aboutsummaryrefslogtreecommitdiff
path: root/src/video_core/vertex_shader.cpp
diff options
context:
space:
mode:
Diffstat (limited to 'src/video_core/vertex_shader.cpp')
-rw-r--r--src/video_core/vertex_shader.cpp210
1 files changed, 165 insertions, 45 deletions
diff --git a/src/video_core/vertex_shader.cpp b/src/video_core/vertex_shader.cpp
index ff825e2e1..bc8c0041c 100644
--- a/src/video_core/vertex_shader.cpp
+++ b/src/video_core/vertex_shader.cpp
@@ -85,8 +85,12 @@ struct VertexShaderState {
};
struct CallStackElement {
- u32 final_address;
- u32 return_address;
+ u32 final_address; // Address upon which we jump to return_address
+ u32 return_address; // Where to jump when leaving scope
+ u8 repeat_counter; // How often to repeat until this call stack element is removed
+ u8 loop_increment; // Which value to add to the loop counter after an iteration
+ // TODO: Should this be a signed value? Does it even matter?
+ u32 loop_address; // The address where we'll return to after each loop iteration
};
// TODO: Is there a maximal size for this?
@@ -105,9 +109,16 @@ static void ProcessShaderCode(VertexShaderState& state) {
while (true) {
if (!state.call_stack.empty()) {
- if (state.program_counter - shader_memory.data() == state.call_stack.top().final_address) {
- state.program_counter = &shader_memory[state.call_stack.top().return_address];
- state.call_stack.pop();
+ auto& top = state.call_stack.top();
+ if (state.program_counter - shader_memory.data() == top.final_address) {
+ state.address_registers[2] += top.loop_increment;
+
+ if (top.repeat_counter-- == 0) {
+ state.program_counter = &shader_memory[top.return_address];
+ state.call_stack.pop();
+ } else {
+ state.program_counter = &shader_memory[top.loop_address];
+ }
// TODO: Is "trying again" accurate to hardware?
continue;
@@ -118,9 +129,10 @@ static void ProcessShaderCode(VertexShaderState& state) {
const Instruction& instr = *(const Instruction*)state.program_counter;
const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.common.operand_desc_id];
- auto call = [&](VertexShaderState& state, u32 offset, u32 num_instructions, u32 return_offset) {
+ static auto call = [](VertexShaderState& state, u32 offset, u32 num_instructions,
+ u32 return_offset, u8 repeat_count, u8 loop_increment) {
state.program_counter = &shader_memory[offset] - 1; // -1 to make sure when incrementing the PC we end up at the correct offset
- state.call_stack.push({ offset + num_instructions, return_offset });
+ state.call_stack.push({ offset + num_instructions, return_offset, repeat_count, loop_increment, offset });
};
u32 binary_offset = state.program_counter - shader_memory.data();
@@ -146,13 +158,10 @@ static void ProcessShaderCode(VertexShaderState& state) {
case Instruction::OpCodeType::Arithmetic:
{
bool is_inverted = 0 != (instr.opcode.GetInfo().subtype & Instruction::OpCodeInfo::SrcInversed);
- if (is_inverted) {
- // TODO: We don't really support this properly: For instance, the address register
- // offset needs to be applied to SRC2 instead, etc.
- // For now, we just abort in this situation.
- LOG_CRITICAL(HW_GPU, "Bad condition...");
- exit(0);
- }
+ // TODO: We don't really support this properly: For instance, the address register
+ // offset needs to be applied to SRC2 instead, etc.
+ // For now, we just abort in this situation.
+ ASSERT_MSG(!is_inverted, "Bad condition...");
const int address_offset = (instr.common.address_register_index == 0)
? 0 : state.address_registers[instr.common.address_register_index - 1];
@@ -255,7 +264,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
// TODO: Be stable against division by zero!
// TODO: I think this might be wrong... we should only use one component here
- dest[i] = float24::FromFloat32(1.0 / src1[i].ToFloat32());
+ dest[i] = float24::FromFloat32(1.0f / src1[i].ToFloat32());
}
break;
@@ -270,7 +279,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
// TODO: Be stable against division by zero!
// TODO: I think this might be wrong... we should only use one component here
- dest[i] = float24::FromFloat32(1.0 / sqrt(src1[i].ToFloat32()));
+ dest[i] = float24::FromFloat32(1.0f / sqrt(src1[i].ToFloat32()));
}
break;
@@ -342,24 +351,143 @@ static void ProcessShaderCode(VertexShaderState& state) {
default:
LOG_ERROR(HW_GPU, "Unhandled arithmetic instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value(), instr.opcode.GetInfo().name, instr.hex);
- _dbg_assert_(HW_GPU, 0);
+ DEBUG_ASSERT(false);
break;
}
break;
}
+
+ case Instruction::OpCodeType::MultiplyAdd:
+ {
+ if (instr.opcode.EffectiveOpCode() == Instruction::OpCode::MAD) {
+ const SwizzlePattern& swizzle = *(SwizzlePattern*)&swizzle_data[instr.mad.operand_desc_id];
+
+ const float24* src1_ = LookupSourceRegister(instr.mad.src1);
+ const float24* src2_ = LookupSourceRegister(instr.mad.src2);
+ const float24* src3_ = LookupSourceRegister(instr.mad.src3);
+
+ const bool negate_src1 = ((bool)swizzle.negate_src1 != false);
+ const bool negate_src2 = ((bool)swizzle.negate_src2 != false);
+ const bool negate_src3 = ((bool)swizzle.negate_src3 != false);
+
+ float24 src1[4] = {
+ src1_[(int)swizzle.GetSelectorSrc1(0)],
+ src1_[(int)swizzle.GetSelectorSrc1(1)],
+ src1_[(int)swizzle.GetSelectorSrc1(2)],
+ src1_[(int)swizzle.GetSelectorSrc1(3)],
+ };
+ if (negate_src1) {
+ src1[0] = src1[0] * float24::FromFloat32(-1);
+ src1[1] = src1[1] * float24::FromFloat32(-1);
+ src1[2] = src1[2] * float24::FromFloat32(-1);
+ src1[3] = src1[3] * float24::FromFloat32(-1);
+ }
+ float24 src2[4] = {
+ src2_[(int)swizzle.GetSelectorSrc2(0)],
+ src2_[(int)swizzle.GetSelectorSrc2(1)],
+ src2_[(int)swizzle.GetSelectorSrc2(2)],
+ src2_[(int)swizzle.GetSelectorSrc2(3)],
+ };
+ if (negate_src2) {
+ src2[0] = src2[0] * float24::FromFloat32(-1);
+ src2[1] = src2[1] * float24::FromFloat32(-1);
+ src2[2] = src2[2] * float24::FromFloat32(-1);
+ src2[3] = src2[3] * float24::FromFloat32(-1);
+ }
+ float24 src3[4] = {
+ src3_[(int)swizzle.GetSelectorSrc3(0)],
+ src3_[(int)swizzle.GetSelectorSrc3(1)],
+ src3_[(int)swizzle.GetSelectorSrc3(2)],
+ src3_[(int)swizzle.GetSelectorSrc3(3)],
+ };
+ if (negate_src3) {
+ src3[0] = src3[0] * float24::FromFloat32(-1);
+ src3[1] = src3[1] * float24::FromFloat32(-1);
+ src3[2] = src3[2] * float24::FromFloat32(-1);
+ src3[3] = src3[3] * float24::FromFloat32(-1);
+ }
+
+ float24* dest = (instr.mad.dest < 0x08) ? state.output_register_table[4*instr.mad.dest.GetIndex()]
+ : (instr.mad.dest < 0x10) ? dummy_vec4_float24
+ : (instr.mad.dest < 0x20) ? &state.temporary_registers[instr.mad.dest.GetIndex()][0]
+ : dummy_vec4_float24;
+
+ for (int i = 0; i < 4; ++i) {
+ if (!swizzle.DestComponentEnabled(i))
+ continue;
+
+ dest[i] = src1[i] * src2[i] + src3[i];
+ }
+ } else {
+ LOG_ERROR(HW_GPU, "Unhandled multiply-add instruction: 0x%02x (%s): 0x%08x",
+ (int)instr.opcode.Value(), instr.opcode.GetInfo().name, instr.hex);
+ }
+ break;
+ }
+
default:
+ {
+ static auto evaluate_condition = [](const VertexShaderState& state, bool refx, bool refy, Instruction::FlowControlType flow_control) {
+ bool results[2] = { refx == state.conditional_code[0],
+ refy == state.conditional_code[1] };
+
+ switch (flow_control.op) {
+ case flow_control.Or:
+ return results[0] || results[1];
+
+ case flow_control.And:
+ return results[0] && results[1];
+
+ case flow_control.JustX:
+ return results[0];
+
+ case flow_control.JustY:
+ return results[1];
+ }
+ };
+
// Handle each instruction on its own
switch (instr.opcode) {
case Instruction::OpCode::END:
exit_loop = true;
break;
+ case Instruction::OpCode::JMPC:
+ if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) {
+ state.program_counter = &shader_memory[instr.flow_control.dest_offset] - 1;
+ }
+ break;
+
+ case Instruction::OpCode::JMPU:
+ if (shader_uniforms.b[instr.flow_control.bool_uniform_id]) {
+ state.program_counter = &shader_memory[instr.flow_control.dest_offset] - 1;
+ }
+ break;
+
case Instruction::OpCode::CALL:
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- binary_offset + 1);
+ binary_offset + 1, 0, 0);
+ break;
+
+ case Instruction::OpCode::CALLU:
+ if (shader_uniforms.b[instr.flow_control.bool_uniform_id]) {
+ call(state,
+ instr.flow_control.dest_offset,
+ instr.flow_control.num_instructions,
+ binary_offset + 1, 0, 0);
+ }
+ break;
+
+ case Instruction::OpCode::CALLC:
+ if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) {
+ call(state,
+ instr.flow_control.dest_offset,
+ instr.flow_control.num_instructions,
+ binary_offset + 1, 0, 0);
+ }
break;
case Instruction::OpCode::NOP:
@@ -370,12 +498,12 @@ static void ProcessShaderCode(VertexShaderState& state) {
call(state,
binary_offset + 1,
instr.flow_control.dest_offset - binary_offset - 1,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
}
break;
@@ -384,43 +512,34 @@ static void ProcessShaderCode(VertexShaderState& state) {
{
// TODO: Do we need to consider swizzlers here?
- auto flow_control = instr.flow_control;
- bool results[3] = { (bool)flow_control.refx == state.conditional_code[0],
- (bool)flow_control.refy == state.conditional_code[1] };
-
- switch (flow_control.op) {
- case flow_control.Or:
- results[2] = results[0] || results[1];
- break;
-
- case flow_control.And:
- results[2] = results[0] && results[1];
- break;
-
- case flow_control.JustX:
- results[2] = results[0];
- break;
-
- case flow_control.JustY:
- results[2] = results[1];
- break;
- }
-
- if (results[2]) {
+ if (evaluate_condition(state, instr.flow_control.refx, instr.flow_control.refy, instr.flow_control)) {
call(state,
binary_offset + 1,
instr.flow_control.dest_offset - binary_offset - 1,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
} else {
call(state,
instr.flow_control.dest_offset,
instr.flow_control.num_instructions,
- instr.flow_control.dest_offset + instr.flow_control.num_instructions);
+ instr.flow_control.dest_offset + instr.flow_control.num_instructions, 0, 0);
}
break;
}
+ case Instruction::OpCode::LOOP:
+ {
+ state.address_registers[2] = shader_uniforms.i[instr.flow_control.int_uniform_id].y;
+
+ call(state,
+ binary_offset + 1,
+ instr.flow_control.dest_offset - binary_offset + 1,
+ instr.flow_control.dest_offset + 1,
+ shader_uniforms.i[instr.flow_control.int_uniform_id].x,
+ shader_uniforms.i[instr.flow_control.int_uniform_id].z);
+ break;
+ }
+
default:
LOG_ERROR(HW_GPU, "Unhandled instruction: 0x%02x (%s): 0x%08x",
(int)instr.opcode.Value(), instr.opcode.GetInfo().name, instr.hex);
@@ -429,6 +548,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
break;
}
+ }
++state.program_counter;
generated by cgit v1.2.3 (git 2.25.1) at 2025-12-17 13:48:52 +0000