[AMDGPU] gfx11 EXECZ and VCCZ are no longer allowed to be used as

sources to SALU and VALU instructions.

Contributors:
Baptiste Saleil <baptiste.saleil@amd.com>

Patch 20/N for upstreaming of AMDGPU gfx11 architecture

Depends on D126989

Reviewed By: rampitec, foad, #amdgpu

Differential Revision: https://reviews.llvm.org/D127143

diff --git a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
index 7b52d3e..cf9eeb4 100644 (file)
--- a/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
+++ b/llvm/lib/Target/AMDGPU/AsmParser/AMDGPUAsmParser.cpp
@@ -1650,6 +1650,7 @@ private:
                              const SMLoc &IDLoc);
   bool validateFlatLdsDMA(const MCInst &Inst, const OperandVector &Operands,
                           const SMLoc &IDLoc);
+  bool validateExeczVcczOperands(const OperandVector &Operands);
   Optional<StringRef> validateLdsDirect(const MCInst &Inst);
   unsigned getConstantBusLimit(unsigned Opcode) const;
   bool usesConstantBus(const MCInst &Inst, unsigned OpIdx);
@@ -4506,6 +4507,22 @@ bool AMDGPUAsmParser::validateFlatLdsDMA(const MCInst &Inst,
   return true;
 }
 
+bool AMDGPUAsmParser::validateExeczVcczOperands(const OperandVector &Operands) {
+  if (!isGFX11Plus())
+    return true;
+  for (auto &Operand : Operands) {
+    if (!Operand->isReg())
+      continue;
+    unsigned Reg = Operand->getReg();
+    if (Reg == SRC_EXECZ || Reg == SRC_VCCZ) {
+      Error(getRegLoc(Reg, Operands),
+            "execz and vccz are not supported on this GPU");
+      return false;
+    }
+  }
+  return true;
+}
+
 bool AMDGPUAsmParser::validateInstruction(const MCInst &Inst,
                                           const SMLoc &IDLoc,
                                           const OperandVector &Operands) {
@@ -4620,6 +4637,9 @@ bool AMDGPUAsmParser::validateInstruction(const MCInst &Inst,
   if (!validateCoherencyBits(Inst, Operands, IDLoc)) {
     return false;
   }
+  if (!validateExeczVcczOperands(Operands)) {
+    return false;
+  }
 
   if (!validateFlatLdsDMA(Inst, Operands, IDLoc)) {
     return false;

diff --git a/llvm/test/MC/AMDGPU/gfx10-constant-bus.s b/llvm/test/MC/AMDGPU/gfx10-constant-bus.s
index 37e853c..c28a167 100644 (file)
--- a/llvm/test/MC/AMDGPU/gfx10-constant-bus.s
+++ b/llvm/test/MC/AMDGPU/gfx10-constant-bus.s
@@ -1,63 +1,72 @@
-// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1010 -show-encoding %s | FileCheck -check-prefix=GFX10 %s
-// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1010 %s 2>&1 | FileCheck -check-prefix=GFX10-ERR --implicit-check-not=error: %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1010 -show-encoding %s | FileCheck -check-prefixes=GCN,GFX10 %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1010 %s 2>&1 | FileCheck -check-prefixes=GCN-ERR --implicit-check-not=error: %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 -show-encoding %s | FileCheck -check-prefixes=GCN,GFX11 %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 %s 2>&1 | FileCheck -check-prefixes=GCN-ERR,GFX11-ERR --implicit-check-not=error: %s
 
 //-----------------------------------------------------------------------------------------
 // On GFX10 we can use two scalar operands (except for 64-bit shift instructions)
 
 v_add_f32 v0, s0, s1
-// GFX10: v_add_f32_e64 v0, s0, s1        ; encoding: [0x00,0x00,0x03,0xd5,0x00,0x02,0x00,0x00]
+// GCN: v_add_f32_e64 v0, s0, s1        ; encoding: [0x00,0x00,0x03,0xd5,0x00,0x02,0x00,0x00]
 
 v_madak_f32 v0, s0, v1, 42.42
 // GFX10: v_madak_f32 v0, s0, v1, 0x4229ae14 ; encoding: [0x00,0x02,0x00,0x42,0x14,0xae,0x29,0x42]
+// GFX11-ERR: error: instruction not supported on this GPU
 
 v_med3_f32 v0, s0, s0, s1
 // GFX10: v_med3_f32 v0, s0, s0, s1       ; encoding: [0x00,0x00,0x57,0xd5,0x00,0x00,0x04,0x00]
+// GFX11: v_med3_f32 v0, s0, s0, s1       ; encoding: [0x00,0x00,0x1f,0xd6,0x00,0x00,0x04,0x00]
 
 //-----------------------------------------------------------------------------------------
 // 64-bit shift instructions can use only one scalar value input
 
 v_ashrrev_i64 v[0:1], 0x100, s[0:1]
-// GFX10-ERR: error: invalid operand (violates constant bus restrictions)
+// GCN-ERR: error: invalid operand (violates constant bus restrictions)
 
 v_ashrrev_i64 v[0:1], s2, s[0:1]
-// GFX10-ERR: error: invalid operand (violates constant bus restrictions)
+// GCN-ERR: error: invalid operand (violates constant bus restrictions)
 
 //-----------------------------------------------------------------------------------------
 // v_div_fmas implicitly reads VCC, so only one scalar operand is possible
 
 v_div_fmas_f32 v5, s3, s3, s3
 // GFX10: v_div_fmas_f32 v5, s3, s3, s3   ; encoding: [0x05,0x00,0x6f,0xd5,0x03,0x06,0x0c,0x00]
+// GFX11: v_div_fmas_f32 v5, s3, s3, s3   ; encoding: [0x05,0x00,0x37,0xd6,0x03,0x06,0x0c,0x00]
 
 v_div_fmas_f32 v5, s3, s3, s2
-// GFX10-ERR: error: invalid operand (violates constant bus restrictions)
+// GCN-ERR: error: invalid operand (violates constant bus restrictions)
 
 v_div_fmas_f32 v5, s3, 0x123, v3
-// GFX10-ERR: error: invalid operand (violates constant bus restrictions)
+// GCN-ERR: error: invalid operand (violates constant bus restrictions)
 
 v_div_fmas_f64 v[5:6], 0x12345678, 0x12345678, 0x12345678
 // GFX10: v_div_fmas_f64 v[5:6], 0x12345678, 0x12345678, 0x12345678 ; encoding: [0x05,0x00,0x70,0xd5,0xff,0xfe,0xfd,0x03,0x78,0x56,0x34,0x12]
+// GFX11: v_div_fmas_f64 v[5:6], 0x12345678, 0x12345678, 0x12345678 ; encoding: [0x05,0x00,0x38,0xd6,0xff,0xfe,0xfd,0x03,0x78,0x56,0x34,0x12]
 
 v_div_fmas_f64 v[5:6], v[1:2], s[2:3], v[3:4]
 // GFX10: v_div_fmas_f64 v[5:6], v[1:2], s[2:3], v[3:4] ; encoding: [0x05,0x00,0x70,0xd5,0x01,0x05,0x0c,0x04]
+// GFX11: v_div_fmas_f64 v[5:6], v[1:2], s[2:3], v[3:4] ; encoding: [0x05,0x00,0x38,0xd6,0x01,0x05,0x0c,0x04]
 
 v_div_fmas_f64 v[5:6], v[1:2], s[2:3], 0x123456
-// GFX10-ERR: error: invalid operand (violates constant bus restrictions)
+// GCN-ERR: error: invalid operand (violates constant bus restrictions)
 
 //-----------------------------------------------------------------------------------------
 // v_mad_u64_u32 has operands of different sizes.
 // When these operands are literals, they are counted as 2 scalar values even if literals are identical.
 
 v_lshlrev_b64 v[5:6], 0x3f717273, 0x3f717273
-// GFX10-ERR: error: invalid operand (violates constant bus restrictions)
+// GCN-ERR: error: invalid operand (violates constant bus restrictions)
 
 v_mad_u64_u32 v[5:6], s12, v1, 0x12345678, 0x12345678
 // GFX10: v_mad_u64_u32 v[5:6], s12, v1, 0x12345678, 0x12345678 ; encoding: [0x05,0x0c,0x76,0xd5,0x01,0xff,0xfd,0x03,0x78,0x56,0x34,0x12]
+// GFX11: v_mad_u64_u32 v[5:6], s12, v1, 0x12345678, 0x12345678 ; encoding: [0x05,0x0c,0xfe,0xd6,0x01,0xff,0xfd,0x03,0x78,0x56,0x34,0x12]
 
 v_mad_u64_u32 v[5:6], s12, s1, 0x12345678, 0x12345678
-// GFX10-ERR: error: invalid operand (violates constant bus restrictions)
+// GCN-ERR: error: invalid operand (violates constant bus restrictions)
 
 //-----------------------------------------------------------------------------------------
 // null is free
 
 v_bfe_u32 v5, s1, s2, null
 // GFX10: v_bfe_u32 v5, s1, s2, null      ; encoding: [0x05,0x00,0x48,0xd5,0x01,0x04,0xf4,0x01]
+// GFX11: v_bfe_u32 v5, s1, s2, null      ; encoding: [0x05,0x00,0x10,0xd6,0x01,0x04,0xf0,0x01]

diff --git a/llvm/test/MC/AMDGPU/gfx11_asm_operands.s b/llvm/test/MC/AMDGPU/gfx11_asm_operands.s
new file mode 100644 (file)
index 0000000..d77962e
--- /dev/null
+++ b/llvm/test/MC/AMDGPU/gfx11_asm_operands.s
@@ -0,0 +1,107 @@
+// RUN: llvm-mc -arch=amdgcn -mcpu=gfx1030 -show-encoding %s | FileCheck --check-prefix=GFX10 %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 -show-encoding %s | FileCheck --check-prefix=GFX11 %s
+// RUN: not llvm-mc -arch=amdgcn -mcpu=gfx1100 -show-encoding %s 2>&1 | FileCheck --check-prefixes=GFX11-ERR %s
+
+// On GFX11+, EXECZ and VCCZ are no longer allowed to be used as sources to SALU and VALU instructions.
+// The inline constants are removed. VCCZ and EXECZ still exist and can be use for conditional branches.
+// LDS_DIRECT and POPS_EXITING_WAVE_ID are also no longer allowed.
+
+//---------------------------------------------------------------------------//
+// EXECZ
+//---------------------------------------------------------------------------//
+
+s_cbranch_execz 0x100
+// GFX10: encoding: [0x00,0x01,0x88,0xbf]
+// GFX11: encoding: [0x00,0x01,0xa5,0xbf]
+
+s_add_i32 s0, execz, s2
+// GFX10: encoding: [0xfc,0x02,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+s_add_i32 s0, src_execz, s2
+// GFX10: encoding: [0xfc,0x02,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+s_add_i32 s0, s1, execz
+// GFX10: encoding: [0x01,0xfc,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+s_add_i32 s0, s1, src_execz
+// GFX10: encoding: [0x01,0xfc,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], execz, v[2:3]
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0xfc,0x04,0x02,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], src_execz, v[2:3]
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0xfc,0x04,0x02,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], v[1:2], execz
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0x01,0xf9,0x01,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], v[1:2], src_execz
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0x01,0xf9,0x01,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+//---------------------------------------------------------------------------//
+// VCCZ
+//---------------------------------------------------------------------------//
+
+s_cbranch_vccz 0x100
+// GFX10: encoding: [0x00,0x01,0x86,0xbf]
+// GFX11: encoding: [0x00,0x01,0xa3,0xbf]
+
+s_add_i32 s0, vccz, s2
+// GFX10: encoding: [0xfb,0x02,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+s_add_i32 s0, src_vccz, s2
+// GFX10: encoding: [0xfb,0x02,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+s_add_i32 s0, s1, vccz
+// GFX10: encoding: [0x01,0xfb,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+s_add_i32 s0, s1, src_vccz
+// GFX10: encoding: [0x01,0xfb,0x00,0x81]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], vccz, v[2:3]
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0xfb,0x04,0x02,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], src_vccz, v[2:3]
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0xfb,0x04,0x02,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], v[1:2], vccz
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0x01,0xf7,0x01,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+v_add_f64 v[0:1], v[1:2], src_vccz
+// GFX10: encoding: [0x00,0x00,0x64,0xd5,0x01,0xf7,0x01,0x00]
+// GFX11-ERR: error: execz and vccz are not supported on this GPU
+
+//---------------------------------------------------------------------------//
+// LDS_DIRECT
+//---------------------------------------------------------------------------//
+
+v_readfirstlane_b32 s0, lds_direct
+// GFX10: encoding: [0xfe,0x04,0x00,0x7e]
+// GFX11-ERR: error: lds_direct is not supported on this GPU
+
+v_readfirstlane_b32 s0, src_lds_direct
+// GFX10: encoding: [0xfe,0x04,0x00,0x7e]
+// GFX11-ERR: error: lds_direct is not supported on this GPU
+
+v_mov_b32 v0, lds_direct
+// GFX10: encoding: [0xfe,0x02,0x00,0x7e]
+// GFX11-ERR: error: lds_direct is not supported on this GPU
+
+v_mov_b32 v0, src_lds_direct
+// GFX10: encoding: [0xfe,0x02,0x00,0x7e]
+// GFX11-ERR: error: lds_direct is not supported on this GPU

diff --git a/llvm/test/MC/Disassembler/AMDGPU/decode-err.txt b/llvm/test/MC/Disassembler/AMDGPU/decode-err.txt
index b90ce05..c016970 100644 (file)
--- a/llvm/test/MC/Disassembler/AMDGPU/decode-err.txt
+++ b/llvm/test/MC/Disassembler/AMDGPU/decode-err.txt
@@ -1,4 +1,9 @@
 # RUN: llvm-mc -arch=amdgcn -mcpu=gfx900 -disassemble -show-encoding < %s 2>&1 | FileCheck -check-prefix=GCN %s
+# RUN: llvm-mc -arch=amdgcn -mcpu=gfx1100 -disassemble -show-encoding < %s 2>&1 | FileCheck -check-prefix=GFX11 %s
 
 # GCN: warning: invalid instruction encoding
 0xdf,0x00,0x00,0x02
+
+# this is buffer_atomic_csub_u32 v5, off, s[8:11], s3 offset:4095. Invalid without glc
+# GFX11: warning: invalid instruction encoding
+0xff,0x0f,0xdc,0xe0,0x00,0x05,0x02,0x03