[AMDGPU] Revert "Preliminary patch for divergence driven instruction selection. Operands Folding 1." #71710
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
jayfoad
commented
Nov 8, 2023
| } | ||
|
|
||
| for (auto &F : CopyUses) { | ||
| foldOperand(*F.OpToFold, F.UseMI, F.UseOpNo, FoldList, |
There was a problem hiding this comment.
The motivation is to remove these recursive calls to foldOperand which are one cause of #71685.
| ; GFX7-NEXT: s_movk_i32 s4, 0xffc0 | ||
| ; GFX7-NEXT: v_add_i32_e32 v0, vcc, s4, v0 | ||
| ; GFX7-NEXT: v_add_i32_e32 v1, vcc, s4, v1 | ||
| ; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xffffffc0, v0 |
There was a problem hiding this comment.
Folding improvements here.
| ; GFX906-NEXT: s_movk_i32 s4, 0xff | ||
| ; GFX906-NEXT: v_lshlrev_b32_sdwa v1, s5, v1 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0 | ||
| ; GFX906-NEXT: v_and_or_b32 v0, v0, s4, v1 | ||
| ; GFX906-NEXT: v_mov_b32_e32 v10, 8 |
There was a problem hiding this comment.
Lots of GlobalISel tests show regressions like this where constants are materialized in vgprs instead of sgprs. This does not seem to affect SelectionDAG.
There was a problem hiding this comment.
Really we need a GlobalISel solution for picking the register banks of constants. I wouldn't worry about these for now
| @@ -2465,8 +2465,7 @@ define hidden amdgpu_kernel void @negativeoffset(ptr addrspace(1) nocapture %buf | |||
| ; GFX9-NEXT: v_lshlrev_b32_sdwa v0, v3, v0 dst_sel:DWORD dst_unused:UNUSED_PAD src0_sel:DWORD src1_sel:BYTE_0 | |||
| ; GFX9-NEXT: v_add_co_u32_e32 v2, vcc, v2, v0 | |||
| ; GFX9-NEXT: v_addc_co_u32_e32 v3, vcc, 0, v1, vcc | |||
| ; GFX9-NEXT: s_movk_i32 s0, 0x1000 | |||
| ; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, s0, v2 | |||
| ; GFX9-NEXT: v_add_co_u32_e32 v0, vcc, 0x1000, v2 | |||
There was a problem hiding this comment.
Folding improvement here.
| ; SI-GISEL-NEXT: s_waitcnt vmcnt(0) | ||
| ; SI-GISEL-NEXT: v_lshrrev_b32_e32 v3, 16, v2 | ||
| ; SI-GISEL-NEXT: v_add_i32_e32 v2, vcc, s2, v2 | ||
| ; SI-GISEL-NEXT: v_add_i32_e32 v3, vcc, s2, v3 | ||
| ; SI-GISEL-NEXT: v_add_i32_e32 v2, vcc, 0xffffffe0, v2 |
There was a problem hiding this comment.
Folding improvement here.
| @@ -4,7 +4,8 @@ | |||
|
|
|||
| ; GCN-LABEL: {{^}}test_mfma_loop_zeroinit: | |||
|
|
|||
| ; GCN-COUNT-32: v_accvgpr_write_b32 a{{[0-9]+}}, 0{{$}} | |||
| ; GCN: v_mov_b32_e32 v0, 0{{$}} | |||
| ; GCN-COUNT-32: v_accvgpr_write_b32 a{{[0-9]+}}, v0{{$}} | |||
There was a problem hiding this comment.
Lots of agpr test cases show regressions like this where we fail to fold inline constants into v_accvgpr_write_b32.
|
For some reason the description is missing the reference to the original commit hash |
arsenm
reviewed
Nov 9, 2023
Comment on lines
757
to
744
| if (DestRC == &AMDGPU::AGPR_32RegClass && | ||
| TII->isInlineConstant(OpToFold, AMDGPU::OPERAND_REG_INLINE_C_INT32)) { | ||
| UseMI->setDesc(TII->get(AMDGPU::V_ACCVGPR_WRITE_B32_e64)); | ||
| UseMI->getOperand(1).ChangeToImmediate(OpToFold.getImm()); | ||
| CopiesToReplace.push_back(UseMI); | ||
| return; | ||
| } |
There was a problem hiding this comment.
Is preserving this part sufficient to avoid the AGPR regressions?
That's because I didn't actually use "git revert". Too much stuff has changed since the original commit. But I'll fix up the description manually to look more like a "git revert". |
…selection. Operands Folding 1." This reverts commit 201f892.
arsenm
approved these changes
Nov 13, 2023
|
@llvm/pr-subscribers-backend-amdgpu @llvm/pr-subscribers-llvm-globalisel Author: Jay Foad (jayfoad) ChangesThis reverts commit 201f892. Patch is 1.27 MiB, truncated to 20.00 KiB below, full version: https://github.com/llvm/llvm-project/pull/71710.diff 55 Files Affected:
diff --git a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
index 45294da5df5916a..0ec0370e21dfc16 100644
--- a/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
+++ b/llvm/lib/Target/AMDGPU/SIFoldOperands.cpp
@@ -717,24 +717,6 @@ void SIFoldOperands::foldOperand(
const TargetRegisterClass *DestRC = TRI->getRegClassForReg(*MRI, DestReg);
if (!DestReg.isPhysical()) {
- if (TRI->isSGPRClass(SrcRC) && TRI->hasVectorRegisters(DestRC)) {
- SmallVector<FoldCandidate, 4> CopyUses;
- for (auto &Use : MRI->use_nodbg_operands(DestReg)) {
- // There's no point trying to fold into an implicit operand.
- if (Use.isImplicit())
- continue;
-
- CopyUses.emplace_back(Use.getParent(),
- Use.getParent()->getOperandNo(&Use),
- &UseMI->getOperand(1));
- }
-
- for (auto &F : CopyUses) {
- foldOperand(*F.OpToFold, F.UseMI, F.UseOpNo, FoldList,
- CopiesToReplace);
- }
- }
-
if (DestRC == &AMDGPU::AGPR_32RegClass &&
TII->isInlineConstant(OpToFold, AMDGPU::OPERAND_REG_INLINE_C_INT32)) {
UseMI->setDesc(TII->get(AMDGPU::V_ACCVGPR_WRITE_B32_e64));
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/add.v2i16.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.v2i16.ll
index 26d1fbb09210c64..e4cababfe1c919a 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/add.v2i16.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/add.v2i16.ll
@@ -165,9 +165,8 @@ define <2 x i16> @v_add_v2i16_neg_inline_imm_splat(<2 x i16> %a) {
; GFX7-LABEL: v_add_v2i16_neg_inline_imm_splat:
; GFX7: ; %bb.0:
; GFX7-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX7-NEXT: s_movk_i32 s4, 0xffc0
-; GFX7-NEXT: v_add_i32_e32 v0, vcc, s4, v0
-; GFX7-NEXT: v_add_i32_e32 v1, vcc, s4, v1
+; GFX7-NEXT: v_add_i32_e32 v0, vcc, 0xffffffc0, v0
+; GFX7-NEXT: v_add_i32_e32 v1, vcc, 0xffffffc0, v1
; GFX7-NEXT: s_setpc_b64 s[30:31]
;
; GFX9-LABEL: v_add_v2i16_neg_inline_imm_splat:
@@ -180,10 +179,10 @@ define <2 x i16> @v_add_v2i16_neg_inline_imm_splat(<2 x i16> %a) {
; GFX8-LABEL: v_add_v2i16_neg_inline_imm_splat:
; GFX8: ; %bb.0:
; GFX8-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX8-NEXT: v_mov_b32_e32 v2, 0xffffffc0
-; GFX8-NEXT: v_add_u16_e32 v1, 0xffc0, v0
-; GFX8-NEXT: v_add_u16_sdwa v0, v0, v2 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
-; GFX8-NEXT: v_or_b32_e32 v0, v1, v0
+; GFX8-NEXT: v_mov_b32_e32 v1, 0xffffffc0
+; GFX8-NEXT: v_add_u16_e32 v2, 0xffc0, v0
+; GFX8-NEXT: v_add_u16_sdwa v0, v0, v1 dst_sel:WORD_1 dst_unused:UNUSED_PAD src0_sel:WORD_1 src1_sel:DWORD
+; GFX8-NEXT: v_or_b32_e32 v0, v2, v0
; GFX8-NEXT: s_setpc_b64 s[30:31]
;
; GFX10-LABEL: v_add_v2i16_neg_inline_imm_splat:
diff --git a/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f32.ll b/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f32.ll
index 9da75b093fc9cb5..1a8529b9101c3fb 100644
--- a/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f32.ll
+++ b/llvm/test/CodeGen/AMDGPU/GlobalISel/fdiv.f32.ll
@@ -436,15 +436,15 @@ define float @v_fdiv_f32_ulp25(float %a, float %b) {
; GFX6-IEEE-LABEL: v_fdiv_f32_ulp25:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v3, v0
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v4, v0
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v3
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v3, v0, v3, vcc
+; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v3, v0, v4, vcc
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-IEEE-NEXT: v_mul_f32_e32 v2, v3, v2
; GFX6-IEEE-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -569,15 +569,15 @@ define float @v_fdiv_f32_dynamic_25ulp(float %x, float %y) #0 {
; GFX6-IEEE-LABEL: v_fdiv_f32_dynamic_25ulp:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v3, v0
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v4, v0
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v3
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v3, v0, v3, vcc
+; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v3, v0, v4, vcc
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-IEEE-NEXT: v_mul_f32_e32 v2, v3, v2
; GFX6-IEEE-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -587,15 +587,15 @@ define float @v_fdiv_f32_dynamic_25ulp(float %x, float %y) #0 {
; GFX6-FLUSH-LABEL: v_fdiv_f32_dynamic_25ulp:
; GFX6-FLUSH: ; %bb.0:
; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-FLUSH-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-FLUSH-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v3, v0
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v4, v0
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v3
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v3, v0, v3, vcc
+; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v3, v0, v4, vcc
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-FLUSH-NEXT: v_mul_f32_e32 v2, v3, v2
; GFX6-FLUSH-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -1527,25 +1527,25 @@ define <2 x float> @v_fdiv_v2f32_ulp25(<2 x float> %a, <2 x float> %b) {
; GFX6-IEEE-LABEL: v_fdiv_v2f32_ulp25:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v5, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v4, v2
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v2|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v2|, v5
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v4, v2, v4, vcc
-; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v5, v0
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v6, v0
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v5
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v2, v2
-; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v5, v0, v5, vcc
+; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v6, v0, v6, vcc
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v4
; GFX6-IEEE-NEXT: v_sub_i32_e32 v0, vcc, v0, v2
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v3
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v3|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v3|, v5
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v5, v4
+; GFX6-IEEE-NEXT: v_mul_f32_e32 v4, v6, v4
; GFX6-IEEE-NEXT: v_ldexp_f32_e32 v0, v4, v0
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v4, v1
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v5
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v3, v3
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v4, v1, v4, vcc
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
@@ -1557,19 +1557,19 @@ define <2 x float> @v_fdiv_v2f32_ulp25(<2 x float> %a, <2 x float> %b) {
; GCN-FLUSH-LABEL: v_fdiv_v2f32_ulp25:
; GCN-FLUSH: ; %bb.0:
; GCN-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GCN-FLUSH-NEXT: s_mov_b32 s4, 0x6f800000
-; GCN-FLUSH-NEXT: v_mov_b32_e32 v4, 0x2f800000
-; GCN-FLUSH-NEXT: v_cmp_gt_f32_e64 vcc, |v2|, s4
-; GCN-FLUSH-NEXT: v_cndmask_b32_e32 v5, 1.0, v4, vcc
-; GCN-FLUSH-NEXT: v_cmp_gt_f32_e64 vcc, |v3|, s4
-; GCN-FLUSH-NEXT: v_cndmask_b32_e32 v4, 1.0, v4, vcc
-; GCN-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v5
+; GCN-FLUSH-NEXT: v_mov_b32_e32 v4, 0x6f800000
+; GCN-FLUSH-NEXT: v_mov_b32_e32 v5, 0x2f800000
+; GCN-FLUSH-NEXT: v_cmp_gt_f32_e64 vcc, |v2|, v4
+; GCN-FLUSH-NEXT: v_cndmask_b32_e32 v6, 1.0, v5, vcc
+; GCN-FLUSH-NEXT: v_cmp_gt_f32_e64 vcc, |v3|, v4
+; GCN-FLUSH-NEXT: v_cndmask_b32_e32 v4, 1.0, v5, vcc
+; GCN-FLUSH-NEXT: v_mul_f32_e32 v2, v2, v6
; GCN-FLUSH-NEXT: v_mul_f32_e32 v3, v3, v4
; GCN-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
; GCN-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
; GCN-FLUSH-NEXT: v_mul_f32_e32 v0, v0, v2
; GCN-FLUSH-NEXT: v_mul_f32_e32 v1, v1, v3
-; GCN-FLUSH-NEXT: v_mul_f32_e32 v0, v5, v0
+; GCN-FLUSH-NEXT: v_mul_f32_e32 v0, v6, v0
; GCN-FLUSH-NEXT: v_mul_f32_e32 v1, v4, v1
; GCN-FLUSH-NEXT: s_setpc_b64 s[30:31]
;
@@ -2316,16 +2316,16 @@ define <2 x float> @v_rcp_v2f32_ulp25(<2 x float> %x) {
; GFX6-IEEE-LABEL: v_rcp_v2f32_ulp25:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v0
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v3
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v0, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-IEEE-NEXT: v_sub_i32_e32 v0, vcc, 0, v0
; GFX6-IEEE-NEXT: v_ldexp_f32_e32 v0, v2, v0
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
@@ -2425,9 +2425,9 @@ define <2 x float> @v_fdiv_v2f32_arcp_ulp25(<2 x float> %a, <2 x float> %b) {
; GFX6-IEEE-LABEL: v_fdiv_v2f32_arcp_ulp25:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v5, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v4, v2
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v2|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v2|, v5
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v4, v2, v4, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v4, v4
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v2, v2
@@ -2435,7 +2435,7 @@ define <2 x float> @v_fdiv_v2f32_arcp_ulp25(<2 x float> %a, <2 x float> %b) {
; GFX6-IEEE-NEXT: v_ldexp_f32_e32 v2, v4, v2
; GFX6-IEEE-NEXT: v_mul_f32_e32 v0, v0, v2
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v3
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v3|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v3|, v5
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v3, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v3, v3
@@ -2863,15 +2863,15 @@ define float @v_fdiv_f32_dynamic_25ulp__nnan_ninf(float %x, float %y, float %z)
; GFX6-IEEE-LABEL: v_fdiv_f32_dynamic_25ulp__nnan_ninf:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v3, v0
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v4, v0
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v3
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v3, v0, v3, vcc
+; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v3, v0, v4, vcc
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-IEEE-NEXT: v_mul_f32_e32 v2, v3, v2
; GFX6-IEEE-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -2881,15 +2881,15 @@ define float @v_fdiv_f32_dynamic_25ulp__nnan_ninf(float %x, float %y, float %z)
; GFX6-FLUSH-LABEL: v_fdiv_f32_dynamic_25ulp__nnan_ninf:
; GFX6-FLUSH: ; %bb.0:
; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-FLUSH-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-FLUSH-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v3, v0
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v4, v0
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v3
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v3, v0, v3, vcc
+; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v3, v0, v4, vcc
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-FLUSH-NEXT: v_mul_f32_e32 v2, v3, v2
; GFX6-FLUSH-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -2983,15 +2983,15 @@ define float @v_fdiv_f32_dynamic_25ulp__nnan_ninf_contractable_user(float %x, fl
; GFX6-IEEE-LABEL: v_fdiv_f32_dynamic_25ulp__nnan_ninf_contractable_user:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v4, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v3, v1
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v4
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v3, v1, v3, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v3, v3
-; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v4, v0
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v5, v0
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v4
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v4, v0, v4, vcc
+; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v4, v0, v5, vcc
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-IEEE-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX6-IEEE-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -3002,15 +3002,15 @@ define float @v_fdiv_f32_dynamic_25ulp__nnan_ninf_contractable_user(float %x, fl
; GFX6-FLUSH-LABEL: v_fdiv_f32_dynamic_25ulp__nnan_ninf_contractable_user:
; GFX6-FLUSH: ; %bb.0:
; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-FLUSH-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-FLUSH-NEXT: v_mov_b32_e32 v4, 0x7f800000
; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v3, v1
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v4
; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v3, v1, v3, vcc
; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v3, v3
-; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v4, v0
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, s4
+; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v5, v0
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v0|, v4
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v4, v0, v4, vcc
+; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v4, v0, v5, vcc
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e32 v0, v0
; GFX6-FLUSH-NEXT: v_mul_f32_e32 v3, v4, v3
; GFX6-FLUSH-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -3391,15 +3391,15 @@ define float @v_fdiv_neglhs_f32_dynamic_25ulp(float %x, float %y) #0 {
; GFX6-IEEE-LABEL: v_fdiv_neglhs_f32_dynamic_25ulp:
; GFX6-IEEE: ; %bb.0:
; GFX6-IEEE-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-IEEE-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-IEEE-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-IEEE-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-IEEE-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-IEEE-NEXT: v_frexp_mant_f32_e64 v3, -v0
-; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 s[4:5], |v0|, s4
+; GFX6-IEEE-NEXT: v_frexp_mant_f32_e64 v4, -v0
+; GFX6-IEEE-NEXT: v_cmp_lt_f32_e64 s[4:5], |v0|, v3
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-IEEE-NEXT: v_cndmask_b32_e64 v3, -v0, v3, s[4:5]
+; GFX6-IEEE-NEXT: v_cndmask_b32_e64 v3, -v0, v4, s[4:5]
; GFX6-IEEE-NEXT: v_frexp_exp_i32_f32_e64 v0, -v0
; GFX6-IEEE-NEXT: v_mul_f32_e32 v2, v3, v2
; GFX6-IEEE-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -3409,15 +3409,15 @@ define float @v_fdiv_neglhs_f32_dynamic_25ulp(float %x, float %y) #0 {
; GFX6-FLUSH-LABEL: v_fdiv_neglhs_f32_dynamic_25ulp:
; GFX6-FLUSH: ; %bb.0:
; GFX6-FLUSH-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-FLUSH-NEXT: s_mov_b32 s4, 0x7f800000
+; GFX6-FLUSH-NEXT: v_mov_b32_e32 v3, 0x7f800000
; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e32 v2, v1
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, s4
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 vcc, |v1|, v3
; GFX6-FLUSH-NEXT: v_cndmask_b32_e32 v2, v1, v2, vcc
; GFX6-FLUSH-NEXT: v_rcp_f32_e32 v2, v2
-; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e64 v3, -v0
-; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 s[4:5], |v0|, s4
+; GFX6-FLUSH-NEXT: v_frexp_mant_f32_e64 v4, -v0
+; GFX6-FLUSH-NEXT: v_cmp_lt_f32_e64 s[4:5], |v0|, v3
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e32 v1, v1
-; GFX6-FLUSH-NEXT: v_cndmask_b32_e64 v3, -v0, v3, s[4:5]
+; GFX6-FLUSH-NEXT: v_cndmask_b32_e64 v3, -v0, v4, s[4:5]
; GFX6-FLUSH-NEXT: v_frexp_exp_i32_f32_e64 v0, -v0
; GFX6-FLUSH-NEXT: v_mul_f32_e32 v2, v3, v2
; GFX6-FLUSH-NEXT: v_sub_i32_e32 v0, vcc, v0, v1
@@ -3820,121 +3820,121 @@ define float @v_fdiv_f32_constrhs0_dynamic(float %x) #0 {
; GFX6-IEEE-FASTFMA-LABEL: v_fdiv_f32_constrhs0_dynamic:
; GFX6-IEEE-FASTFMA: ; %bb.0:
; GFX6-IEEE-FASTFMA-NEXT: s_waitcnt vmcnt(0) expcnt(0) lgkmcnt(0)
-; GFX6-IEEE-FASTFMA-NEXT: s_mov_b32 s6, 0x4640e400
-; GFX6-IEEE-FASTFMA-NEXT: v_div_scale_f32 v1, s[4:5], s6, s6, v0
-; GFX6-IEEE-FASTFMA-NEXT: v_rcp_f32_e32 v2, v1
-; GFX6-IEEE-FASTFMA-NEXT: v_div_scale_f32 v3, vcc, v0, s6, v0
+; GFX6-IEEE-FASTFMA-NEXT: v_mov_b32_e32 v1, 0x4640e400
+; GFX6-IEEE-FASTFMA-NEXT: v_div_scale_f32 v2, s[4:5], v1, v1, v0
+; GFX6-IEEE-FASTFMA-NEXT: v_rcp_f32_e32 v3, v2
+; GFX6-IEEE-FASTFMA-NEXT: v_div_scale_f32 v4, vcc, v0, v1, v0
; GFX6-IEEE-FASTFMA-NEXT: s_getreg_b32 s4, hwreg(HW_REG_MODE, 4, 2)
; GFX6-IEEE-FASTFMA-NEXT: s_setreg_imm32_b32 hwreg(HW_REG_MODE, 4, 2), 3
-; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v4, -v1, v2, 1.0
-; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v2, v4, v2, v2
-; GFX6-IEEE-FASTFMA-NEXT: v_mul_f32_e32 v4, v3, v2
-; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v5, -v1, v4, v3
-; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v4, v5, v2, v4
-; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v1, -v1, v4, v3
+; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v5, -v2, v3, 1.0
+; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v3, v5, v3, v3
+; GFX6-IEEE-FASTFMA-NEXT: v_mul_f32_e32 v5, v4, v3
+; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v6, -v2, v5, v4
+; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v5, v6, v3, v5
+; GFX6-IEEE-FASTFMA-NEXT: v_fma_f32 v2, -v2, v5, v4
; GFX6-IEEE-FASTFMA-NEXT: s_setreg_b32 hwreg(HW_REG_MODE, 4, 2), s4
-; GFX6-IEEE-FASTFMA-NEXT: v_div_fmas_f32 v1, v1, v2, v4
-; GFX6-IEEE-FASTFMA-NEXT: v_div_fixup_f32 v0, v1, s6, v0
+; GFX6-IEEE-FASTFMA-NEXT: v_div_fmas_f32 v2, v2, v3, v5
+; GFX6-IEEE-FASTFMA-N...
[truncated]
|
zahiraam
pushed a commit
to zahiraam/llvm-project
that referenced
this pull request
Nov 20, 2023
…selection. Operands Folding 1." (llvm#71710) This reverts commit 201f892.
zahiraam
pushed a commit
to zahiraam/llvm-project
that referenced
this pull request
Nov 20, 2023
The bug was fixed by llvm#71710.
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.
This reverts commit 201f892.