[AMDGPU][MC][GFX8+][DISASSEMBLER] Corrected 1/2pi value for 64-bit operands

[dp]

Summary

See https://bugs.llvm.org/show_bug.cgi?id=39332

[artem.tamazov]

lib/Target/AMDGPU/InstPrinter/AMDGPUInstPrinter.cpp

494

0.15915494309189532 shall be enough to represent 1/2pi in 64-bit ieee-754 fp number (raw representation is 0x3FC45F306DC9C882)

[dp]

Thanks! Corrected.

[artem.tamazov]

LGTM

docs/AMDGPUOperandSyntax.rst

 Only a small fraction of floating-point numbers may be encoded as *inline constants*.
 They are enumerated in the table below.
 Other floating-point numbers have to be encoded as :ref:`literals<amdgpu_synid_literal>`.
 
 Floating-point *inline constants* are converted to
 :ref:`expected operand type<amdgpu_syn_instruction_type>`
 as described :ref:`here<amdgpu_synid_fp_const_conv>`.
 
-    ================================== ===================================================== ==================
-    Value                              Note                                                  Availability
-    ================================== ===================================================== ==================
-    0.0                                The same as integer constant 0.                       All GPUs
-    0.5                                Floating-point constant 0.5                           All GPUs
-    1.0                                Floating-point constant 1.0                           All GPUs
-    2.0                                Floating-point constant 2.0                           All GPUs
-    4.0                                Floating-point constant 4.0                           All GPUs
-    -0.5                               Floating-point constant -0.5                          All GPUs
-    -1.0                               Floating-point constant -1.0                          All GPUs
-    -2.0                               Floating-point constant -2.0                          All GPUs
-    -4.0                               Floating-point constant -4.0                          All GPUs
-    0.1592                             1.0/(2.0*pi). Use only for 16-bit operands.           GFX8, GFX9
-    0.15915494                         1.0/(2.0*pi). Use only for 16- and 32-bit operands.   GFX8, GFX9
-    0.159154943091895317852646485335   1.0/(2.0*pi).                                         GFX8, GFX9
-    ================================== ===================================================== ==================
+    ===================== ===================================================== ==================
+    Value                 Note                                                  Availability
+    ===================== ===================================================== ==================
+    0.0                   The same as integer constant 0.                       All GPUs
+    0.5                   Floating-point constant 0.5                           All GPUs
+    1.0                   Floating-point constant 1.0                           All GPUs
+    2.0                   Floating-point constant 2.0                           All GPUs
+    4.0                   Floating-point constant 4.0                           All GPUs
+    -0.5                  Floating-point constant -0.5                          All GPUs
+    -1.0                  Floating-point constant -1.0                          All GPUs
+    -2.0                  Floating-point constant -2.0                          All GPUs
+    -4.0                  Floating-point constant -4.0                          All GPUs
+    0.1592                1.0/(2.0*pi). Use only for 16-bit operands.           GFX8, GFX9
+    0.15915494            1.0/(2.0*pi). Use only for 16- and 32-bit operands.   GFX8, GFX9
+    0.15915494309189532   1.0/(2.0*pi).                                         GFX8, GFX9
+    ===================== ===================================================== ==================
 
 .. WARNING:: GFX7 does not support inline constants for *f16* operands.
 
 .. _amdgpu_synid_literal:
 
 literal
 -------
 

lib/Target/AMDGPU/InstPrinter/AMDGPUInstPrinter.cpp

void AMDGPUInstPrinter::printImmediate64(uint64_t Imm,
   else if (Imm == DoubleToBits(-2.0))
     O << "-2.0";
   else if (Imm == DoubleToBits(4.0))
     O << "4.0";
   else if (Imm == DoubleToBits(-4.0))
     O << "-4.0";
   else if (Imm == 0x3fc45f306dc9c882 &&
            STI.getFeatureBits()[AMDGPU::FeatureInv2PiInlineImm])
-  O << "0.15915494";
+    O << "0.15915494309189532";
   else {
     assert(isUInt<32>(Imm) || Imm == 0x3fc45f306dc9c882);
 
     // In rare situations, we will have a 32-bit literal in a 64-bit
     // operand. This is technically allowed for the encoding of s_mov_b64.
     O << formatHex(static_cast<uint64_t>(Imm));
   }
 }

test/CodeGen/AMDGPU/imm.ll

 
 ; GCN-LABEL: {{^}}add_inline_imm_inv_2pi_f64:
 ; SI: s_load_dwordx2 [[VAL:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x13
 ; SI-DAG: v_mov_b32_e32 v[[LO_VREG:[0-9]+]], 0x6dc9c882
 ; SI-DAG: v_mov_b32_e32 v[[HI_VREG:[0-9]+]], 0x3fc45f30
 ; SI: v_add_f64 v{{\[[0-9]+:[0-9]+\]}}, s{{\[[0-9]+:[0-9]+\]}}, v{{\[}}[[LO_VREG]]:[[HI_VREG]]{{\]}}
 
 ; VI: s_load_dwordx2 [[VAL:s\[[0-9]+:[0-9]+\]]], {{s\[[0-9]+:[0-9]+\]}}, 0x4c
-; VI: v_add_f64 [[REG:v\[[0-9]+:[0-9]+\]]], [[VAL]], 0.15915494{{$}}
+; VI: v_add_f64 [[REG:v\[[0-9]+:[0-9]+\]]], [[VAL]], 0.15915494309189532{{$}}
 ; VI: buffer_store_dwordx2 [[REG]]
 define amdgpu_kernel void @add_inline_imm_inv_2pi_f64(double addrspace(1)* %out, [8 x i32], double %x) {
   %y = fadd double %x, 0x3fc45f306dc9c882
   store double %y, double addrspace(1)* %out
   ret void
 }
 
 ; GCN-LABEL: {{^}}add_m_inv_2pi_f64:

test/MC/AMDGPU/literals.s

 // 1/(2*PI)
 //---------------------------------------------------------------------------//
 
 // NOSICI: error: invalid operand for instruction
 // NOVI: error: invalid operand for instruction
 v_trunc_f32_e32 v0, 0x3fc45f306dc9c882
 
 // NOSICI: error: invalid operand for instruction
-// VI: v_fract_f64_e32 v[0:1], 0.15915494 ; encoding: [0xf8,0x64,0x00,0x7e]
+// VI: v_fract_f64_e32 v[0:1], 0.15915494309189532 ; encoding: [0xf8,0x64,0x00,0x7e]
 v_fract_f64_e32 v[0:1], 0x3fc45f306dc9c882
 
 // SICI: v_trunc_f32_e32 v0, 0x3e22f983 ; encoding: [0xff,0x42,0x00,0x7e,0x83,0xf9,0x22,0x3e]
 // VI: v_trunc_f32_e32 v0, 0.15915494 ; encoding: [0xf8,0x38,0x00,0x7e]
 v_trunc_f32_e32 v0, 0x3e22f983
 
 // SICI: v_fract_f64_e32 v[0:1], 0x3e22f983 ; encoding: [0xff,0x7c,0x00,0x7e,0x83,0xf9,0x22,0x3e]
 // VI: v_fract_f64_e32 v[0:1], 0x3e22f983 ; encoding: [0xff,0x64,0x00,0x7e,0x83,0xf9,0x22,0x3e]
 v_fract_f64_e32 v[0:1], 0x3e22f983
 
 // NOSICI: error: invalid operand for instruction
 // NOVI: error: invalid operand for instruction
 v_trunc_f32_e64 v0, 0x3fc45f306dc9c882
 
 // NOSICI: error: invalid operand for instruction
-// VI: v_fract_f64_e64 v[0:1], 0.15915494 ; encoding: [0x00,0x00,0x72,0xd1,0xf8,0x00,0x00,0x00]
+// VI: v_fract_f64_e64 v[0:1], 0.15915494309189532 ; encoding: [0x00,0x00,0x72,0xd1,0xf8,0x00,0x00,0x00]
 v_fract_f64_e64 v[0:1], 0x3fc45f306dc9c882
 
 // NOSICI: error: invalid operand for instruction
 // VI: v_trunc_f32_e64 v0, 0.15915494 ; encoding: [0x00,0x00,0x5c,0xd1,0xf8,0x00,0x00,0x00]
 v_trunc_f32_e64 v0, 0x3e22f983
 
 // NOSICI: error: invalid operand for instruction
 // NOVI: error: invalid operand for instruction
 v_fract_f64_e64 v[0:1], 0x3e22f983
 
 // NOSICI: error: invalid operand for instruction
-// VI: s_mov_b64 s[0:1], 0.15915494 ; encoding: [0xf8,0x01,0x80,0xbe]
+// VI: s_mov_b64 s[0:1], 0.15915494309189532 ; encoding: [0xf8,0x01,0x80,0xbe]
 s_mov_b64_e32 s[0:1], 0.159154943091895317852646485335
 
 // SICI: v_and_b32_e32 v0, 0x3e22f983, v1 ; encoding: [0xff,0x02,0x00,0x36,0x83,0xf9,0x22,0x3e]
 // VI: v_and_b32_e32 v0, 0.15915494, v1 ; encoding: [0xf8,0x02,0x00,0x26]
 v_and_b32_e32 v0, 0.159154943091895317852646485335, v1
 
 // NOSICI: error: invalid operand for instruction
 // VI: v_and_b32_e64 v0, 0.15915494, v1 ; encoding: [0x00,0x00,0x13,0xd1,0xf8,0x02,0x02,0x00]
 v_and_b32_e64 v0, 0.159154943091895317852646485335, v1
 
 // SICI: v_fract_f64_e32 v[0:1], 0x3fc45f30 ; encoding: [0xff,0x7c,0x00,0x7e,0x30,0x5f,0xc4,0x3f]
-// VI: v_fract_f64_e32 v[0:1], 0.15915494 ; encoding: [0xf8,0x64,0x00,0x7e]
+// VI: v_fract_f64_e32 v[0:1], 0.15915494309189532 ; encoding: [0xf8,0x64,0x00,0x7e]
 v_fract_f64 v[0:1], 0.159154943091895317852646485335
 
 // SICI: v_trunc_f32_e32 v0, 0x3e22f983 ; encoding: [0xff,0x42,0x00,0x7e,0x83,0xf9,0x22,0x3e]
 // VI: v_trunc_f32_e32 v0, 0.15915494 ; encoding: [0xf8,0x38,0x00,0x7e]
 v_trunc_f32 v0, 0.159154943091895317852646485335

test/MC/Disassembler/AMDGPU/literal_vi.txt

+# RUN: llvm-mc -arch=amdgcn -mcpu=tonga -disassemble -show-encoding %s | FileCheck -check-prefix=VI %s
+
+# VI: v_fract_f64_e32 v[0:1], 0.15915494309189532 ; encoding: [0xf8,0x64,0x00,0x7e]
+0xf8,0x64,0x00,0x7e
+
+# VI: v_trunc_f32_e32 v0, 0.15915494 ; encoding: [0xf8,0x38,0x00,0x7e]
+0xf8,0x38,0x00,0x7e