should be done for spill-vgpr.ll as it has lost a lot of spills after this patch, maybe @rampitec knows. Patch is 558.66 KiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/72975.diff 30 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
index 951ed9420594b19..ef8f92a8272fe52 100644
--- a/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
+++ b/llvm/lib/Target/AMDGPU/AMDGPUTargetMachine.cpp
@@ -346,7 +346,7 @@ static cl::opt<bool> EnableMaxIlpSchedStrategy(
static cl::opt<bool> EnableRewritePartialRegUses(
"amdgpu-enable-rewrite-partial-reg-uses",
- cl::desc("Enable rewrite partial reg uses pass"), cl::init(false),
+ cl::desc("Enable rewrite partial reg uses pass"), cl::init(true),
cl::Hidden);
static cl::opt<bool> EnableHipStdPar(
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
index 33a4d3c5494f7c9..d6d7266bbe2afc8 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/extractelement-stack-lower.ll
@@ -11,14 +11,13 @@ define i32 @v_extract_v64i32_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GCN-NEXT: s_mov_b32 s4, s33
; GCN-NEXT: s_add_i32 s33, s32, 0x3fc0
; GCN-NEXT: s_and_b32 s33, s33, 0xffffc000
-; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:60 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:56 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:52 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:48 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:44 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:40 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:36 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:28 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:24 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:20 ; 4-byte Folded Spill
@@ -27,133 +26,115 @@ define i32 @v_extract_v64i32_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:8 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 offset:4 ; 4-byte Folded Spill
; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s33 ; 4-byte Folded Spill
-; GCN-NEXT: v_mov_b32_e32 v6, v2
-; GCN-NEXT: global_load_dwordx4 v[2:5], v[0:1], off
-; GCN-NEXT: global_load_dwordx4 v[16:19], v[0:1], off offset:16
-; GCN-NEXT: global_load_dwordx4 v[56:59], v[0:1], off offset:32
-; GCN-NEXT: global_load_dwordx4 v[48:51], v[0:1], off offset:48
-; GCN-NEXT: global_load_dwordx4 v[20:23], v[0:1], off offset:64
-; GCN-NEXT: global_load_dwordx4 v[44:47], v[0:1], off offset:80
-; GCN-NEXT: global_load_dwordx4 v[40:43], v[0:1], off offset:96
-; GCN-NEXT: global_load_dwordx4 v[60:63], v[0:1], off offset:112
-; GCN-NEXT: global_load_dwordx4 v[36:39], v[0:1], off offset:128
-; GCN-NEXT: global_load_dwordx4 v[32:35], v[0:1], off offset:144
-; GCN-NEXT: global_load_dwordx4 v[28:31], v[0:1], off offset:160
-; GCN-NEXT: global_load_dwordx4 v[52:55], v[0:1], off offset:176
-; GCN-NEXT: global_load_dwordx4 v[24:27], v[0:1], off offset:192
-; GCN-NEXT: global_load_dwordx4 v[7:10], v[0:1], off offset:208
+; GCN-NEXT: global_load_dwordx4 v[3:6], v[0:1], off
+; GCN-NEXT: global_load_dwordx4 v[7:10], v[0:1], off offset:16
+; GCN-NEXT: global_load_dwordx4 v[11:14], v[0:1], off offset:32
+; GCN-NEXT: global_load_dwordx4 v[15:18], v[0:1], off offset:48
+; GCN-NEXT: global_load_dwordx4 v[19:22], v[0:1], off offset:64
+; GCN-NEXT: global_load_dwordx4 v[23:26], v[0:1], off offset:80
+; GCN-NEXT: global_load_dwordx4 v[27:30], v[0:1], off offset:96
+; GCN-NEXT: global_load_dwordx4 v[31:34], v[0:1], off offset:112
+; GCN-NEXT: global_load_dwordx4 v[35:38], v[0:1], off offset:128
+; GCN-NEXT: global_load_dwordx4 v[48:51], v[0:1], off offset:144
+; GCN-NEXT: global_load_dwordx4 v[52:55], v[0:1], off offset:160
+; GCN-NEXT: global_load_dwordx4 v[39:42], v[0:1], off offset:176
+; GCN-NEXT: global_load_dwordx4 v[43:46], v[0:1], off offset:192
+; GCN-NEXT: global_load_dwordx4 v[56:59], v[0:1], off offset:208
+; GCN-NEXT: global_load_dwordx4 v[60:63], v[0:1], off offset:224
; GCN-NEXT: s_add_i32 s32, s32, 0x10000
; GCN-NEXT: s_add_i32 s32, s32, 0xffff0000
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:528 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:544 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:548 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:552 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:556 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:560 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:564 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:568 ; 4-byte Folded Spill
-; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:572 ; 4-byte Folded Spill
-; GCN-NEXT: global_load_dwordx4 v[8:11], v[0:1], off offset:224
-; GCN-NEXT: global_load_dwordx4 v[12:15], v[0:1], off offset:240
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:532 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 offset:536 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s33 offset:540 ; 4-byte Folded Spill
+; GCN-NEXT: global_load_dwordx4 v[60:63], v[0:1], off offset:240
+; GCN-NEXT: v_and_b32_e32 v0, 63, v2
; GCN-NEXT: v_lshrrev_b32_e64 v1, 6, s33
-; GCN-NEXT: v_add_u32_e32 v1, 0x100, v1
-; GCN-NEXT: buffer_store_dword v2, off, s[0:3], s33 offset:256
-; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:260
-; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:264
-; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:268
-; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:272
-; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:276
-; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:280
-; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:284
-; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:288
-; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:292
-; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:296
-; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:300
-; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:304
-; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:308
-; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:312
-; GCN-NEXT: buffer_store_dword v51, off, s[0:3], s33 offset:316
-; GCN-NEXT: buffer_store_dword v20, off, s[0:3], s33 offset:320
-; GCN-NEXT: buffer_store_dword v21, off, s[0:3], s33 offset:324
-; GCN-NEXT: buffer_store_dword v22, off, s[0:3], s33 offset:328
-; GCN-NEXT: buffer_store_dword v23, off, s[0:3], s33 offset:332
-; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:336
-; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:340
-; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:344
-; GCN-NEXT: buffer_store_dword v47, off, s[0:3], s33 offset:348
-; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:352
-; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:356
-; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:360
-; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:364
-; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:368
-; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:372
-; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 offset:376
-; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s33 offset:380
-; GCN-NEXT: buffer_store_dword v36, off, s[0:3], s33 offset:384
-; GCN-NEXT: buffer_store_dword v37, off, s[0:3], s33 offset:388
-; GCN-NEXT: buffer_store_dword v38, off, s[0:3], s33 offset:392
-; GCN-NEXT: buffer_store_dword v39, off, s[0:3], s33 offset:396
-; GCN-NEXT: buffer_store_dword v32, off, s[0:3], s33 offset:400
-; GCN-NEXT: buffer_store_dword v33, off, s[0:3], s33 offset:404
-; GCN-NEXT: buffer_store_dword v34, off, s[0:3], s33 offset:408
-; GCN-NEXT: buffer_store_dword v35, off, s[0:3], s33 offset:412
-; GCN-NEXT: buffer_store_dword v28, off, s[0:3], s33 offset:416
-; GCN-NEXT: buffer_store_dword v29, off, s[0:3], s33 offset:420
-; GCN-NEXT: buffer_store_dword v30, off, s[0:3], s33 offset:424
-; GCN-NEXT: buffer_store_dword v31, off, s[0:3], s33 offset:428
-; GCN-NEXT: buffer_store_dword v52, off, s[0:3], s33 offset:432
-; GCN-NEXT: buffer_store_dword v53, off, s[0:3], s33 offset:436
-; GCN-NEXT: buffer_store_dword v54, off, s[0:3], s33 offset:440
-; GCN-NEXT: buffer_store_dword v55, off, s[0:3], s33 offset:444
-; GCN-NEXT: buffer_store_dword v24, off, s[0:3], s33 offset:448
-; GCN-NEXT: buffer_store_dword v25, off, s[0:3], s33 offset:452
-; GCN-NEXT: buffer_store_dword v26, off, s[0:3], s33 offset:456
-; GCN-NEXT: buffer_store_dword v27, off, s[0:3], s33 offset:460
-; GCN-NEXT: buffer_load_dword v16, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v17, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v18, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v19, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v20, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v21, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v22, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v23, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v24, off, s[0:3], s33 offset:544 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v25, off, s[0:3], s33 offset:548 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v26, off, s[0:3], s33 offset:552 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v27, off, s[0:3], s33 offset:556 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v28, off, s[0:3], s33 offset:560 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v29, off, s[0:3], s33 offset:564 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v30, off, s[0:3], s33 offset:568 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v31, off, s[0:3], s33 offset:572 ; 4-byte Folded Reload
-; GCN-NEXT: v_and_b32_e32 v0, 63, v6
; GCN-NEXT: v_lshlrev_b32_e32 v0, 2, v0
+; GCN-NEXT: v_add_u32_e32 v1, 0x100, v1
; GCN-NEXT: v_add_u32_e32 v0, v1, v0
; GCN-NEXT: s_waitcnt vmcnt(0)
-; GCN-NEXT: v_mov_b32_e32 v16, v20
-; GCN-NEXT: v_mov_b32_e32 v17, v21
-; GCN-NEXT: v_mov_b32_e32 v18, v22
-; GCN-NEXT: v_mov_b32_e32 v19, v23
-; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:464
-; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:468
-; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:472
-; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:476
-; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:480
-; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:484
-; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:488
-; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:492
-; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:496
-; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:500
-; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:504
-; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:508
+; GCN-NEXT: buffer_store_dword v60, off, s[0:3], s33 offset:512 ; 4-byte Folded Spill
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: buffer_store_dword v61, off, s[0:3], s33 offset:516 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v62, off, s[0:3], s33 offset:520 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v63, off, s[0:3], s33 offset:524 ; 4-byte Folded Spill
+; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:256
+; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:260
+; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:264
+; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:268
+; GCN-NEXT: buffer_store_dword v7, off, s[0:3], s33 offset:272
+; GCN-NEXT: buffer_store_dword v8, off, s[0:3], s33 offset:276
+; GCN-NEXT: buffer_store_dword v9, off, s[0:3], s33 offset:280
+; GCN-NEXT: buffer_store_dword v10, off, s[0:3], s33 offset:284
+; GCN-NEXT: buffer_store_dword v11, off, s[0:3], s33 offset:288
+; GCN-NEXT: buffer_store_dword v12, off, s[0:3], s33 offset:292
+; GCN-NEXT: buffer_store_dword v13, off, s[0:3], s33 offset:296
+; GCN-NEXT: buffer_store_dword v14, off, s[0:3], s33 offset:300
+; GCN-NEXT: buffer_store_dword v15, off, s[0:3], s33 offset:304
+; GCN-NEXT: buffer_store_dword v16, off, s[0:3], s33 offset:308
+; GCN-NEXT: buffer_store_dword v17, off, s[0:3], s33 offset:312
+; GCN-NEXT: buffer_store_dword v18, off, s[0:3], s33 offset:316
+; GCN-NEXT: buffer_store_dword v19, off, s[0:3], s33 offset:320
+; GCN-NEXT: buffer_store_dword v20, off, s[0:3], s33 offset:324
+; GCN-NEXT: buffer_store_dword v21, off, s[0:3], s33 offset:328
+; GCN-NEXT: buffer_store_dword v22, off, s[0:3], s33 offset:332
+; GCN-NEXT: buffer_store_dword v23, off, s[0:3], s33 offset:336
+; GCN-NEXT: buffer_store_dword v24, off, s[0:3], s33 offset:340
+; GCN-NEXT: buffer_store_dword v25, off, s[0:3], s33 offset:344
+; GCN-NEXT: buffer_store_dword v26, off, s[0:3], s33 offset:348
+; GCN-NEXT: buffer_store_dword v27, off, s[0:3], s33 offset:352
+; GCN-NEXT: buffer_store_dword v28, off, s[0:3], s33 offset:356
+; GCN-NEXT: buffer_store_dword v29, off, s[0:3], s33 offset:360
+; GCN-NEXT: buffer_store_dword v30, off, s[0:3], s33 offset:364
+; GCN-NEXT: buffer_store_dword v31, off, s[0:3], s33 offset:368
+; GCN-NEXT: buffer_store_dword v32, off, s[0:3], s33 offset:372
+; GCN-NEXT: buffer_store_dword v33, off, s[0:3], s33 offset:376
+; GCN-NEXT: buffer_store_dword v34, off, s[0:3], s33 offset:380
+; GCN-NEXT: buffer_store_dword v35, off, s[0:3], s33 offset:384
+; GCN-NEXT: buffer_store_dword v36, off, s[0:3], s33 offset:388
+; GCN-NEXT: buffer_store_dword v37, off, s[0:3], s33 offset:392
+; GCN-NEXT: buffer_store_dword v38, off, s[0:3], s33 offset:396
+; GCN-NEXT: buffer_store_dword v48, off, s[0:3], s33 offset:400
+; GCN-NEXT: buffer_store_dword v49, off, s[0:3], s33 offset:404
+; GCN-NEXT: buffer_store_dword v50, off, s[0:3], s33 offset:408
+; GCN-NEXT: buffer_store_dword v51, off, s[0:3], s33 offset:412
+; GCN-NEXT: buffer_store_dword v52, off, s[0:3], s33 offset:416
+; GCN-NEXT: buffer_store_dword v53, off, s[0:3], s33 offset:420
+; GCN-NEXT: buffer_store_dword v54, off, s[0:3], s33 offset:424
+; GCN-NEXT: buffer_store_dword v55, off, s[0:3], s33 offset:428
+; GCN-NEXT: buffer_store_dword v39, off, s[0:3], s33 offset:432
+; GCN-NEXT: buffer_store_dword v40, off, s[0:3], s33 offset:436
+; GCN-NEXT: buffer_store_dword v41, off, s[0:3], s33 offset:440
+; GCN-NEXT: buffer_store_dword v42, off, s[0:3], s33 offset:444
+; GCN-NEXT: buffer_store_dword v43, off, s[0:3], s33 offset:448
+; GCN-NEXT: buffer_store_dword v44, off, s[0:3], s33 offset:452
+; GCN-NEXT: buffer_store_dword v45, off, s[0:3], s33 offset:456
+; GCN-NEXT: buffer_store_dword v46, off, s[0:3], s33 offset:460
+; GCN-NEXT: buffer_store_dword v56, off, s[0:3], s33 offset:464
+; GCN-NEXT: buffer_store_dword v57, off, s[0:3], s33 offset:468
+; GCN-NEXT: buffer_store_dword v58, off, s[0:3], s33 offset:472
+; GCN-NEXT: buffer_store_dword v59, off, s[0:3], s33 offset:476
+; GCN-NEXT: buffer_load_dword v3, off, s[0:3], s33 offset:528 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:532 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:536 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:540 ; 4-byte Folded Reload
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:480
+; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:484
+; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:488
+; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:492
+; GCN-NEXT: buffer_load_dword v3, off, s[0:3], s33 offset:512 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v4, off, s[0:3], s33 offset:516 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v5, off, s[0:3], s33 offset:520 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v6, off, s[0:3], s33 offset:524 ; 4-byte Folded Reload
+; GCN-NEXT: s_waitcnt vmcnt(0)
+; GCN-NEXT: buffer_store_dword v3, off, s[0:3], s33 offset:496
+; GCN-NEXT: buffer_store_dword v4, off, s[0:3], s33 offset:500
+; GCN-NEXT: buffer_store_dword v5, off, s[0:3], s33 offset:504
+; GCN-NEXT: buffer_store_dword v6, off, s[0:3], s33 offset:508
; GCN-NEXT: buffer_load_dword v0, v0, s[0:3], 0 offen
; GCN-NEXT: buffer_load_dword v63, off, s[0:3], s33 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v62, off, s[0:3], s33 offset:4 ; 4-byte Folded Reload
@@ -163,14 +144,13 @@ define i32 @v_extract_v64i32_varidx(ptr addrspace(1) %ptr, i32 %idx) {
; GCN-NEXT: buffer_load_dword v58, off, s[0:3], s33 offset:20 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v57, off, s[0:3], s33 offset:24 ; 4-byte Folded Reload
; GCN-NEXT: buffer_load_dword v56, off, s[0:3], s33 offset:28 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v47, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:36 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v45, off, s[0:3], s33 offset:40 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v44, off, s[0:3], s33 offset:44 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v43, off, s[0:3], s33 offset:48 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v42, off, s[0:3], s33 offset:52 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v41, off, s[0:3], s33 offset:56 ; 4-byte Folded Reload
-; GCN-NEXT: buffer_load_dword v40, off, s[0:3], s33 offset:60 ; 4-byte Folded Reload
+; GCN-NEXT: buffer_load_dword v46, off, s[0:3], s33 offset:32 ; 4-byte Folded Reload
+; GCN-NEXT:...
[truncated]
|
Comment on lines
50
to
63
| ; GFX9-NEXT: s_set_gpr_idx_on s2, gpr_idx(SRC0) | ||
| ; GFX9-NEXT: v_mov_b32_e32 v0, v2 | ||
| ; GFX9-NEXT: v_mov_b32_e32 v1, v3 | ||
| ; GFX9-NEXT: v_mov_b32_e32 v18, v2 | ||
| ; GFX9-NEXT: s_set_gpr_idx_off | ||
| ; GFX9-NEXT: v_readfirstlane_b32 s0, v0 | ||
| ; GFX9-NEXT: s_set_gpr_idx_on s2, gpr_idx(SRC0) | ||
| ; GFX9-NEXT: v_mov_b32_e32 v0, v2 | ||
| ; GFX9-NEXT: s_set_gpr_idx_off | ||
| ; GFX9-NEXT: v_readfirstlane_b32 s1, v1 | ||
| ; GFX9-NEXT: s_set_gpr_idx_on s2, gpr_idx(SRC0) | ||
| ; GFX9-NEXT: v_mov_b32_e32 v3, v3 | ||
| ; GFX9-NEXT: v_mov_b32_e32 v1, v3 | ||
| ; GFX9-NEXT: s_set_gpr_idx_off | ||
| ; GFX9-NEXT: v_readfirstlane_b32 s2, v18 | ||
| ; GFX9-NEXT: v_readfirstlane_b32 s3, v3 | ||
| ; GFX9-NEXT: v_readfirstlane_b32 s2, v0 | ||
| ; GFX9-NEXT: v_readfirstlane_b32 s3, v1 |
There was a problem hiding this comment.
V_INDIRECT_REG_READ_GPR_IDX_B32_V16 is expanded into bundle that starts with S_SET_GPR_IDX_ON and ends with S_SET_GPR_IDX_OFF
If there is nothing in between two bundles si-pre-emit-peephole will remove pair of S_SET_GPR_IDX_OFF/S_SET_GPR_IDX_ON
What happens here is that machine-scheduler moves V_READFIRSTLANE_B32 between V_INDIRECT_REG_READ_GPR_IDX_B32_V16 in a different way (there are more moves later)
We could get even better result there were no readfirstlanes between V_INDIRECT_REG_READ_GPR_IDX_B32_V16 (one pair of s_set_gpr_idx_on / s_set_gpr_idx_off)
s_set_gpr_idx_on s2, gpr_idx(SRC0)
v_mov_b32_e32 v0, v2
v_mov_b32_e32 v1, v3
v_mov_b32_e32 v18, v2
v_mov_b32_e32 v3, v3
s_set_gpr_idx_off
v_readfirstlane_b32 s0, v0
v_readfirstlane_b32 s1, v1
v_readfirstlane_b32 s2, v18
v_readfirstlane_b32 s3, v3
I see this as different issue
There was a problem hiding this comment.
Thank you for the explanation Petar, it's different issue indeed.
| # RUN: llc -march=amdgcn -mcpu=tonga %s -start-before detect-dead-lanes -stop-before machine-scheduler -verify-machineinstrs -o - | FileCheck -check-prefix=GCN %s | ||
| # RUN: llc -march=amdgcn -mcpu=tonga %s -start-before detect-dead-lanes -stop-before machine-scheduler -verify-machineinstrs -early-live-intervals -o - | FileCheck -check-prefix=GCN %s | ||
| # RUN: llc -march=amdgcn -mcpu=tonga %s -amdgpu-enable-rewrite-partial-reg-uses=false -start-before detect-dead-lanes -stop-before machine-scheduler -verify-machineinstrs -o - | FileCheck -check-prefix=GCN %s | ||
| # RUN: llc -march=amdgcn -mcpu=tonga %s -amdgpu-enable-rewrite-partial-reg-uses=false -start-before detect-dead-lanes -stop-before machine-scheduler -verify-machineinstrs -early-live-intervals -o - | FileCheck -check-prefix=GCN %s | ||
|
|
||
| # GCN-LABEL: name: dead_lane | ||
| # GCN: bb.0: |
There was a problem hiding this comment.
Why don't you update the test, it looks fine to me, it uses vgpr_32 instead of sub0:vreg_64
# GCN-NEXT: %5:vgpr_32 = nofpexcept V_MAC_F32_e32 undef %1:vgpr_32, undef %1:vgpr_32, undef %5, implicit $mode, implicit $exec
# GCN-NEXT: FLAT_STORE_DWORD undef %4:vreg_64, %5,
There was a problem hiding this comment.
Right, I was confused by the name dead-lane, enabled the test.
perlfu
reviewed
Nov 22, 2023
| ; GFX908: NumVgprs: 254 | ||
| ; GFX900: ScratchSize: 1796 | ||
| ; GFX908: NumVgprs: 252 | ||
| ; GFX900: ScratchSize: 132 |
There was a problem hiding this comment.
I think this is probably fine.
It matches the test without a branch above.
More concerningly, the spill behaviour is so bad originally because rename-independent-subregs creates a new vreg_1024 for each disconnected dwordx4 load (from the 32 x float loads).
The register allocator then spills each of these as a vreg_1024 (with only 4 lanes used).
Looks like rename-independent-subregs should be shrinking/narrowing registers, but will need a target hook to be able to find if that is possible and which register class to use.
There was a problem hiding this comment.
Looks like rename-independent-subregs should be shrinking/narrowing registers, but will need a target hook to be able to find if that is possible and which register class to use.
That is what I originally thought to do, but I didn't want to complicate things dealing with other targets. Maybe once this pass get matured it can be made target-independent (it's almost indepent at the moment). I also remember @arsenm had an intention to make it more like an utility class to be used when needed.
arsenm
approved these changes
Nov 28, 2023
|
I'm going to submit this tomorrow if there are no objections, internal testing passed. |
vpykhtin
force-pushed
the
enable_rewrite_partial_regs
branch
from
1acf542
to
9f9dd26
Compare
vpykhtin
force-pushed
the
enable_rewrite_partial_regs
branch
from
9f9dd26
to
882d881
Compare
|
vpykhtin
force-pushed
the
enable_rewrite_partial_regs
branch
from
882d881
to
6836676
Compare
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Let's try once again after #69957 has landed.
Most of the tests has changed their register numbers, but GlobalISel/extractelement.i128.ll got additional s_set_gpr_idx_on/s_set_gpr_idx_off in extractelement_vgpr_v4i128_sgpr_idx function.
I had to disable the pass for some tests involving subregisters, I'm not sure if this also should be done for spill-vgpr.ll as it has lost a lot of spills after this patch, maybe @rampitec knows.