Skip to content

S
swiftshader
Commit 9b3388ed by Alexis Hetu Committed by Alexis Hétu
Options

Fixed non square matrix multiplication

A few cases of matrix multiplication weren't working properly due to using the wrong matrix dimension. For example, if we have the following matrix operation (using matColRow): mat32 = mat22 * mat32 This results in 3 multiplication of 2 element vectors for each row of the result and the current code (before this cl) would only perform 2 multiplications and produce the wrong result. This cl should fix all uses of the * operator and of the matrixCompMult function when using non square matrices. Change-Id: Id0dbfd9d65c20102220049c34435c37e3db7f9da Reviewed-on: https://swiftshader-review.googlesource.com/3966Tested-by: 's avatarAlexis Hétu <sugoi@google.com> Reviewed-by: 's avatarNicolas Capens <capn@google.com>
parent 8e098376
Showing with 12 additions and 6 deletions
src/OpenGL/compiler/OutputASM.cpp
......@@ -533,11 +533,12 @@ namespace glsl
case EOpVectorTimesMatrix:
if(visit == PostVisit)
{
int size = leftType.getNominalSize();
sw::Shader::Opcode dpOpcode = sw::Shader::OPCODE_DP(leftType.getNominalSize());
int size = rightType.getNominalSize();
for(int i = 0; i < size; i++)
{
Instruction *dot = emit(sw::Shader::OPCODE_DP(size), result, left, right);
Instruction *dot = emit(dpOpcode, result, left, right);
dot->dst.mask = 1 << i;
argument(dot->src[1], right, i);
}
......@@ -549,7 +550,8 @@ namespace glsl
Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
mul->src[1].swizzle = 0x00;
for(int i = 1; i < leftType.getNominalSize(); i++)
int size = rightType.getNominalSize();
for(int i = 1; i < size; i++)
{
Instruction *mad = emit(sw::Shader::OPCODE_MAD, result, left, right, result);
argument(mad->src[0], left, i);
......@@ -562,7 +564,8 @@ namespace glsl
{
int dim = leftType.getNominalSize();
for(int i = 0; i < dim; i++)
int size = rightType.getNominalSize();
for(int i = 0; i < size; i++)
{
Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, left, right);
mul->dst.index += i;
......@@ -1209,9 +1212,12 @@ namespace glsl
case EOpMul:
if(visit == PostVisit)
{
ASSERT(dim2(arg[0]) == dim2(arg[1]));
TIntermTyped *arg0 = arg[0]->getAsTyped();
TIntermTyped *arg1 = arg[1]->getAsTyped();
ASSERT((arg0->getNominalSize() == arg1->getNominalSize()) && (arg0->getSecondarySize() == arg1->getSecondarySize()));
for(int i = 0; i < dim2(arg[0]); i++)
int size = arg0->getNominalSize();
for(int i = 0; i < size; i++)
{
Instruction *mul = emit(sw::Shader::OPCODE_MUL, result, arg[0], arg[1]);
mul->dst.index += i;
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment