mirror of
https://github.com/KytyPS5/KytyPS5.git
synced 2026-10-21 11:26:56 +02:00
shader: complete signed 64-bit integer comparisons
This commit is contained in:
@@ -487,11 +487,14 @@ enum class Opcode {
|
||||
V_CMP_NE_U32,
|
||||
V_CMP_GE_U32,
|
||||
V_CMP_T_U32,
|
||||
// Keep V_CMP_F_I64 to V_CMPX_T_U64 contiguous: VectorAluOps.cpp tests this range.
|
||||
V_CMP_F_I64,
|
||||
V_CMP_EQ_I64,
|
||||
V_CMP_LT_I64,
|
||||
V_CMP_EQ_I64,
|
||||
V_CMP_LE_I64,
|
||||
V_CMP_GT_I64,
|
||||
V_CMP_NE_I64,
|
||||
V_CMP_GE_I64,
|
||||
V_CMP_T_I64,
|
||||
V_CMP_F_U64,
|
||||
V_CMP_LT_U64,
|
||||
@@ -505,7 +508,9 @@ enum class Opcode {
|
||||
V_CMPX_LT_I64,
|
||||
V_CMPX_EQ_I64,
|
||||
V_CMPX_LE_I64,
|
||||
V_CMPX_GT_I64,
|
||||
V_CMPX_NE_I64,
|
||||
V_CMPX_GE_I64,
|
||||
V_CMPX_T_I64,
|
||||
V_CMPX_F_U64,
|
||||
V_CMPX_LT_U64,
|
||||
|
||||
@@ -278,25 +278,26 @@ constexpr VopcOpcodeInfo VOPC_OPCODE_LIST[] = {
|
||||
{0xc1u, Opcode::V_CMP_LT_U32}, {0xc2u, Opcode::V_CMP_EQ_U32},
|
||||
{0xc3u, Opcode::V_CMP_LE_U32}, {0xc4u, Opcode::V_CMP_GT_U32},
|
||||
{0xc5u, Opcode::V_CMP_NE_U32}, {0xc6u, Opcode::V_CMP_GE_U32},
|
||||
{0xc7u, Opcode::V_CMP_T_U32}, {0xa2u, Opcode::V_CMP_EQ_I64, false},
|
||||
{0xa1u, Opcode::V_CMP_LT_I64, false}, {0xa3u, Opcode::V_CMP_LE_I64, false},
|
||||
{0xa5u, Opcode::V_CMP_NE_I64, false},
|
||||
{0xa0u, Opcode::V_CMP_F_I64, false}, {0xa7u, Opcode::V_CMP_T_I64, false},
|
||||
{0xb0u, Opcode::V_CMPX_F_I64, false}, {0xb1u, Opcode::V_CMPX_LT_I64, false},
|
||||
{0xb2u, Opcode::V_CMPX_EQ_I64, false}, {0xb3u, Opcode::V_CMPX_LE_I64, false},
|
||||
{0xb7u, Opcode::V_CMPX_T_I64, false},
|
||||
{0xb5u, Opcode::V_CMPX_NE_I64, false}, {0xd1u, Opcode::V_CMPX_LT_U32},
|
||||
{0xc7u, Opcode::V_CMP_T_U32}, {0xd1u, Opcode::V_CMPX_LT_U32},
|
||||
{0xd2u, Opcode::V_CMPX_EQ_U32}, {0xd3u, Opcode::V_CMPX_LE_U32},
|
||||
{0xd4u, Opcode::V_CMPX_GT_U32}, {0xd5u, Opcode::V_CMPX_NE_U32},
|
||||
{0xd6u, Opcode::V_CMPX_GE_U32}, {0xe1u, Opcode::V_CMP_LT_U64, false},
|
||||
{0xe0u, Opcode::V_CMP_F_U64, false}, {0xe7u, Opcode::V_CMP_T_U64, false},
|
||||
{0xf0u, Opcode::V_CMPX_F_U64, false}, {0xf1u, Opcode::V_CMPX_LT_U64, false},
|
||||
{0xf2u, Opcode::V_CMPX_EQ_U64, false}, {0xf4u, Opcode::V_CMPX_GT_U64, false},
|
||||
{0xf6u, Opcode::V_CMPX_GE_U64, false}, {0xf7u, Opcode::V_CMPX_T_U64, false},
|
||||
{0xe2u, Opcode::V_CMP_EQ_U64, false}, {0xe3u, Opcode::V_CMP_LE_U64, false},
|
||||
{0xe4u, Opcode::V_CMP_GT_U64, false}, {0xe5u, Opcode::V_CMP_NE_U64, false},
|
||||
{0xe6u, Opcode::V_CMP_GE_U64, false}, {0xf3u, Opcode::V_CMPX_LE_U64, false},
|
||||
{0xf5u, Opcode::V_CMPX_NE_U64, false}, {0xc9u, Opcode::V_CMP_LT_F16},
|
||||
{0xd6u, Opcode::V_CMPX_GE_U32}, {0xa0u, Opcode::V_CMP_F_I64, false},
|
||||
{0xa1u, Opcode::V_CMP_LT_I64, false}, {0xa2u, Opcode::V_CMP_EQ_I64, false},
|
||||
{0xa3u, Opcode::V_CMP_LE_I64, false}, {0xa4u, Opcode::V_CMP_GT_I64, false},
|
||||
{0xa5u, Opcode::V_CMP_NE_I64, false}, {0xa6u, Opcode::V_CMP_GE_I64, false},
|
||||
{0xa7u, Opcode::V_CMP_T_I64, false}, {0xb0u, Opcode::V_CMPX_F_I64, false},
|
||||
{0xb1u, Opcode::V_CMPX_LT_I64, false}, {0xb2u, Opcode::V_CMPX_EQ_I64, false},
|
||||
{0xb3u, Opcode::V_CMPX_LE_I64, false}, {0xb4u, Opcode::V_CMPX_GT_I64, false},
|
||||
{0xb5u, Opcode::V_CMPX_NE_I64, false}, {0xb6u, Opcode::V_CMPX_GE_I64, false},
|
||||
{0xb7u, Opcode::V_CMPX_T_I64, false}, {0xe0u, Opcode::V_CMP_F_U64, false},
|
||||
{0xe1u, Opcode::V_CMP_LT_U64, false}, {0xe2u, Opcode::V_CMP_EQ_U64, false},
|
||||
{0xe3u, Opcode::V_CMP_LE_U64, false}, {0xe4u, Opcode::V_CMP_GT_U64, false},
|
||||
{0xe5u, Opcode::V_CMP_NE_U64, false}, {0xe6u, Opcode::V_CMP_GE_U64, false},
|
||||
{0xe7u, Opcode::V_CMP_T_U64, false}, {0xf0u, Opcode::V_CMPX_F_U64, false},
|
||||
{0xf1u, Opcode::V_CMPX_LT_U64, false}, {0xf2u, Opcode::V_CMPX_EQ_U64, false},
|
||||
{0xf3u, Opcode::V_CMPX_LE_U64, false}, {0xf4u, Opcode::V_CMPX_GT_U64, false},
|
||||
{0xf5u, Opcode::V_CMPX_NE_U64, false}, {0xf6u, Opcode::V_CMPX_GE_U64, false},
|
||||
{0xf7u, Opcode::V_CMPX_T_U64, false}, {0xc9u, Opcode::V_CMP_LT_F16},
|
||||
{0xcau, Opcode::V_CMP_EQ_F16}, {0xcbu, Opcode::V_CMP_LE_F16},
|
||||
{0xccu, Opcode::V_CMP_GT_F16}, {0xcdu, Opcode::V_CMP_LG_F16},
|
||||
{0xceu, Opcode::V_CMP_GE_F16}, {0xe9u, Opcode::V_CMP_NGE_F16},
|
||||
@@ -1245,9 +1246,10 @@ VopcSdwaFields DecodeVopcSdwaFields(uint32_t modifier) {
|
||||
}
|
||||
|
||||
bool SupportsVopcSdwa(Opcode opcode) {
|
||||
// F64 SDWA operand handling is not implemented.
|
||||
return opcode != Opcode::UNSUPPORTED && opcode != Opcode::V_CMP_NE_I64 &&
|
||||
(opcode < Opcode::V_CMP_F_F64 || opcode > Opcode::V_CMPX_TRU_F64);
|
||||
// SDWA operand handling is not implemented for 64-bit compares.
|
||||
return opcode != Opcode::UNSUPPORTED &&
|
||||
(opcode < Opcode::V_CMP_F_F64 || opcode > Opcode::V_CMPX_TRU_F64) &&
|
||||
(opcode < Opcode::V_CMP_F_I64 || opcode > Opcode::V_CMPX_T_U64);
|
||||
}
|
||||
|
||||
void DecodeVopcSdwa(uint32_t pc, std::span<const uint32_t> code, uint32_t word_index,
|
||||
@@ -1628,7 +1630,9 @@ bool IsVopcCompareExec(Opcode opcode) {
|
||||
case Opcode::V_CMPX_LT_I64:
|
||||
case Opcode::V_CMPX_EQ_I64:
|
||||
case Opcode::V_CMPX_LE_I64:
|
||||
case Opcode::V_CMPX_GT_I64:
|
||||
case Opcode::V_CMPX_NE_I64:
|
||||
case Opcode::V_CMPX_GE_I64:
|
||||
case Opcode::V_CMPX_T_I64:
|
||||
case Opcode::V_CMPX_F_U64:
|
||||
case Opcode::V_CMPX_LT_U64:
|
||||
|
||||
Reference in New Issue
Block a user