From: lhez <redacted>
Date: Thu, 30 Oct 2025 23:00:20 +0000 (-0700)
Subject: opencl: fix boundary handling for mul_mm (llama/16875)
X-Git-Tag: upstream/0.9.4.185~54
X-Git-Url: https://git.djapps.eu/?a=commitdiff_plain;h=f14754dfd817bd611ce108c17bdcbda21a93667d;p=pkg%2Fggml%2Fsources%2Fggml

opencl: fix boundary handling for mul_mm (llama/16875)
---

diff --git a/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl b/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl
index 1a1bfe14..6982f8f5 100644
--- a/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl
+++ b/src/ggml-opencl/kernels/mul_mm_f16_f32_l4_lm.cl
@@ -79,8 +79,8 @@ kernel void kernel_mul_mm_f16_f32_l4_lm(
 
     for (int block = 0; block < ne00; block += BK) {
         for (int l = 0; l < BM; l += loadstride_a) {
-            if (loadc_a + l < ne01) {
-            const int idx = pos_a + (loadc_a + l) * stride_a / LOAD_VEC_A + loadr_a;
+            if (ir*BM + loadc_a + l < ne01) {
+                const int idx = pos_a + (loadc_a + l) * stride_a / LOAD_VEC_A + loadr_a;
                 buf_a[(loadr_a * LOAD_VEC_A + 0) * BM + loadc_a + l] = src0[idx].s0;
                 buf_a[(loadr_a * LOAD_VEC_A + 1) * BM + loadc_a + l] = src0[idx].s1;
                 buf_a[(loadr_a * LOAD_VEC_A + 2) * BM + loadc_a + l] = src0[idx].s2;
@@ -94,7 +94,7 @@ kernel void kernel_mul_mm_f16_f32_l4_lm(
         }
 
         for (int l = 0; l < BN; l += loadstride_b) {
-            if (loadc_b + l < ne11) {
+            if (ic*BN + loadc_b + l < ne11) {
                 const int idx = pos_b + (loadc_b + l) * stride_b / LOAD_VEC_B + loadr_b;
                 buf_b[(loadr_b * LOAD_VEC_B + 0) * BN + loadc_b + l] = src1[idx].s0;
                 buf_b[(loadr_b * LOAD_VEC_B + 1) * BN + loadc_b + l] = src1[idx].s1;
diff --git a/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl b/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl
index 39a5d486..d7d5ba64 100644
--- a/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl
+++ b/src/ggml-opencl/kernels/mul_mm_f32_f32_l4_lm.cl
@@ -79,7 +79,7 @@ kernel void kernel_mul_mm_f32_f32_l4_lm(
 
     for (int block = 0; block < ne00; block += BK) {
         for (int l = 0; l < BM; l += loadstride_a) {
-            if (loadc_a + l < ne01) {
+            if (ir*BM + loadc_a + l < ne01) {
                 const int idx = pos_a + (loadc_a + l) * stride_a / LOAD_VEC_A + loadr_a;
                 buf_a[(loadr_a * LOAD_VEC_A + 0) * BM + loadc_a + l] = src0[idx].s0;
                 buf_a[(loadr_a * LOAD_VEC_A + 1) * BM + loadc_a + l] = src0[idx].s1;
@@ -94,7 +94,7 @@ kernel void kernel_mul_mm_f32_f32_l4_lm(
         }
 
         for (int l = 0; l < BN; l += loadstride_b) {
-            if (loadc_b + l < ne11) {
+            if (ic*BN + loadc_b + l < ne11) {
                 const int idx = pos_b + (loadc_b + l) * stride_b / LOAD_VEC_B + loadr_b;
                 buf_b[(loadr_b * LOAD_VEC_B + 0) * BN + loadc_b + l] = src1[idx].s0;
                 buf_b[(loadr_b * LOAD_VEC_B + 1) * BN + loadc_b + l] = src1[idx].s1;
diff --git a/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl b/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl
index fd47e8a8..147b66f6 100644
--- a/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl
+++ b/src/ggml-opencl/kernels/mul_mm_q8_0_f32_l4_lm.cl
@@ -78,7 +78,7 @@ kernel void kernel_mul_mm_q8_0_f32_l4_lm(
 
     for (int block = 0; block < ne00; block += BK) {
         for (int l = 0; l < BM; l += loadstride_a) {
-            if (loadc_a + l < ne01) {
+            if (ir*BM + loadc_a + l < ne01) {
                 int idx = pos_a + (loadc_a + l) * stride_a / LOAD_VEC_A + loadr_a;
                 int ib  = idx / 8;
                 int iqs = idx % 8;
@@ -101,7 +101,7 @@ kernel void kernel_mul_mm_q8_0_f32_l4_lm(
         }
 
         for (int l = 0; l < BN; l += loadstride_b) {
-            if (loadc_b + l < ne11) {
+            if (ic*BN + loadc_b + l < ne11) {
                 int idx = pos_b + (loadc_b + l) * stride_b / LOAD_VEC_B + loadr_b;
                 buf_b[(loadr_b * LOAD_VEC_B + 0) * BN + loadc_b + l] = src1[idx].s0;
                 buf_b[(loadr_b * LOAD_VEC_B + 1) * BN + loadc_b + l] = src1[idx].s1;