[PATCH] D69888: [AMDGPU][MC] Corrected src0 for v_movrelsd_b32 and v_movrelsd_2_b32
Dmitry Preobrazhensky via Phabricator via llvm-commits
llvm-commits at lists.llvm.org
Wed Nov 6 05:14:38 PST 2019
dp created this revision.
dp added reviewers: rampitec, arsenm.
Herald added subscribers: llvm-commits, hiraditya, t-tye, tpr, dstuttard, yaxunl, nhaehnle, wdng, jvesely, kzhuravl.
Herald added a project: LLVM.
These opcodes should not accept SGPRs as src0, only VGPRs are legal.
See https://bugs.llvm.org/show_bug.cgi?id=40903
https://reviews.llvm.org/D69888
Files:
llvm/lib/Target/AMDGPU/VOP1Instructions.td
llvm/test/MC/AMDGPU/gfx10_asm_all.s
llvm/test/MC/AMDGPU/vop1.s
Index: llvm/test/MC/AMDGPU/vop1.s
===================================================================
--- llvm/test/MC/AMDGPU/vop1.s
+++ llvm/test/MC/AMDGPU/vop1.s
@@ -273,6 +273,10 @@
// VI: v_movrelsd_b32_e32 v1, v2 ; encoding: [0x02,0x71,0x02,0x7e]
v_movrelsd_b32_e32 v1, v2
+// NOSICI: error: invalid operand for instruction
+// NOVI: error: invalid operand for instruction
+v_movrelsd_b32_e32 v1, s2
+
// NOSI: error: instruction not supported on this GPU
// NOSI: v_log_legacy_f32 v1, v2
// CI: v_log_legacy_f32_e32 v1, v2 ; encoding: [0x02,0x8b,0x02,0x7e]
Index: llvm/test/MC/AMDGPU/gfx10_asm_all.s
===================================================================
--- llvm/test/MC/AMDGPU/gfx10_asm_all.s
+++ llvm/test/MC/AMDGPU/gfx10_asm_all.s
@@ -32199,6 +32199,9 @@
v_movrelsd_b32_e64 v5, v255
// GFX10: encoding: [0x05,0x00,0xc4,0xd5,0xff,0x01,0x00,0x00]
+v_movrelsd_b32 v5, s1
+// GFX10-ERR: error: invalid operand for instruction
+
v_movrelsd_2_b32 v5, v1
// GFX10: encoding: [0x01,0x91,0x0a,0x7e]
@@ -32217,6 +32220,9 @@
v_movrelsd_2_b32_e64 v5, v255
// GFX10: encoding: [0x05,0x00,0xc8,0xd5,0xff,0x01,0x00,0x00]
+v_movrelsd_2_b32 v5, s1
+// GFX10-ERR: error: invalid operand for instruction
+
v_cvt_f16_u16_e32 v5, v1
// GFX10: encoding: [0x01,0xa1,0x0a,0x7e]
Index: llvm/lib/Target/AMDGPU/VOP1Instructions.td
===================================================================
--- llvm/lib/Target/AMDGPU/VOP1Instructions.td
+++ llvm/lib/Target/AMDGPU/VOP1Instructions.td
@@ -274,13 +274,13 @@
// to be a src operand. The custom inserter must add a tied implicit
// def and use of the super register since there seems to be no way to
// add an implicit def of a virtual register in tablegen.
-def VOP_MOVRELD : VOPProfile<[untyped, i32, untyped, untyped]> {
+class VOP_MOVREL<RegisterOperand Src1RC> : VOPProfile<[untyped, i32, untyped, untyped]> {
let Src0RC32 = VOPDstOperand<VGPR_32>;
let Src0RC64 = VOPDstOperand<VGPR_32>;
let Outs = (outs);
- let Ins32 = (ins Src0RC32:$vdst, VSrc_b32:$src0);
- let Ins64 = (ins Src0RC64:$vdst, VSrc_b32:$src0);
+ let Ins32 = (ins Src0RC32:$vdst, Src1RC:$src0);
+ let Ins64 = (ins Src0RC64:$vdst, Src1RC:$src0);
let InsDPP = (ins DstRC:$vdst, DstRC:$old, Src0RC32:$src0,
dpp_ctrl:$dpp_ctrl, row_mask:$row_mask,
bank_mask:$bank_mask, bound_ctrl:$bound_ctrl);
@@ -306,6 +306,9 @@
let EmitDst = 1; // force vdst emission
}
+def VOP_MOVRELD : VOP_MOVREL<VSrc_b32>;
+def VOP_MOVRELSD : VOP_MOVREL<VRegSrc_32>;
+
let SubtargetPredicate = HasMovrel, Uses = [M0, EXEC] in {
// v_movreld_b32 is a special case because the destination output
// register is really a source. It isn't actually read (but may be
@@ -315,7 +318,7 @@
// so this must have an implicit def of the register added to it.
defm V_MOVRELD_B32 : VOP1Inst <"v_movreld_b32", VOP_MOVRELD>;
defm V_MOVRELS_B32 : VOP1Inst <"v_movrels_b32", VOP_I32_VI32_NO_EXT>;
-defm V_MOVRELSD_B32 : VOP1Inst <"v_movrelsd_b32", VOP_NO_EXT<VOP_I32_I32>>;
+defm V_MOVRELSD_B32 : VOP1Inst <"v_movrelsd_b32", VOP_MOVRELSD>;
} // End Uses = [M0, EXEC]
defm V_MOV_FED_B32 : VOP1Inst <"v_mov_fed_b32", VOP_I32_I32>;
@@ -430,9 +433,8 @@
defm V_PIPEFLUSH : VOP1Inst<"v_pipeflush", VOP_NONE>;
let Uses = [M0] in {
- // FIXME-GFX10: Should V_MOVRELSD_2_B32 be VOP_NO_EXT?
defm V_MOVRELSD_2_B32 :
- VOP1Inst<"v_movrelsd_2_b32", VOP_NO_EXT<VOP_I32_I32>>;
+ VOP1Inst<"v_movrelsd_2_b32", VOP_MOVRELSD>;
def V_SWAPREL_B32 : VOP1_Pseudo<"v_swaprel_b32", VOP_SWAP_I32, [], 1> {
let Constraints = "$vdst = $src1, $vdst1 = $src0";
-------------- next part --------------
A non-text attachment was scrubbed...
Name: D69888.228043.patch
Type: text/x-patch
Size: 3659 bytes
Desc: not available
URL: <http://lists.llvm.org/pipermail/llvm-commits/attachments/20191106/64d867e7/attachment.bin>
More information about the llvm-commits
mailing list