[AMDGPU] Fix encoding of VOP3P dpp on GFX11 and GFX12 #82710

rampitec · 2024-02-23T00:10:31Z

The bug affects dpp forms of v_dot2_f32_f16. The encoding does not match SP3 and does not set op_sel_hi bits properly.

llvmbot · 2024-02-23T00:11:05Z

@llvm/pr-subscribers-backend-amdgpu

Author: Stanislav Mekhanoshin (rampitec)

Changes

The bug affects dpp forms of v_dot2_f32_f16. The encoding does not match SP3 and does not set op_sel_hi bits properly.

Full diff: https://github.com/llvm/llvm-project/pull/82710.diff

11 Files Affected:

(modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+2)
(modified) llvm/lib/Target/AMDGPU/VOPInstructions.td (+1)
(modified) llvm/test/MC/AMDGPU/gfx11-promotions.s (+4-4)
(modified) llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s (+2-2)
(modified) llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s (+1-1)
(modified) llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s (+2-2)
(modified) llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s (+1-1)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt (+4-4)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt (+2-2)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt (+4-4)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt (+2-2)

diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index cf76de40aef412..ac3c8f95306bc9 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -1353,6 +1353,7 @@ class VOP3P_DPP16<bits<7> op, VOP_DPP_Pseudo ps, int subtarget,
   let AssemblerPredicate = HasDPP16;
   let SubtargetPredicate = HasDPP16;
   let OtherPredicates = ps.OtherPredicates;
+  let IsPacked = ps.IsPacked;
 }
 
 class VOP3P_DPP8_Base<bits<7> op, VOP_Pseudo ps, string opName = ps.OpName>
@@ -1362,6 +1363,7 @@ class VOP3P_DPP8_Base<bits<7> op, VOP_Pseudo ps, string opName = ps.OpName>
   let SchedRW = ps.SchedRW;
   let Uses = ps.Uses;
   let OtherPredicates = ps.OtherPredicates;
+  let IsPacked = ps.IsPacked;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/VOPInstructions.td b/llvm/lib/Target/AMDGPU/VOPInstructions.td
index 2989d05e968ef9..80d7d96a5e3cc8 100644
--- a/llvm/lib/Target/AMDGPU/VOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOPInstructions.td
@@ -818,6 +818,7 @@ class VOP_DPP_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[],
   let VALU = 1;
   let DPP = 1;
   let Size = 8;
+  let IsPacked = P.IsPacked;
 
   let ReadsModeReg = !or(P.DstVT.isFP, P.Src0VT.isFP);
 
diff --git a/llvm/test/MC/AMDGPU/gfx11-promotions.s b/llvm/test/MC/AMDGPU/gfx11-promotions.s
index 0bd90266457ee0..67e7beaa262f47 100644
--- a/llvm/test/MC/AMDGPU/gfx11-promotions.s
+++ b/llvm/test/MC/AMDGPU/gfx11-promotions.s
@@ -337,17 +337,17 @@ v_dot2_f32_f16_e64 v0, v1, v2, v3
 //===----------------------------------------------------------------------===//
 
 v_dot2_f32_f16 v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x40,0x13,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x77,0x39,0x05]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x40,0x13,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x77,0x39,0x05]
 
 //===----------------------------------------------------------------------===//
 // VOP3P.DPP16.
 //===----------------------------------------------------------------------===//
 
 v_dot2_f32_f16 v0, v1, v2, v3 quad_perm:[1,2,3,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s
index 2cfb8abd4e9790..3ff4ed27f1b256 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s
@@ -2,10 +2,10 @@
 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefixes=GFX11 %s
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[0,0,0] neg_hi:[0,0,0] quad_perm:[2,2,3,1] bound_ctrl:0 fi:1
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x04,0xff]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] bank_mask:0xe
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
 
 v_fma_mix_f32 v0, v1, v2, v3 op_sel:[0,0,0] row_ror:7 bank_mask:0x1 bound_ctrl:0
 // GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x01,0xf1]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s
index 2656ba0cf1807e..3fb993dc8bec48 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s
@@ -15,4 +15,4 @@ v_fma_mixlo_f16 v0, abs(v1), -v2, abs(v3) op_sel:[1,0,0] op_sel_hi:[1,0,0] dpp8:
 // GFX11: encoding: [0x00,0x0d,0x21,0xcc,0xe9,0x04,0x0e,0x4c,0x01,0x92,0x44,0x92]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
+// GFX11: encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s
index 75bd1696e10bb7..a6360684f1d0e3 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s
@@ -2,10 +2,10 @@
 // RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefixes=GFX12 %s
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[0,0,0] neg_hi:[0,0,0] quad_perm:[2,2,3,1] bound_ctrl:0 fi:1
-// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x04,0xff]
+// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] bank_mask:0xe
-// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
+// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
 
 v_fma_mix_f32 v0, v1, v2, v3 op_sel:[0,0,0] row_ror:7 bank_mask:0x1 bound_ctrl:0
 // GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x01,0xf1]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s
index 14cf169d4b4249..299339339e8c66 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s
@@ -15,7 +15,7 @@ v_fma_mixlo_f16 v0, abs(v1), -v2, abs(v3) op_sel:[1,0,0] op_sel_hi:[1,0,0] dpp8:
 // GFX12: encoding: [0x00,0x0d,0x21,0xcc,0xe9,0x04,0x0e,0x4c,0x01,0x92,0x44,0x92]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
+// GFX12: encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
 
 v_dot4_f32_fp8_bf8 v0, v1, v2, v3 dpp8:[0,1,2,3,4,5,6,7]
 // GFX12: v_dot4_f32_fp8_bf8_e64_dpp v0, v1, v2, v3 dpp8:[0,1,2,3,4,5,6,7] ; encoding: [0x00,0x40,0x24,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x88,0xc6,0xfa]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt
index 6b230367c8313f..ceca6d9fc3faa7 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt
@@ -1,11 +1,11 @@
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-WavefrontSize32,+WavefrontSize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 
-# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
-0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe
+# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
+0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
 
-# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff]
-0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff
+# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff]
+0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
 
 # GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 bound_ctrl:1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1]
 0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt
index 89c9b54d7cfeec..57c96170eadce0 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt
@@ -1,8 +1,8 @@
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-WavefrontSize32,+WavefrontSize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 
-# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
-0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05
+# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
+0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05
 
 # GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 clamp dpp8:[2,2,2,2,4,4,4,4] fi:1 ; encoding: [0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92]
 0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt
index 52fd0530681cf1..10f438465d65e1 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt
@@ -1,11 +1,11 @@
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 
-# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
-0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe
+# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
+0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
 
-# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff]
-0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff
+# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff]
+0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
 
 # GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 bound_ctrl:1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1]
 0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt
index 688212e51c427d..2fb9c23ed5ec5b 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt
@@ -1,8 +1,8 @@
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 
-# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
-0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05
+# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
+0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05
 
 # GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 clamp dpp8:[2,2,2,2,4,4,4,4] fi:1 ; encoding: [0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92]
 0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92

llvmbot · 2024-02-23T00:11:05Z

@llvm/pr-subscribers-mc

Author: Stanislav Mekhanoshin (rampitec)

Changes

The bug affects dpp forms of v_dot2_f32_f16. The encoding does not match SP3 and does not set op_sel_hi bits properly.

Full diff: https://github.com/llvm/llvm-project/pull/82710.diff

11 Files Affected:

(modified) llvm/lib/Target/AMDGPU/VOP3PInstructions.td (+2)
(modified) llvm/lib/Target/AMDGPU/VOPInstructions.td (+1)
(modified) llvm/test/MC/AMDGPU/gfx11-promotions.s (+4-4)
(modified) llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s (+2-2)
(modified) llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s (+1-1)
(modified) llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s (+2-2)
(modified) llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s (+1-1)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt (+4-4)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt (+2-2)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt (+4-4)
(modified) llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt (+2-2)

diff --git a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
index cf76de40aef412..ac3c8f95306bc9 100644
--- a/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOP3PInstructions.td
@@ -1353,6 +1353,7 @@ class VOP3P_DPP16<bits<7> op, VOP_DPP_Pseudo ps, int subtarget,
   let AssemblerPredicate = HasDPP16;
   let SubtargetPredicate = HasDPP16;
   let OtherPredicates = ps.OtherPredicates;
+  let IsPacked = ps.IsPacked;
 }
 
 class VOP3P_DPP8_Base<bits<7> op, VOP_Pseudo ps, string opName = ps.OpName>
@@ -1362,6 +1363,7 @@ class VOP3P_DPP8_Base<bits<7> op, VOP_Pseudo ps, string opName = ps.OpName>
   let SchedRW = ps.SchedRW;
   let Uses = ps.Uses;
   let OtherPredicates = ps.OtherPredicates;
+  let IsPacked = ps.IsPacked;
 }
 
 //===----------------------------------------------------------------------===//
diff --git a/llvm/lib/Target/AMDGPU/VOPInstructions.td b/llvm/lib/Target/AMDGPU/VOPInstructions.td
index 2989d05e968ef9..80d7d96a5e3cc8 100644
--- a/llvm/lib/Target/AMDGPU/VOPInstructions.td
+++ b/llvm/lib/Target/AMDGPU/VOPInstructions.td
@@ -818,6 +818,7 @@ class VOP_DPP_Pseudo <string OpName, VOPProfile P, list<dag> pattern=[],
   let VALU = 1;
   let DPP = 1;
   let Size = 8;
+  let IsPacked = P.IsPacked;
 
   let ReadsModeReg = !or(P.DstVT.isFP, P.Src0VT.isFP);
 
diff --git a/llvm/test/MC/AMDGPU/gfx11-promotions.s b/llvm/test/MC/AMDGPU/gfx11-promotions.s
index 0bd90266457ee0..67e7beaa262f47 100644
--- a/llvm/test/MC/AMDGPU/gfx11-promotions.s
+++ b/llvm/test/MC/AMDGPU/gfx11-promotions.s
@@ -337,17 +337,17 @@ v_dot2_f32_f16_e64 v0, v1, v2, v3
 //===----------------------------------------------------------------------===//
 
 v_dot2_f32_f16 v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x40,0x13,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x77,0x39,0x05]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x00,0x13,0xcc,0xe9,0x04,0x0e,0x04,0x01,0x77,0x39,0x05]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x40,0x13,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x77,0x39,0x05]
 
 //===----------------------------------------------------------------------===//
 // VOP3P.DPP16.
 //===----------------------------------------------------------------------===//
 
 v_dot2_f32_f16 v0, v1, v2, v3 quad_perm:[1,2,3,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0]
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x39,0x00,0xff]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[1,2,3,0] row_mask:0xf bank_mask:0xf ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x39,0x00,0xff]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s
index 2cfb8abd4e9790..3ff4ed27f1b256 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s
@@ -2,10 +2,10 @@
 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefixes=GFX11 %s
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[0,0,0] neg_hi:[0,0,0] quad_perm:[2,2,3,1] bound_ctrl:0 fi:1
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x04,0xff]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] bank_mask:0xe
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
 
 v_fma_mix_f32 v0, v1, v2, v3 op_sel:[0,0,0] row_ror:7 bank_mask:0x1 bound_ctrl:0
 // GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x01,0xf1]
diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s
index 2656ba0cf1807e..3fb993dc8bec48 100644
--- a/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp8.s
@@ -15,4 +15,4 @@ v_fma_mixlo_f16 v0, abs(v1), -v2, abs(v3) op_sel:[1,0,0] op_sel_hi:[1,0,0] dpp8:
 // GFX11: encoding: [0x00,0x0d,0x21,0xcc,0xe9,0x04,0x0e,0x4c,0x01,0x92,0x44,0x92]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0]
-// GFX11: encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
+// GFX11: encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s
index 75bd1696e10bb7..a6360684f1d0e3 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp16.s
@@ -2,10 +2,10 @@
 // RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefixes=GFX12 %s
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[0,0,0] neg_hi:[0,0,0] quad_perm:[2,2,3,1] bound_ctrl:0 fi:1
-// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x04,0xff]
+// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]
 
 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] bank_mask:0xe
-// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
+// GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
 
 v_fma_mix_f32 v0, v1, v2, v3 op_sel:[0,0,0] row_ror:7 bank_mask:0x1 bound_ctrl:0
 // GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x01,0xf1]
diff --git a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s
index 14cf169d4b4249..299339339e8c66 100644
--- a/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s
+++ b/llvm/test/MC/AMDGPU/gfx12_asm_vop3p_dpp8.s
@@ -15,7 +15,7 @@ v_fma_mixlo_f16 v0, abs(v1), -v2, abs(v3) op_sel:[1,0,0] op_sel_hi:[1,0,0] dpp8:
 // GFX12: encoding: [0x00,0x0d,0x21,0xcc,0xe9,0x04,0x0e,0x4c,0x01,0x92,0x44,0x92]
 
 v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0]
-// GFX12: encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
+// GFX12: encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
 
 v_dot4_f32_fp8_bf8 v0, v1, v2, v3 dpp8:[0,1,2,3,4,5,6,7]
 // GFX12: v_dot4_f32_fp8_bf8_e64_dpp v0, v1, v2, v3 dpp8:[0,1,2,3,4,5,6,7] ; encoding: [0x00,0x40,0x24,0xcc,0xe9,0x04,0x0e,0x1c,0x01,0x88,0xc6,0xfa]
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt
index 6b230367c8313f..ceca6d9fc3faa7 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp16.txt
@@ -1,11 +1,11 @@
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-WavefrontSize32,+WavefrontSize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 
-# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
-0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe
+# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
+0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
 
-# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff]
-0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff
+# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff]
+0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
 
 # GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 bound_ctrl:1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1]
 0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt
index 89c9b54d7cfeec..57c96170eadce0 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_vop3p_dpp8.txt
@@ -1,8 +1,8 @@
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 # RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-WavefrontSize32,+WavefrontSize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX11 %s
 
-# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
-0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05
+# GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
+0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05
 
 # GFX11: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 clamp dpp8:[2,2,2,2,4,4,4,4] fi:1 ; encoding: [0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92]
 0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt
index 52fd0530681cf1..10f438465d65e1 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp16.txt
@@ -1,11 +1,11 @@
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 
-# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe]
-0x00,0x05,0x13,0xcc,0xfa,0x04,0x0e,0x64,0x01,0x1b,0x00,0xfe
+# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[1,1,0] neg_hi:[1,0,1] quad_perm:[3,2,1,0] row_mask:0xf bank_mask:0xe ; encoding: [0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe]
+0x00,0x45,0x13,0xcc,0xfa,0x04,0x0e,0x7c,0x01,0x1b,0x00,0xfe
 
-# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff]
-0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x0c,0xff
+# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf bound_ctrl:1 fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff]
+0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x0c,0xff
 
 # GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 row_ror:7 row_mask:0xf bank_mask:0x1 bound_ctrl:1 ; encoding: [0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1]
 0x00,0x00,0x20,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x27,0x09,0xf1
diff --git a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt
index 688212e51c427d..2fb9c23ed5ec5b 100644
--- a/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/gfx12_dasm_vop3p_dpp8.txt
@@ -1,8 +1,8 @@
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx1200 -mattr=-wavefrontsize32,+wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=GFX12 %s
 
-# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05]
-0x00,0x05,0x13,0xcc,0xe9,0x04,0x0e,0xc4,0x01,0x77,0x39,0x05
+# GFX12: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 neg_lo:[0,1,1] neg_hi:[1,0,1] dpp8:[7,6,5,4,3,2,1,0] ; encoding: [0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05]
+0x00,0x45,0x13,0xcc,0xe9,0x04,0x0e,0xdc,0x01,0x77,0x39,0x05
 
 # GFX12: v_fma_mix_f32_e64_dpp v0, v1, v2, v3 clamp dpp8:[2,2,2,2,4,4,4,4] fi:1 ; encoding: [0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92]
 0x00,0x80,0x20,0xcc,0xea,0x04,0x0e,0x04,0x01,0x92,0x44,0x92

rampitec · 2024-02-23T09:48:19Z

llvm/test/MC/AMDGPU/gfx11_asm_vop3p_dpp16.s

@@ -2,10 +2,10 @@
 // RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=-wavefrontsize32,+wavefrontsize64 -show-encoding %s | FileCheck --check-prefixes=GFX11 %s

 v_dot2_f32_f16 v0, v1, v2, v3 neg_lo:[0,0,0] neg_hi:[0,0,0] quad_perm:[2,2,3,1] bound_ctrl:0 fi:1
-// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x00,0x13,0xcc,0xfa,0x04,0x0e,0x04,0x01,0x7a,0x04,0xff]
+// GFX11: v_dot2_f32_f16_e64_dpp v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1 ; encoding: [0x00,0x40,0x13,0xcc,0xfa,0x04,0x0e,0x1c,0x01,0x7a,0x04,0xff]


So that reviewers can judge, SP3 of v_dot2_f32_f16 v0, v1, v2, v3 quad_perm:[2,2,3,1] row_mask:0xf bank_mask:0xf fi:1:

0xcc134000 0x1c0e04fa 0xff047a01 Fields: CLAMP 0x0 ENC_VOP3P[ 15: 15] COMP_OP_SEL_HI 0x1 ENC_VOP3P[ -1: -1] ENCODING 0xcc ENC_VOP3P[ 31: 24] NEG 0x0 ENC_VOP3P[ 63: 61] NEG_HI 0x0 ENC_VOP3P[ 10: 8] OP 0x13 ENC_VOP3P[ 22: 16] OP_SEL 0x0 ENC_VOP3P[ 13: 11] OP_SEL_HI 0x3 ENC_VOP3P[ 60: 59] OP_SEL_HI_2 0x1 ENC_VOP3P[ 14: 14] SRC0 0xfa ENC_VOP3P[ 40: 32] SRC1 0x102 ENC_VOP3P[ 49: 41] SRC2 0x103 ENC_VOP3P[ 58: 50] VDST 0x0 ENC_VOP3P[ 7: 0]

Not saying that neg_lo:[0,0,0] neg_hi:[0,0,0] is a useless test.

And this is the decode of what llvm-mc has produced before the patch out of these bits:

Decode: v_dot2_f32_f16 v0, sel_lo(v1), sel_lo(v2), sel_lo(v3) quad_perm:[2,2,3,1] fi:1 // 000000000000: CC130000 040E04FA FF047A01 Fields: CLAMP 0x0 ENC_VOP3P[ 15: 15] COMP_OP_SEL_HI 0x1 ENC_VOP3P[ -1: -1] ENCODING 0xcc ENC_VOP3P[ 31: 24] NEG 0x0 ENC_VOP3P[ 63: 61] NEG_HI 0x0 ENC_VOP3P[ 10: 8] OP 0x13 ENC_VOP3P[ 22: 16] OP_SEL 0x0 ENC_VOP3P[ 13: 11] OP_SEL_HI 0x0 ENC_VOP3P[ 60: 59] OP_SEL_HI_2 0x0 ENC_VOP3P[ 14: 14] SRC0 0xfa ENC_VOP3P[ 40: 32] SRC1 0x102 ENC_VOP3P[ 49: 41] SRC2 0x103 ENC_VOP3P[ 58: 50] VDST 0x0 ENC_VOP3P[ 7: 0]

rampitec · 2024-02-23T10:45:14Z

Then the root of the problem is that VOP3P had no DPP when the code was written, so GFX9 code is OK and e64 encoding is correct. Then it was not properly verified when DPP support was added in GFX11.

rampitec · 2024-02-23T11:39:12Z

As a side note: I have written scripts to produce MC tests. But I am verifying produced bits with my naked eyes before I submit these script results. Yes, my eyes are bleeding. This is the path.

jayfoad

I do not have a deep understanding of VOP3P DPP encodings. But the code change seems obvious, and I spot-checked some of the test diffs against SP3.

rampitec · 2024-02-23T11:49:11Z

I do not have a deep understanding of VOP3P DPP encodings. But the code change seems obvious, and I spot-checked some of the test diffs against SP3.

It is not different in the 3-rd dword, which is DPP. It is just DPP looses bits in the low 64-bits.

Sisyph · 2024-02-23T14:57:17Z

Nice job spotting it! I found an old email where I was confused about this exact topic, whether op_sel_hi should be set for these bits on this instruction, but never resolved it. Thanks.

Sisyph · 2024-02-23T15:02:01Z

Do we need a change for v_dot2_f32_bf16? and src2 is f32. I will check it a bit more

Sisyph · 2024-02-23T16:48:40Z

I don't think dpp is a special case. This is how it should work for both vop3p and vop3p with dpp encoding:

https://www.amd.com/content/dam/amd/en/documents/radeon-tech-docs/instruction-set-architectures/rdna3-shader-instruction-set-architecture-feb-2023_0.pdf

From the packed math section 7.5 of the programming guide table
"If either the source operand or destination operand is 32bits, the corresponding OPSEL bit must set
to zero."
"If either the source operand or destination operand is 32bits or is a constant, the corresponding OPSEL_HI
bit must set to zero"
For v_dot2_F32_f16 and v_dot2_f32_bf16 src2 is an f32. Therefore the OPSEL_HI_2 bit of these should be 0.

For op_sel and op_sel_hi for src0 and src1, the only description I see for what those bits will do for these instructions is for inline constants in section 7.5.1.
DOT2_F32_BF16 says it ignores OPSEL, so it doesn't matter.
DOT2_F32_F16 says "use FP32 inline, supports OPSEL"
So I think the baseline vop3p behavior you have enabled here for those operands is correct.

rampitec · 2024-02-23T20:45:14Z

I don't think dpp is a special case. This is how it should work for both vop3p and vop3p with dpp encoding:

https://www.amd.com/content/dam/amd/en/documents/radeon-tech-docs/instruction-set-architectures/rdna3-shader-instruction-set-architecture-feb-2023_0.pdf

From the packed math section 7.5 of the programming guide table "If either the source operand or destination operand is 32bits, the corresponding OPSEL bit must set to zero." "If either the source operand or destination operand is 32bits or is a constant, the corresponding OPSEL_HI bit must set to zero" For v_dot2_F32_f16 and v_dot2_f32_bf16 src2 is an f32. Therefore the OPSEL_HI_2 bit of these should be 0.

For op_sel and op_sel_hi for src0 and src1, the only description I see for what those bits will do for these instructions is for inline constants in section 7.5.1. DOT2_F32_BF16 says it ignores OPSEL, so it doesn't matter. DOT2_F32_F16 says "use FP32 inline, supports OPSEL" So I think the baseline vop3p behavior you have enabled here for those operands is correct.

It is the same as SP3 and it is the same as base VOP3P without DPP. That is just DPP was missing these bits.

[AMDGPU] Fix encoding on VOP3P dpp on GFX11 and GFX12

c33dd47

The bug affects dpp forms of v_dot2_f32_f16. The encoding does not match SP3 and does not set op_sel_hi bits properly.

rampitec requested review from jayfoad and Sisyph February 23, 2024 00:10

llvmbot added backend:AMDGPU mc Machine (object) code labels Feb 23, 2024

rampitec requested a review from kzhuravl February 23, 2024 00:59

jayfoad requested review from kosarev and mbrkusanin February 23, 2024 09:12

rampitec commented Feb 23, 2024

View reviewed changes

rampitec changed the title ~~[AMDGPU] Fix encoding on VOP3P dpp on GFX11 and GFX12~~ [AMDGPU] Fix encoding of VOP3P dpp on GFX11 and GFX12 Feb 23, 2024

Merge branch 'main' into vop3p-dpp-encoding

f4851ee

rampitec requested a review from changpeng February 23, 2024 11:45

jayfoad approved these changes Feb 23, 2024

View reviewed changes

rampitec merged commit 3dfca24 into llvm:main Feb 23, 2024
3 of 4 checks passed

rampitec deleted the vop3p-dpp-encoding branch February 23, 2024 11:50

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

[AMDGPU] Fix encoding of VOP3P dpp on GFX11 and GFX12 #82710

[AMDGPU] Fix encoding of VOP3P dpp on GFX11 and GFX12 #82710

rampitec commented Feb 23, 2024

llvmbot commented Feb 23, 2024

llvmbot commented Feb 23, 2024

rampitec Feb 23, 2024

rampitec Feb 23, 2024

rampitec Feb 23, 2024

rampitec commented Feb 23, 2024

rampitec commented Feb 23, 2024

jayfoad left a comment

rampitec commented Feb 23, 2024

Sisyph commented Feb 23, 2024

Sisyph commented Feb 23, 2024

Sisyph commented Feb 23, 2024

rampitec commented Feb 23, 2024

[AMDGPU] Fix encoding of VOP3P dpp on GFX11 and GFX12 #82710

[AMDGPU] Fix encoding of VOP3P dpp on GFX11 and GFX12 #82710

Conversation

rampitec commented Feb 23, 2024

llvmbot commented Feb 23, 2024

llvmbot commented Feb 23, 2024

rampitec Feb 23, 2024

Choose a reason for hiding this comment

rampitec Feb 23, 2024

Choose a reason for hiding this comment

rampitec Feb 23, 2024

Choose a reason for hiding this comment

rampitec commented Feb 23, 2024

rampitec commented Feb 23, 2024

jayfoad left a comment

Choose a reason for hiding this comment

rampitec commented Feb 23, 2024

Sisyph commented Feb 23, 2024

Sisyph commented Feb 23, 2024

Sisyph commented Feb 23, 2024

rampitec commented Feb 23, 2024