Jump to content

  • Log In with Google      Sign In   
  • Create Account

Help with shader disassembling


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
6 replies to this topic

#1 Big Muscle   Members   -  Reputation: 135

Like
0Likes
Like

Posted 13 February 2013 - 04:26 AM

Hello,

I have a few short shaders in assembler code and I would like to understand what it really does. Or better, I know what it does but I am not able to get it work. So please, could someone help?

 

I need to get the alpha value of the pixels written by pixel shader. This is the pixel shader:



Microsoft ® Direct3D Shader Compiler 9.30.9200.16384
Copyright © Microsoft Corporation 2002-2011. All rights reserved.

//
// Generated by Microsoft ® D3D Shader Disassembler
//
//
///
// Note: shader requires additional functionality:
//       Minimum-precision data types
//
//
// Input signature:
//
// Name                 Index   Mask Register SysValue  Format   Used
// -------------------- ----- ------ -------- -------- ------- ------
// SV_POSITION              0   xyzw        0      POS   float       
// TEXCOORD                 0   xy          1     NONE   float   xy  
// TEXCOORD                 1   xyzw        2     NONE min2_8f   xyzw
//
//
// Output signature:
//
// Name                 Index   Mask Register SysValue  Format   Used
// -------------------- ----- ------ -------- -------- ------- ------
// SV_TARGET                0   xyzw        0   TARGET min2_8f   xyzw
//
//
// Sampler/Resource to DX9 shader sampler mappings:
//
// Target Sampler Source Sampler  Source Resource
// -------------- --------------- ----------------
// s0             s0              t0               
//
//
// Level9 shader bytecode:
//
    ps_2_0
    dcl t0.xy
    dcl t1 {min2_8f}
    dcl_2d s0
    texld r0 {min2_8f}, t0, s0
    mul r0.xyz, r0 {min2_8f}, t1 {min2_8f}
    mov r0.w, t1.w {min2_8f}
    mov oC0 {min2_8f}, r0

// approximately 4 instruction slots used (1 texture, 3 arithmetic)
ps_4_0
dcl_globalFlags refactoringAllowed | enableMinimumPrecision
dcl_sampler s0, mode_default
dcl_resource_texture2d (float,float,float,float) t0
dcl_input_ps linear v1.xy
dcl_input_ps linear v2.xyzw {min2_8f}
dcl_output o0.xyzw {min2_8f}
dcl_temps 1
sample r0.xyzw {min2_8f}, v1.xyxx, t0.xyzw, s0
mul r0.xyz, r0.xyzx {min2_8f as def32}, v2.xyzx {min2_8f as def32}
mov r0.w, v2.w {min2_8f as def32}
mov o0.xyzw {min2_8f}, r0.xyzw {def32 as min2_8f}
ret 
// Approximately 0 instruction slots used

 

As I understand correctly, The output alpha value is stored in o0.w which is just copied from the input (v2.w which the second input TEXCOORD.w). And there is a vertex shader which should compute the alpha and put it on its output:

 



Microsoft ® Direct3D Shader Compiler 9.30.9200.16384
Copyright © Microsoft Corporation 2002-2011. All rights reserved.

//
// Generated by Microsoft ® D3D Shader Disassembler
//
//
///
// Note: shader requires additional functionality:
//       Minimum-precision data types
//
//
// Input signature:
//
// Name                 Index   Mask Register SysValue  Format   Used
// -------------------- ----- ------ -------- -------- ------- ------
// POSITION                 0   xy          0     NONE   float   xy  
// TEXCOORD                 0   xy          1     NONE     int   x   
//
//
// Output signature:
//
// Name                 Index   Mask Register SysValue  Format   Used
// -------------------- ----- ------ -------- -------- ------- ------
// SV_POSITION              0   xyzw        0      POS   float   xyzw
// TEXCOORD                 0   xy          1     NONE   float   xy  
// TEXCOORD                 1   xyzw        2     NONE min2_8f   xyzw
//
//
// Constant buffer to DX9 shader constant mappings:
//
// Target Reg Buffer  Start Reg # of Regs        Data Conversion
// ---------- ------- --------- --------- ----------------------
// c0         cb1             0       250  ( FLT, FLT, FLT, FLT)
// c251       cb0             0         1  ( FLT, FLT, FLT, FLT)
//
//
// Runtime generated constant mappings:
//
// Target Reg                               Constant Description
// ---------- --------------------------------------------------
// c250                            Vertex Shader position offset
//
//
// Level9 shader bytecode:
//
    vs_2_0
    def c252, 0.5, 1, 0, 0
    dcl_texcoord v0
    dcl_texcoord1 v1
    mova a0.x, v1.x
    mul r0.xy, v0, c0[a0.x]
    add r0.x, r0.y, r0.x
    add oT0.x, r0.x, c0[a0.x].z
    mul r0.xy, v0, c1[a0.x]
    add r0.x, r0.y, r0.x
    add oT0.y, r0.x, c1[a0.x].z
    mov oT1 {min2_8f}, c2[a0.x]
    mad r0.xy, v0, c251.xzzw, c251.ywzw
    add oPos.xy, r0, c250
    mov oPos.zw, c252.xyxy

// approximately 11 instruction slots used
vs_4_0
dcl_globalFlags refactoringAllowed | enableMinimumPrecision
dcl_constantbuffer cb0[1], immediateIndexed
dcl_constantbuffer cb1[250], dynamicIndexed
dcl_input v0.xy
dcl_input v1.x
dcl_output_siv o0.xyzw, position
dcl_output o1.xy
dcl_output o2.xyzw {min2_8f}
dcl_temps 1
mad o0.xy, v0.xyxx, cb0[0].xzxx, cb0[0].ywyy
mov o0.zw, l(0,0,0.500000,1.000000)
mov r0.x, v1.x
dp2 r0.y, v0.xyxx, cb1[r0.x + 0].xyxx
add o1.x, r0.y, cb1[r0.x + 0].z
iadd r0.xy, v1.xxxx, l(1, 2, 0, 0)
dp2 r0.z, v0.xyxx, cb1[r0.x + 0].xyxx
add o1.y, r0.z, cb1[r0.x + 0].z
mov o2.xyzw {min2_8f}, cb1[r0.y + 0].xyzw {def32 as min2_8f}
ret 
// Approximately 0 instruction slots used

 

From this code, I can understand that the requested value is stored at cb1[r0.y+0].w which should be just a value in the constant buffer #2. And index to constant buffer #2 is stored at offset pointed by r0.y. But I am not able to compute this offset (which should just be value in constant buffer #1). I probably miss something, because if I read the constant buffers by

 



ID3D11Buffer* constantBuffers[2];
This->VSGetConstantBuffers(0, 2, constantBuffers);

 

the requested value is not there. The constantBuffers[0] is 16 bytes long, constantBuffers[1] is 4000 bytes long which correspond to the data written in shader disassembly.

 

I know I have a little bit non-standard task. But I would really appreciate any help with finding where the requested value (alpha component of pixel shader output) is stored. Thanks in advance!



Sponsor:

#2 Zaoshi Kaba   Crossbones+   -  Reputation: 4356

Like
0Likes
Like

Posted 13 February 2013 - 05:01 AM

cb1 seems to be 250 registers big, so it has to be that 4000 byte buffer, it's probably bunch of float4s, should be able to find it among 250 vectors.



#3 Big Muscle   Members   -  Reputation: 135

Like
0Likes
Like

Posted 13 February 2013 - 05:45 AM

I tried to dump that buffer and it seems to be stored there, but not as pure alpha, but rather (1.0 - alpha). The blend state is following if it depends:

 

blendStateDesc.RenderTarget[0].BlendEnable = TRUE
blendStateDesc.RenderTarget[0].SrcBlend = D3D11_BLEND_ONE

blendStateDesc.RenderTarget[0].DestBlend = D3D11_BLEND_INV_SRC_ALPHA

blendStateDesc.RenderTarget[0].BlendOp =  D3D11_BLEND_OP_ADD

blendStateDesc.RenderTarget[0].SrcBlendAlpha = D3D11_BLEND_ONE

blendStateDesc.RenderTarget[0].DestBlendAlpha = D3D11_BLEND_INV_SRC_ALPHA6
blendStateDesc.RenderTarget[0].BlendOpAlpha = D3D11_BLEND_OP_ADD

blendStateDesc.RenderTarget[0].RenderTargetWriteMask = 15

 

Now I need to find the correct index in constant buffer as it is on different position than I computed. I would expect that R-G-B-A has offsets 0-4-8-12, so alpha value is at offset 12. But here it seems to be at 8, so I think I have overlooked something in the shader assembly.



#4 Zaoshi Kaba   Crossbones+   -  Reputation: 4356

Like
0Likes
Like

Posted 13 February 2013 - 08:10 AM

It seems to be cb1[dot(input.pos.xy, cb1[input.tex0.x].xy) + 0].xyzw.
Somehow I cannot see what this could be; noise?


#5 Ponkx   Members   -  Reputation: 113

Like
1Likes
Like

Posted 13 February 2013 - 08:24 AM

Seems to me, that offset is computed as

  iadd r0.xy, v1.xxxx, l(1, 2, 0, 0)

that is TEXCOORD.x + 2



#6 Big Muscle   Members   -  Reputation: 135

Like
0Likes
Like

Posted 13 February 2013 - 12:47 PM

Yes! +2 works with correct RGBA offset! thanks!



#7 Big Muscle   Members   -  Reputation: 135

Like
0Likes
Like

Posted 13 February 2013 - 01:43 PM

so still something wrong, sometimes the value is incorrect.






Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS