Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Fix usage of vfma/vfms on platforms that don't have FMA extensions #221

Merged
merged 1 commit into from
Dec 13, 2024
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
Fix usage of vfma on platform that don't have it
stellar-aria committed Dec 13, 2024
commit 2e945d7386ae8836d10373c5354650525bf58cee
15 changes: 11 additions & 4 deletions Source/MatrixFunctions/arm_mat_cholesky_f32.c
Original file line number Diff line number Diff line change
@@ -258,12 +258,17 @@ ARM_DSP_ATTRIBUTE arm_status arm_mat_cholesky_f32(
vecGj1=vld1q_f32(&pG[(j + 1) * n + k]);
vecGj2=vld1q_f32(&pG[(j + 2) * n + k]);
vecGj3=vld1q_f32(&pG[(j + 3) * n + k]);

#if defined(__ARM_FEATURE_FMA)
acc0 = vfmaq_f32(acc0, vecGi, vecGj0);
acc1 = vfmaq_f32(acc1, vecGi, vecGj1);
acc2 = vfmaq_f32(acc2, vecGi, vecGj2);
acc3 = vfmaq_f32(acc3, vecGi, vecGj3);

#else
acc0 = vmlaq_f32(acc0, vecGi, vecGj0);
acc1 = vmlaq_f32(acc1, vecGi, vecGj1);
acc2 = vmlaq_f32(acc2, vecGi, vecGj2);
acc3 = vmlaq_f32(acc3, vecGi, vecGj3);
#endif
kCnt--;
k+=4;
}
@@ -319,9 +324,11 @@ ARM_DSP_ATTRIBUTE arm_status arm_mat_cholesky_f32(

vecGi=vld1q_f32(&pG[i * n + k]);
vecGj=vld1q_f32(&pG[j * n + k]);

#if defined(__ARM_FEATURE_FMA)
acc = vfmaq_f32(acc, vecGi, vecGj);

#else
acc = vmlaq_f32(acc, vecGi, vecGj);
#endif
kCnt--;
k+=4;
}
4 changes: 4 additions & 0 deletions Source/MatrixFunctions/arm_mat_solve_lower_triangular_f32.c
Original file line number Diff line number Diff line change
@@ -209,7 +209,11 @@
for(k=0; k < i; k++)
{
vecX = vld1q_f32(&pX[cols*k+j]);
#if defined(__ARM_FEATURE_FMA)
vecA = vfmsq_f32(vecA,vdupq_n_f32(pLT[n*i + k]),vecX);
#else
vecA = vmlsq_f32(vecA,vdupq_n_f32(pLT[n*i + k]),vecX);
#endif
}

if (pLT[n*i + i]==0.0f)
4 changes: 4 additions & 0 deletions Source/MatrixFunctions/arm_mat_solve_upper_triangular_f32.c
Original file line number Diff line number Diff line change
@@ -197,7 +197,11 @@ arm_status status; /* status of matrix inverse */
for(k=n-1; k > i; k--)
{
vecX = vld1q_f32(&pX[cols*k+j]);
#if defined(__ARM_FEATURE_FMA)
vecA = vfmsq_f32(vecA,vdupq_n_f32(pUT[n*i + k]),vecX);
#else
vecA = vmlsq_f32(vecA,vdupq_n_f32(pUT[n*i + k]),vecX);
#endif
}

if (pUT[n*i + i]==0.0f)
5 changes: 4 additions & 1 deletion Source/StatisticsFunctions/arm_mse_f32.c
Original file line number Diff line number Diff line change
@@ -132,8 +132,11 @@ ARM_DSP_ATTRIBUTE void arm_mse_f32(
pSrcB += 4;

vecA = vsubq_f32(vecA, vecB);

#if defined(__ARM_FEATURE_FMA)
vecSum = vfmaq_f32(vecSum, vecA, vecA);
#else
vecSum = vmlaq_f32(vecSum, vecA, vecA);
#endif
/*
* Decrement the blockSize loop counter
*/