gfx11_dasm_wmma.txt | Explore in Territory

# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize32 -disassemble -show-encoding < %s | FileCheck -check-prefix=W32 %s
# RUN: llvm-mc -triple=amdgcn -mcpu=gfx1100 -mattr=+wavefrontsize64 -disassemble -show-encoding < %s | FileCheck -check-prefix=W64 %s


# Test v_wmma_f32_16x16x16_f16

# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]                               ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c]
# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]                               ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c]
0x10,0x40,0x40,0xcc,0x00,0x11,0x42,0x1c

# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b]
# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b]
0x10,0x40,0x40,0xcc,0x00,0x11,0xca,0x1b

# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c]
# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c]
0x10,0x41,0x40,0xcc,0x00,0x11,0x42,0x3c

# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c]
# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c]
0x10,0x42,0x40,0xcc,0x00,0x11,0x42,0x5c

# W32: v_wmma_f32_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c]
# W64: v_wmma_f32_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c]
0x10,0x43,0x40,0xcc,0x00,0x11,0x42,0x7c


# Test v_wmma_f32_16x16x16_bf16

# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23]                               ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c]
# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19]                               ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c]
0x10,0x40,0x41,0xcc,0x00,0x11,0x42,0x1c

# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b]
# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0                                    ; encoding: [0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b]
0x10,0x40,0x41,0xcc,0x00,0x11,0xca,0x1b

# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c]
# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c]
0x10,0x41,0x41,0xcc,0x00,0x11,0x42,0x3c

# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c]
# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c]
0x10,0x42,0x41,0xcc,0x00,0x11,0x42,0x5c

# W32: v_wmma_f32_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c]
# W64: v_wmma_f32_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c]
0x10,0x43,0x41,0xcc,0x00,0x11,0x42,0x7c


# Test v_wmma_f16_16x16x16_f16

# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23]                                ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c]
# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19]                                ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c]
0x10,0x40,0x42,0xcc,0x00,0x11,0x42,0x1c

# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], 1.0                                     ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b]
# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], 1.0                                     ; encoding: [0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b]
0x10,0x40,0x42,0xcc,0x00,0x11,0xca,0x1b

# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1]                 ; encoding: [0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c]
# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1]                 ; encoding: [0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c]
0x10,0x60,0x42,0xcc,0x00,0x11,0x42,0x1c

# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0]  ; encoding: [0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c]
# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0]  ; encoding: [0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c]
0x10,0x41,0x42,0xcc,0x00,0x11,0x42,0x3c

# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0]  ; encoding: [0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c]
# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0]  ; encoding: [0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c]
0x10,0x42,0x42,0xcc,0x00,0x11,0x42,0x5c

# W32: v_wmma_f16_16x16x16_f16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0]  ; encoding: [0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c]
# W64: v_wmma_f16_16x16x16_f16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0]  ; encoding: [0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c]
0x10,0x43,0x42,0xcc,0x00,0x11,0x42,0x7c


# Test v_wmma_bf16_16x16x16_bf16

# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c]
# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c]
0x10,0x40,0x43,0xcc,0x00,0x11,0x42,0x1c

# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b]
# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], 1.0 ; encoding: [0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b]
0x10,0x40,0x43,0xcc,0x00,0x11,0xca,0x1b

# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c]
# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] op_sel:[0,0,1] ; encoding: [0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c]
0x10,0x60,0x43,0xcc,0x00,0x11,0x42,0x1c

# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c]
# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c]
0x10,0x41,0x43,0xcc,0x00,0x11,0x42,0x3c

# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c]
# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c]
0x10,0x42,0x43,0xcc,0x00,0x11,0x42,0x5c

# W32: v_wmma_bf16_16x16x16_bf16 v[16:23], v[0:7], v[8:15], v[16:23] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c]
# W64: v_wmma_bf16_16x16x16_bf16 v[16:19], v[0:7], v[8:15], v[16:19] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c]
0x10,0x43,0x43,0xcc,0x00,0x11,0x42,0x7c


# Test v_wmma_i32_16x16x16_iu8

# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15]                               ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c]
# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11]                               ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c]
0x08,0x40,0x44,0xcc,0x00,0x09,0x22,0x1c

# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], 1                                      ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a]
# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], 1                                      ; encoding: [0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a]
0x08,0x40,0x44,0xcc,0x00,0x09,0x06,0x1a

# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c]
# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c]
0x08,0x41,0x44,0xcc,0x00,0x09,0x22,0x3c

# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c]
# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c]
0x08,0x42,0x44,0xcc,0x00,0x09,0x22,0x5c

# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c]
# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c]
0x08,0x43,0x44,0xcc,0x00,0x09,0x22,0x7c

# W32: v_wmma_i32_16x16x16_iu8 v[8:15], v[0:3], v[4:7], v[8:15] clamp                         ; encoding: [0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c]
# W64: v_wmma_i32_16x16x16_iu8 v[8:11], v[0:3], v[4:7], v[8:11] clamp                         ; encoding: [0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c]
0x08,0xc0,0x44,0xcc,0x00,0x09,0x22,0x1c


# Test v_wmma_i32_16x16x16_iu4

# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11]                             ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c]
# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7]                               ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c]
0x04,0x40,0x45,0xcc,0x00,0x05,0x12,0x1c

# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], 1                                     ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a]
# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], 1                                      ; encoding: [0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a]
0x04,0x40,0x45,0xcc,0x00,0x05,0x06,0x1a

# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c]
# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,0,0] neg_hi:[1,0,0] ; encoding: [0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c]
0x04,0x41,0x45,0xcc,0x00,0x05,0x12,0x3c

# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c]
# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[0,1,0] neg_hi:[0,1,0] ; encoding: [0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c]
0x04,0x42,0x45,0xcc,0x00,0x05,0x12,0x5c

# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c]
# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] neg_lo:[1,1,0] neg_hi:[1,1,0] ; encoding: [0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c]
0x04,0x43,0x45,0xcc,0x00,0x05,0x12,0x7c

# W32: v_wmma_i32_16x16x16_iu4 v[4:11], v[0:1], v[2:3], v[4:11] clamp                         ; encoding: [0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c]
# W64: v_wmma_i32_16x16x16_iu4 v[4:7], v[0:1], v[2:3], v[4:7] clamp                         ; encoding: [0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c]
0x04,0xc0,0x45,0xcc,0x00,0x05,0x12,0x1c
llvm/llvm/test/MC/Disassembler/AMDGPU/gfx11_dasm_wmma.txt
