summaryrefslogtreecommitdiffstats
path: root/src
diff options
context:
space:
mode:
authorYuri Kunde Schlesner <yuriks@yuriks.net>2016-12-03 05:24:24 +0100
committerGitHub <noreply@github.com>2016-12-03 05:24:24 +0100
commit018191c1f01980d6a642b0ef1bd0a4cd636c0178 (patch)
treef37a96f68ab0a1e9a19bb46649963ef1908ed352 /src
parentMerge pull request #2251 from JayFoxRox/remove-version (diff)
parentshader_jit: Load LOOPCOUNT_REG and LOOPINC 4 bit left-shifted (diff)
downloadyuzu-018191c1f01980d6a642b0ef1bd0a4cd636c0178.tar
yuzu-018191c1f01980d6a642b0ef1bd0a4cd636c0178.tar.gz
yuzu-018191c1f01980d6a642b0ef1bd0a4cd636c0178.tar.bz2
yuzu-018191c1f01980d6a642b0ef1bd0a4cd636c0178.tar.lz
yuzu-018191c1f01980d6a642b0ef1bd0a4cd636c0178.tar.xz
yuzu-018191c1f01980d6a642b0ef1bd0a4cd636c0178.tar.zst
yuzu-018191c1f01980d6a642b0ef1bd0a4cd636c0178.zip
Diffstat (limited to 'src')
-rw-r--r--src/video_core/shader/shader_jit_x64.cpp15
1 files changed, 9 insertions, 6 deletions
diff --git a/src/video_core/shader/shader_jit_x64.cpp b/src/video_core/shader/shader_jit_x64.cpp
index 211c703ab..9a3d6ca8f 100644
--- a/src/video_core/shader/shader_jit_x64.cpp
+++ b/src/video_core/shader/shader_jit_x64.cpp
@@ -102,11 +102,11 @@ static const X64Reg SETUP = R9;
/// The two 32-bit VS address offset registers set by the MOVA instruction
static const X64Reg ADDROFFS_REG_0 = R10;
static const X64Reg ADDROFFS_REG_1 = R11;
-/// VS loop count register
+/// VS loop count register (Multiplied by 16)
static const X64Reg LOOPCOUNT_REG = R12;
/// Current VS loop iteration number (we could probably use LOOPCOUNT_REG, but this quicker)
static const X64Reg LOOPCOUNT = RSI;
-/// Number to increment LOOPCOUNT_REG by on each loop iteration
+/// Number to increment LOOPCOUNT_REG by on each loop iteration (Multiplied by 16)
static const X64Reg LOOPINC = RDI;
/// Result of the previous CMP instruction for the X-component comparison
static const X64Reg COND0 = R13;
@@ -718,15 +718,18 @@ void JitShader::Compile_LOOP(Instruction instr) {
looping = true;
+ // This decodes the fields from the integer uniform at index instr.flow_control.int_uniform_id.
+ // The Y (LOOPCOUNT_REG) and Z (LOOPINC) component are kept multiplied by 16 (Left shifted by
+ // 4 bits) to be used as an offset into the 16-byte vector registers later
int offset =
ShaderSetup::UniformOffset(RegisterType::IntUniform, instr.flow_control.int_uniform_id);
MOV(32, R(LOOPCOUNT), MDisp(SETUP, offset));
MOV(32, R(LOOPCOUNT_REG), R(LOOPCOUNT));
- SHR(32, R(LOOPCOUNT_REG), Imm8(8));
- AND(32, R(LOOPCOUNT_REG), Imm32(0xff)); // Y-component is the start
+ SHR(32, R(LOOPCOUNT_REG), Imm8(4));
+ AND(32, R(LOOPCOUNT_REG), Imm32(0xFF0)); // Y-component is the start
MOV(32, R(LOOPINC), R(LOOPCOUNT));
- SHR(32, R(LOOPINC), Imm8(16));
- MOVZX(32, 8, LOOPINC, R(LOOPINC)); // Z-component is the incrementer
+ SHR(32, R(LOOPINC), Imm8(12));
+ AND(32, R(LOOPINC), Imm32(0xFF0)); // Z-component is the incrementer
MOVZX(32, 8, LOOPCOUNT, R(LOOPCOUNT)); // X-component is iteration count
ADD(32, R(LOOPCOUNT), Imm8(1)); // Iteration count is X-component + 1